The present application is a continuation-in-part of U.S. patent application Ser. No. 16/871,869, filed May 11, 2020 and entitled “Privacy Control and Enhancements for Distributed Networks,” which is a continuation-in-part of U.S. patent application Ser. No. 16/802,733, filed Feb. 27, 2020 and entitled “Methods and Apparatus for Device Location Services,” which is a continuation-in-part of U.S. patent application Ser. No. 16/774,832, filed Jan. 28, 2020 and entitled “Privacy Enhancement Using Derived Data Disclosure,” which is a continuation-in-part of U.S. patent application Ser. No. 16/585,438, filed Sep. 27, 2019 and entitled “Methods and Apparatus for Determining Preferences and Events and Generating Associated Outreach Therefrom,” which claims priority to U.S. Provisional Patent Application Ser. No. 62/737,448, filed Sep. 27, 2018 and also entitled “Methods and Apparatus for Determining Preferences and Events and Generating Associated Outreach Therefrom,” all of which are incorporated by reference herein in their entirety. The above-cited U.S. patent application Ser. No. 16/774,832 also claims priority to U.S. Provisional Patent Application Ser. No. 62/797,550, filed Jan. 28, 2019 and entitled “Privacy Enhancement Using Derived Data Disclosure,” which is incorporated by reference herein in its entirety. The above-cited U.S. patent application Ser. No. 16/802,733 also claims priority to U.S. Provisional Patent Application Ser. No. 62/811,240, filed Feb. 27, 2019 and entitled “Methods and Apparatus for Device Location Services,” which is incorporated by reference herein in its entirety. The above-cited U.S. patent application Ser. No. 16/871,869 also claims priority to U.S. Provisional Patent Application Ser. No. 62/846,109, filed May 10, 2019 and entitled “Privacy Control and Enhancements for Distributed Networks,” which is incorporated by reference herein in its entirety. The present application further claims priority to U.S. Provisional Patent Application Ser. No. 62/963,230, filed Jan. 20, 2020 and entitled “Infrastructure Support to Enhance Resource-Constrained Device Capabilities,” which is incorporated by reference herein in its entirety.
The field relates generally to information processing systems, and more particularly to systems that process sensor data.
Traditional sensor-based systems, such as alarm systems, are not intelligent and require the triggering of sensors connected to a controller that generates alarms in response. Although various types of artificial intelligence (AI) using inputs from sensors are also known in the art, a need remains for accurate and efficient techniques that can provide deeper analysis, without requiring vast amounts of tagged data. For example, it would be highly desirable to have techniques that can both provide an understanding of an observed space as well as react to the derived understanding in a manner that provides optimal benefit to users, which includes ensuring that users are provided with expected levels of privacy relating to sensor data.
Privacy techniques have been well studied in the context of Internet communication and electronic payments. For example, privacy researchers have developed privacy enhancing communication methods such as onion routers, and have developed privacy-preserving payment schemes of many types. Regulation has been developed for storage and access of medical data, e.g., HIPAA, and associated technology solutions have been deployed. Many countries have regulations dictating how personal identifying information (PII) can be collected and stored, e.g., in the context of web browser trackers such as HTML cookies. The latter requires that an end user is asked to agree for certain information to be collected and stored for such features to be used relative to a communication session.
Conventional privacy techniques are often specific to the respective contexts for which they were developed. For example, an onion router is useful for implementing communications privacy, but is not directly applicable to protections relating to collection of HTML cookie data, as the onion router does not remove or control HTML cookies, and does not verify that a user is willing for HTML cookie information to be collected and stored. Similarly, since the collection of medical data typically has nothing at all to do with detecting an identifier relating to a browser, an onion router does not address HIPAA related requirements in general. Another example is that of methods to enhance payment privacy, such as blind digital signatures. These are specific to the context for which they were developed, and do not address regulation regarding web browsing nor HIPAA. Developed technologies, similarly, do not readily address other privacy contexts not mentioned above.
In addition, an increasing number of resource-constrained devices are being deployed, but their use is held back by their constraints. This leads to service failures and degradation of functionality, including security and privacy.
It is therefore apparent that a need exists for better approaches, including approaches that can effectively enhance the capabilities of resource-constrained devices while overcoming disadvantages of conventional approaches.
Illustrative embodiments provide techniques for infrastructure support to enhance resource-constrained device capabilities. Some embodiments provide such infrastructure support in the context of techniques for privacy control and enhancements for distributed networks, possibly in conjunction with device location services and/or privacy enhancement using derived data disclosure, as well as in numerous additional or alternative contexts. Some of these embodiments further involve determining preferences and events and generating associated outreach therefrom. Such outreach illustratively comprises various types of automated actions that are performed based at least in part on predicates each computed using data generated by multiple sensor devices.
In one embodiment, an apparatus comprises at least one processing device comprising a processor coupled to a memory. The at least one processing device is configured to receive in a requesting device a service availability notification relating to availability of one or more services implemented by a service providing device, to generate in the requesting device a request for a particular service based at least in part on the received service availability notification, and to send the request for the particular service from the requesting device to the service providing device. The requesting device illustratively comprises a resource-constrained device relative to the service providing device with respect to the particular service, and the particular service comprises a service that is capable of being performed in the requesting device but is offloaded from the requesting device to the service providing device via the request in order to conserve resources of the requesting device. The service providing device performs the particular service for the requesting device in accordance with the request.
At least one of the requesting device and the service providing device illustratively comprises a mobile device, such as a mobile telephone, although a wide variety of other types of user devices can be used in other embodiments.
Some embodiments disclosed herein implement methods to maintain user security and privacy, which are poised to be problems of increasing importance, particularly as people deploy increasing numbers of IoT devices in their home, where IoT denotes Internet of Things. The configuration, maintenance and use of such devices potentially pose risks to end users unless properly managed.
These and other illustrative embodiments include but are not limited to systems, methods, apparatus, and computer program products. The illustrative embodiments are advantageously configured to address and solve one or more significant problems of conventional approaches, as outlined in more detail elsewhere herein.
FIG. 1 is a block diagram of an information processing system comprising a set of sensors and a corresponding set of predicates that can be generated using outputs of the sensors in an illustrative embodiment.
FIG. 2 shows a plurality of network nodes of an information processing system in an illustrative embodiment.
FIG. 3 illustrates association of the network nodes of FIG. 2 with particular rooms of a home in an illustrative embodiment.
FIG. 4 is a block diagram of an information processing system configured with functionality for determining preferences and events and generating associated outreach therefrom in an illustrative embodiment.
FIG. 5 is a block diagram of an information processing system comprising a set of sensors and a corresponding set of predicates, augmented with machine learning functionality in an illustrative embodiment.
FIGS. 6 through 8 show additional examples that involve determining preferences and events and generating associated outreach therefrom in illustrative embodiments.
FIGS. 9 and 10 show examples of user interfaces including permission requests generated in accordance with techniques for privacy enhancement using derived data disclosure in illustrative embodiments.
FIG. 11 is a block diagram of an information processing system with functionality for privacy enhancement using derived data disclosure in an illustrative embodiment.
FIG. 12 is a block diagram of another information processing system with functionality for privacy enhancement using derived data disclosure in an illustrative embodiment.
FIG. 13 shows an example layout of an area comprising multiple sensor devices in an illustrative embodiment.
FIGS. 14 and 15 show additional examples of user interfaces including permission requests generated in accordance with techniques for privacy enhancement using derived data disclosure in illustrative embodiments.
FIGS. 16 and 17 are plots illustrating different privacy impacts for different types of derived data disclosure in illustrative embodiments.
FIG. 18 is a block diagram of an information processing system that implements functionality for privacy enhancement using derived data disclosure to third party service providers in an illustrative embodiment.
FIG. 19 is a flow diagram of a process for privacy enhancement using derived data disclosure to third party service providers in an illustrative embodiment.
FIG. 20 is a block diagram of another information processing system that implements functionality for privacy enhancement using derived data disclosure to third party service providers in an illustrative embodiment.
FIGS. 21 and 22 are block diagrams of information processing systems configured to provide device location services in illustrative embodiments.
FIG. 23 is a flow diagram of a process for privacy control and enhancements for distributed networks in an illustrative embodiment.
FIG. 24 is a block diagram of an information processing system utilized to implement the process of FIG. 23.
FIG. 25 shows an example arrangement of features, profiles and fields utilized in the illustrative embodiments of FIGS. 23 and 24.
FIG. 26 is a block diagram showing a more detailed view of a portion of the information processing system of FIG. 24.
FIGS. 27 through 30 are diagrams showing additional aspects of privacy control and enhancements in illustrative embodiments.
FIGS. 31 through 35 are diagrams illustrating aspects of infrastructure support to enhance resource-constrained device capabilities in illustrative embodiments.
Illustrative embodiments will be described herein with reference to exemplary information processing systems and associated computers, servers, storage devices and other processing devices. It is to be appreciated that the embodiments described below are presented by way of example only, and should not be construed as limiting in any way.
Some of these illustrative embodiments can make use of one or more of the techniques described in one or more of the following patent applications, each incorporated by reference herein in its entirety:
U.S. patent application Ser. No. 16/527,826, filed Jul. 31, 2019 and entitled “Managing Access Rights of Transferable Sensor Systems,”
U.S. patent application Ser. No. 16/598,614, filed Oct. 10, 2019 and entitled “Configuration and Management of Smart Nodes with Limited User Interfaces,” now issued as U.S. Pat. No. 10,887,447,
U.S. patent application Ser. No. 16/676,978, filed Nov. 7, 2019 and entitled “Third Party Application Enablement for Node Networks Deployed in Residential and Commercial Settings,”
U.S. patent application Ser. No. 16/682,627, filed Nov. 13, 2019 and entitled “Managing Power for Residential and Commercial Networks,” and
U.S. patent application Ser. No. 17/005,949, filed Aug. 28, 2020 and entitled “Privacy and the Management of Permissions.”
The example embodiments described herein are for purposes of illustration only. Numerous other arrangements and associated operating scenarios are possible in other embodiments.
The disclosed technology in some embodiments addresses several long-felt needs in the context of the rapidly developing area of home and office automation. One of these aspects include methods addressing how to automatically extract valuable information regarding preferences, events, layout, space use, and the like, without the need for massive quantities of tagged data. Another relates to how to selectively share data with both trusted and untrusted service providers with the goal of enhancing the user value of system ownership, defraying the cost of services, customize content and services, and providing numerous other desirable features. A third aspect relates to security and privacy aspects related to the processing, communication and sharing of data, a core aspect of which is an abstraction of sensitive user data to create anonymized predicates related to the preferences and needs of users, as well as to events related to the system.
In one embodiment, a likely location of a user is determined by computing a presence predicate relative to one or more nodes in a network, where sensor data from one or more nodes is processed and a node is selected in a manner that maximizes the probability that an observed event, observed using the sensors of one or more nodes, is best observed at the node for which the presence is determined. Here, “best” may correspond to the strongest signal, the signal with the least attenuation, a signal corresponding to multiple correlated sensor values that are strongly correlated and for which the signal is strong, or related techniques. In addition, triangulation techniques can be used to determine relative position and movement.
The system illustratively establishes a physical layout of the network of nodes based on multiple presence determinations over time, e.g., a determination of the sequence of selections of one node being the one with the greatest likely proximity to a given observed person. This does not have to be the same person, as the system can determine, using statistical methods, that two nodes are co-located due to them often being selected in temporal series as the most likely location of a user.
Given a network map corresponding to the layout of a space, established in this way, the system also can determine with high precision the location of a user as he or she moves through the space by looking at historical or common movements, and observing a series of presence determinations, in addition to using the outputs of the sensor of the nodes in the network. This increases the accuracy of the determination, and allows predictive capabilities that are useful to proactively respond to likely events, e.g., lighting a path ahead of a person walking through a building, but only lighting areas where the user is, has just been, or is likely to go next. The same predictive provision can also be made to audio content, allowing the user to experience a speaker phone that follows him or her around as he or she moves through an apartment or office, as well as other features based on predictive establishment of location. This is useful both for convenience and for security, where first-responders such as law-enforcement and firemen can benefit from the use of the security features.
Although the location in some embodiments can be computed by either a third party service provider or the system, it may be beneficial for the location to be computed by the system, which then shares the corresponding predicate with the service provider (assuming the user has granted permission to that) since that illustratively reduces the amount of information that has to be disclosed to the third party service provider, and provides or enables other benefits as described in this disclosure.
A user pseudonym can be generated using two types of inputs: a presence assertion, as established above, and an observation, illustratively an event that is associated with one user. Here, one such event is the transmission of a media access control (MAC) address or other hardware identifier, or use of a session key or other software identifier, by a mobile user, where the observation is made using a node equipped with a radio unit, which is an example of what is more generally referred to herein as a “radio sensor.”
This observation can be made independently of whether the node is actively engaged in the transmission or only is a passive recipient. Note that radio signals will be strongest in close proximity to a node (and its associated sensors), and especially so if the power of the radio is very limited. This way, the system correlates location assertions (which may be made in a probabilistic manner with multiple rather likely options being output along with their assessed probabilities) and radio proximity locations. This correlation can also use triangulation, and will benefit from the map of network nodes described above. When there is a strong correlation between the location as asserted by the proximity predicate and the observation using the radio of a sensor, then the MAC address or other identifier is associated with the person whose location was assessed to correspond to the location of the presence assertion.
Multiple identifiers can be assigned with one and the same user, as illustrated in FIG. 1. Multiple users may use one device as well. The pseudonyms generated in this way, which may be simple local identifiers associated with MAC addresses and other identifiers, correspond to tagged data. This tagged data can be used to train machine learning (ML) components that take other sensor data such as sound and motion data, and correlates the user pseudonyms with such data in order to generate a classifier that, given the sensor data such as sound and motion alone, and without any MAC address or other signal received by a radio unit, assigns a set of sensor observations to a pseudonym.
Having profiles associated with pseudonyms permits the system to personalize the treatment of users, which has both convenience benefits and security benefits. These benefits will be further detailed below. A pseudonym can be an identifier, such as a MAC address, or a locally unique value that is associated, in a database, with one or more identifiers, such as MAC addresses. A user can be given multiple pseudonyms, where these pseudonyms may either correspond to the likely user or the likely device. The system can detect that one identifier is used by multiple users, possibly using Global Positioning System (GPS) data to disambiguate the users. For example, knowing that user A is in Chicago based on user A's phone being there, and noticing that user A's iPad is used in user A's home, in the living room, the system knows that somebody else uses user A's iPad. If the use of it is also associated with user B's phone, location-wise, then the system associates user A's iPad with both user A and user B. The system can determine which user it is based on additional sensor data, such as GPS data, distinguishing sound data, data that the system obtains from accessing the user's calendars and email, and other types of data. This enables the system to identify common use, for purposes of convenience, configuration, but also, for enhanced security, where the latter is enabled by detecting anomalies.
In the context of privacy, the sharing of pseudonyms is beneficial in comparison to the sharing of the underlying data that is required to compute the pseudonym predicate. This is generally in accordance with the same principles as for other predicates. However, in the context of pseudonyms, there are further advantages of sharing of predicates. For example, different service providers can be given different pseudonyms for one and the same user, the pseudonyms can be session based (where a session may be a period of presence, a day, or any selected period of time), and/or one pseudonym can correspond to multiple devices of the same believed end user. Pseudonyms can therefore provide additional privacy advantages in illustrative embodiments, as will be understood by a person skilled in the art.
The disclosed system introduces privacy-protecting methods to classify sound content, thereby identifying user preferences. The system distinguishes, for example, the sound profiles of different sports and different music genres from each other, and is configured to store information deduced about these predicates; however, the system illustratively does not store sound recordings, and therefore actively ignores conversations in the observed space, except for voice input that is determined to correspond to voice commands to the system. These are illustratively placed in the context of the observed events. For example, it is important for the classification of an observed sound sample that is “help help help” to know (a) whether this is being said in the context of a higher-than-normal risk for distress; in the context of a known movie that contains this voice segment; or in the context of apparent children's laughter. Thus, the system identifies and classifies contexts, events and situations, and use these classifications to perform actions while minimizing the risk for false positives and false negatives.
The disclosed system is further configured to detect the location of and operation of various appliances, such as washer, dryer, water heater, HVAC (heating, ventilation and air conditioning) appliances, refrigerator, oven, microwave, dishwasher, fans, etc.; and to identify rooms based on the use of identified appliances in these rooms. If the system identifies that an appliance has a change in its usage profile, e.g., generates a sound that was previously not observed; consumes energy in a manner that was not previously observed, and the like, then this is indicative of a risk of malfunction. In addition, if the sensor outputs associated with such appliances are indicative of a known type of malfunction, then this comprises an estimate of a classification of a potential need for maintenance, a potential risk, a potential need for the replacement of the appliance, and/or the need for a user action.
Such information is very important to derive and act on, and accordingly, the system determines events that are likely to match such situations or needs, based on the sensor output profiles observed by the system. These profiles can be in the time range, for example, as in a typical reporting of sensor output values, or in the frequency range, for example, as in the reporting of a Fast Fourier Transform (FFT). The use of correlation between outputs is useful whether the values being processed are in the time or frequency range. Correlation between multiple types of sensor outputs is beneficial to obtain derived sensor profile data taking multiple dimensions of sensor data into consideration.
The benefits of the disclosed technology for explaining data disclosure needs, request permissions, and grant permissions are readily appreciated in this context, as these associated predicates are much less intrusive in terms of the privacy impact than the associated raw data, but also much easier to conceptualize for an end user asked if he or she wishes to grant access to the information. When it comes to static classifications, or mostly static classifications, such as “this is a TV room,” or “this is a room with a laundry machine,” the benefits are also particularly evident, as the disclosure of information occurs only once (assuming no change of classification), and no further sharing of associated data is required once that has been done.
Other risks, such as earthquakes are also detected by the disclosed system, along with associated responses to minimize the risk of such events, and methods to streamline and aid the work of emergency responders. A person skilled in the art will recognize that the disclosed methods are only exemplary, and that they are used to illustrate the operation of the system and the benefits of the system.
The earthquake detector can use inputs from multiple homes in a related geographic area, and, as such, be used as part of an early prediction system. There are many benefits with such systems, including being able to automatically shut off gas, automatically secure elevators, and more; and even just a notice of a fraction of a second can save lives and property. However, the accurate detection of earthquakes illustratively takes into account many sensor data types, from many nodes within many homes, businesses and other installations. Therefore, the theoretical privacy impact is significant in the context of a traditional approach wherein a data consumer receives all data necessary for the computation from the multiple data sources and makes the assessment. Unless this data consumer is trusted by all parties involved, this will not be acceptable.
Accordingly, illustrative embodiments may be configured, for example, to compute, by one or more collaborating systems, a set of predicates indicative of earthquake movement, and to transmit such predicates to a third party in charge of making rapid classifications and issuing alerts, thereby providing significant privacy benefits. This is a good example of a situation where privacy concerns, should traditional approaches be used instead of the disclosed technology, could block the widespread participation by end users, and therefore significantly limit the accuracy and, as a result, benefits of the detection system. Thus, the disclosed system directly improves privacy, but also indirectly improves the safety of users.
The system in some embodiments is configured to convey predicates selectively to trusted third parties, such as one or more advertisers or advertisem*nt providers, and is also configured to enable the outreach to individual third party service providers, e.g., for purposes of predictive maintenance. We disclose methods to make such connections in a way that is compatible with the existing advertisem*nt infrastructure, which is famously demanding in terms of the time constraints it poses on its components in terms of enabling rapid auctions of advertisem*nts based on known criteria and user profiles.
In one embodiment, this is achieved by establishing a connection between a user device and an advertisers or advertisem*nt provider, and to either piggyback or separately convey profile data such as classifications of user actions, needs, spaces, and the like, to the advertiser or advertisem*nt provider. This enables the integration of physical systems such as smart homes and smart offices into the traditional advertisem*nt infrastructure in a backwards compatible manner, and without re-architecting how existing systems operate. For purposes of predictive maintenance, this can be offered to the user by notifications and discount coupons tied to third party trusted service providers, where such third party trusted service providers are billed as they connect to the service provider to obtain details related to the needs of the user. Such needs identify the believed source of the problem, the type of equipment, when known, and what parts may be needed based on such analysis. In addition, the third party trusted service provider is given access to location information and contact information of the user, which the user consents to being provided by enabling the received coupon or other offer.
These are new structures enabling substantial economic benefits for all parties involved, and significant improvements in the precision with which assessments can be made, translating into savings in time and convenience for those involved. In the case of the predictive maintenance scenario, it is also enabling a tight-knit reputation system leveraged by a feedback channel from the user and a measurement-based feedback channel from the sensors associated with the nodes associated with the predictive maintenance or other system modifications, as applicable. Namely, the service provider can determine when the representative of the third party service provider started making modifications; when the modifications concluded; and that the installation resulted in a non-anomalous operational profile. If defective parts were mistakenly or backhandedly used, the service provider would in many instances be able to detect this based on the energy consumption profiles observed in context of the expected profiles.
Thus, third party service providers that do not provide quality service can be excluded by the service provider, or given lower priority, lower pay for the negotiated service, or other discouraging mechanisms for enforcing desirable behavior; analogously, highly dependable and reliable third party service providers can be given priority, bonuses, or other encouragements. This weeding of third party service providers produces an improved service provision to the end user. As will be understood by a person skilled in the art, these are just exemplary embodiments of which there are a large number of variations of related benefit. More generally, commercial offers, including advertisem*nts are illustratively generated in response to detected needs, which comprise needs for maintenance, needs for products, needs for service as well as other needs identified based on the determined example predicates, as will be understood by a person skilled in the art.
A user can choose between two or more versions of an application, e.g., a paid application version (costing, say, $10 a month or a flat fee of $25, etc.) and an advertisem*nt-supported version. The latter may correspond to additional permission requests, e.g., to obtain access to data classifying interests, obtain demographic information, etc. The requests can be presented as a main set of requests corresponding to the common aspect of the two versions and then an optional aspect corresponding to the advertisem*nt version. Alternatively, two separate application descriptions with their corresponding permissions can be presented.
In addition to using heuristic methods for classification, the system illustratively also uses ML models, which are illustratively bootstrapped by training them on the output from heuristic models. These heuristic models for classification can be used to generate predicates. This is another benefit of the disclosed approach, as it increases the convergence speed of the ML model while minimizing the amount of externally provided tagged data and the quantities of sensor data that is needed for the ML model to become accurate.
The predicates that are output from the above-noted ML models can be given practical descriptions that correspond to their conceptual meaning, making it easy for end users to understand the meaning of these predicates, in spite of it most likely not being easy to understand the manner in which the predicates were derived. Thus, as ML and other artificial intelligence becomes more prevalent, the disclosed technology helps protect privacy by creating conceptually comprehensible quantities from raw data, and by pegging permissions and controls to these quantities.
The disclosed system operates well in isolation, as described above and in the details provided with the figures, but does also benefit from interaction with components that are not part of the sensor networks, but which can be accessed using an application programming interface (API) or other feed. For example, knowledge of the GPS location of a registered user is beneficial for predicting when that user will return home (or arrive in the office, for a space corresponding to an office as opposed to a home). This can be obtained using an API to a service provider that reads GPS location, such as some search engines do for example, as well as directly from the user's cellular phone, on which the user has an application (“app”) that can be used to control the space and to query the space about environmental data.
Moreover, the system can also integrate with cleaning robots, such as Roomba™ units, many of which provide layout information as a service. However, even without the active participation of the cleaning robot, the system can infer the layout of the space and the location and shape of obstacles. That is because typical cleaning robots have wireless capabilities, and therefore, can be detected and identified based on their MAC address or other hardware identifier. This detection is illustratively made by multiple nodes equipped with radio units, thereby allowing these to collectively triangulate the location of the cleaning robot with frequent intervals, simply based on signal strength of the observed cleaning robot radio as observed by one or more nodes equipped with a radio unit. The corresponding outputs of processing of this type can be in the form of raw data, or illustratively, predicates that correspond to conceptual quantities.
In one embodiment, at least some nodes comprise an IR transmitter compatible with typical consumer electronics (TVs, stereos, DVD players, etc.), just like universal remote controls, and are used to control associated equipment. In one embodiment, a user indicates in a portal associated with the service provision that a new piece of equipment has been plugged in, along with its position and an identification of its type (such as “Samsung TV, 55-inch flat screen” or “Philips DVD player”). The backend service consults a database to determine the encoding of signals for the corresponding device, and then provisions associated user apps on phones, tablets and similar devices to enable the operation of the introduced equipment.
The location can be determined by the user turning on the equipment in response to a challenge from the system, resulting in a detectable change in the energy consumption or load on the corresponding outlet, assuming the equipment is plugged in to a smart outlet supported by the backend. Alternatively, the user indicates the location of the new equipment from a pull-down list of locations, each location which corresponds to a system identified room; or indicates a location by making a loud noise in the room, or by otherwise responding to a challenge by the system that results in the detection of an action by the user. One such action is to hold up the user's phone or tablet to an outlet or switch or other type of node in the proximity of the installed equipment. Another alternative is that the system cycles through all possible power-on sequences for the IR transmitters in the space associated by the system, and determines from the power consumption, load determination, or sound output from the equipment when the new equipment has been turned on. This is illustratively not done in a way that activates already introduced equipment.
These processes, and variants of these, are used to establish a node in the proximity of the new equipment, and the encoding to be used to control the new equipment, thereby enabling the remote control, using a software remote control in the form of an app, to be used to control the equipment. This can be combined with parental control techniques to limit screen time, whether on an equipment basis, location basis, depending on the individual app or device used to control the equipment, or other techniques.
It can also be used for automated control of equipment, e.g., for using a stereo system as an alarm clock, but without the stereo system having to be able to determine the time. This can be done from a central location of control to which the user establishes and enables the alarm. A person skilled in the art will recognize that this approach can be used for a large array of consumer equipment that is controllable by remote controls, including audio equipment, TVs, recorders, disc players, free-standing and built-in fans, heaters and air conditioners, smart locks, pet doors, pet feeding systems, plant watering systems, and so on.
An important enhancement associated with the disclosed technology is the simple process of introduction of the equipment into the system, requiring a minimum of effort and knowledge; another is the enhanced functionality and control made possible by the connection of the disclosed system to compatible units that can be remote controlled. The latter enables the node network and the associated control structure to advance the benefits of the users greatly while maintaining simplicity, controlling cost, and maintaining centralized management that simplifies security and privacy in an array of ways.
Users would not grant access to control all devices corresponding to an installation, but would instead selectively grant access to the control. Thus, capabilities for control of equipment that are not part of the network can be granted, where this equipment instead is connectable to by the nodes of networks corresponding to the system.
The disclosed techniques can be used to gain information about users that is both privacy sensitive and much more insightful than current methods. For example, the system can determine the presence of one or more users in a TV room, and can determine the programming; then infer the time of the commercial, if applicable; and determine the user actions during the commercial. Did one or more users leave the room at the time of the commercial? Was it to get a snack (go to the kitchen), go to the bathroom, etc.? To the extent that the identity of one or more users leaving the room can be determined, what are their known genders and ages, based on demographic information collected, MAC address and other device identifiers? Was the volume turned off? Based on the sound associated with the programming, and based on the sound in the TV room, was there conversation? Was that timed towards the end of the commercial? Was there laughter? These are important types of insights that can be gained, and in terms of statistics, shared with advertisers, TV networks, and others.
Similarly, it is of relevance to know, both for programming and commercial segments, the information relating to the people in the TV room. Based on motion sensor data and traces from presence data, how many users are present? Based on MAC addresses and other device identifiers, what are the pseudonyms and/or identities of these? Based on collected demographic information and information obtained from user behavior, what are the demographics of these users, and what are their previously observed preferences and behavior? This is valuable to know to determine what programming is appreciated for various users, both on an individual basis in order to determine user preferences and improve recommendations, and in terms of statistical determinations. The disclosed system enables these and other related queries to be answered, based on the disclosed stricture, as will be appreciated by a person skilled in the art.
Using the disclosed permissions-based system, predicates corresponding to the events can be used to improve functionality, while protecting the privacy of the underlying user data. This results in a better system, whether it provides, for example, more relevant advertisem*nts selected based on the context and presence, or otherwise better responds to the situational needs determined from the context.
One benefit of the disclosed technology is how it can inform first responder decisions. Examples of such decisions include firemen prioritizing rescues, police rapidly confronting gunmen, and helping search-and-rescue teams find earthquake survivors trapped in the rubble of collapsed buildings. For all of these, an automated and immediate understanding of the likely location of people is significant; for a case involving active shooters, it is also vital to know the location of offenders as opposed to potential victims. This can be achieved by detection of gunshots and an immediate tracking of the associated location of all people in the presence of the detected shootings, and, as previously explained, while also protecting raw data and the associated user privacy, thereby aligning needs with understandable permission requests.
It may be impossible to eliminate active-shooter situations; but it is possible to provide real-time insights to the professionals that jump into active-shooter situations. An important goal of the system in the present embodiment is to aid the responders and help protect lives. The combination of discretely-placed IoT devices, gunshot audio detection, and cloud-based analysis, control, and notifications enables the following benefits during active-shooter situations:
1. Real-time gunshot detection
2. Real-time gunshot location identification and tracking
3. Estimation of number of active shooters
4. Identification and communication of shelter-in-place warnings
5. Enable identification and communication of areas for safe and urgent evacuation
6. Identification and communication of ideal evacuation routes
7. Remote and secure electrical power control of buildings, circuits, lighting, and individual plugs
8. Estimation of number and types of gunshots for medical first-responders
9. Assistance and testing during active-shooter trainings The disclosed system may be deployed within the confines of the existing building electrical system. There is no need to deploy a stand-alone, expensive, and difficult to conceal shooter detection system. Identification and location of gunshots is primarily made through audio detection with fixed-location devices; though this technology is greatly aided with the addition of multiple types of sensors. Algorithms which process a sensor signal automatically predict future sensor states and perform alerts and notification through software and/or other functionally equivalent firmware, hardware, or electronics, for processing data and digitally performing tasks. Notifications and predictions are enabled through smart computing such as artificial, deep learning, forward chaining, inductive reasoning, and machine learning.
This smart computing accesses past data, such as a device configuration and associated data over time, and analyzes that data with software, such as an algorithm, to identify patterns. For example, the described system, when installed pervasively, and paired with active-shooter training scenarios will intelligently improve its performance over time. While much less common, this same technology can be utilized to detect explosions. The disclosed system enables advanced sensing, communication, and control through outlets and wall switches, and enables a low-cost and discreet gunshot detection system that, when paired with advanced cloud-based functionality, provides valuable insight and control to teachers, students, administrators, and first responders during chaotic active shooter situations. The system, fully integrated into a building's existing infrastructure, is also capable of monitoring environmental conditions, energy usage, movement, and other human behaviors.
First responder services and insights include but are not limited to:
1. Real-time gunshot location identification and tracking
2. Estimation of number of active shooters
3. Identification and communication of shelter-in-place warnings
4. Enable identification and communication of areas for safe and urgent evacuation
5. Identification and communication of ideal evacuation routes
6. Remote, secure, and arc-free electrical power control of buildings, circuits, lighting,
and individual plugs
7. Estimation of number and types of gunshots for medical first-responders
8. Assistance and testing during active-shooter trainings
9. Automatic generation of structure interior map
10. Dog presence alert
11. Number of souls in structure
12. Automatic remote door lock/unlock
13. Car running in garage
14. Whole structure lighting control
The disclosed system processes often sensitive sensor data and generates from it user predicates, most of which are not sensitive and which therefore can be shared with third party service providers such as home security companies, contractors performing predictive maintenance, advertisers interested both in understanding individual users better for reasons of improved product suggestions, and in gaining a better understanding of users as a collective. The latter is enabled by deep insights into the causes and effects related to user observations; how users respond to advertisem*nts and product placements; what motivates end users; and numerous other insights. The disclosed predicates rely on heuristics, which means that they do not depend on large-scale deployments or tagged data.
From these first-level predicates, more complex predicates matching user personalities can be generated. For example, knowing how individual users spend their day at home in the context of their location, actions, search queries, web-site visits, TV shows watched, eating habits, entertainment preferences, as well as their responses to commercials, news and other content, allows unprecedented insights into what motivates people. This will, in turn, improve the provision of services to these users, as well as to others, based on correlation of behaviors and an improved consumer understanding.
This functionality could not have been provided using traditional technology, as existing methods fail to safeguard user privacy and the computation corresponding to most functionalities require either access to large quantities of low-level data (which would be used, at the cost of privacy, in traditional systems) or predicates and an associated permissions-based system, as described in this disclosure.
The disclosed system determines the use of various spaces based on identified use, detected temperature, pressure, humidity, associated energy consumption, and possibly additional or alternative information. This enables the automated distinction of a bedroom from a TV room, for example. It can also be used to identify a space as a basem*nt, attic, garage or an outdoor area. For example, an attic may be very warm in the summer. The direct benefits of installing an attic fan can be computed based on temperature swings of a detected attic space and the associated heating of interior rooms, compared to detected or reported outdoor temperatures, and a resident can be presented with offers for purchase and installation of attic fans, where such offers could potentially be subsidized by electricity service providers wishing to curb mid-day electricity consumption, or simply acting as an aggregator of offers to reach scale. Detecting temperatures in a garden is helpful to determine what plants may be most suitable, to augment national weather forecast data by a dramatic expansion of the number of sensors. Temperature and humidity measurements can be used to identify high fire risk zones.
By limiting the privacy impact of data sharing by use of predicates instead of raw data sharing, and by making both the quantities and the permission requests comprehensible, the end user is able to become responsible for his or her own privacy, being able to make informed decisions that are later policed by way of granting (or not granting) selective access to predicates and capabilities. Accordingly, such techniques will not only help improve privacy, but will also enable the comfortable sharing of data and the creation of new services that would not have been possible without the disclosed system, as end users would not have felt comfortable with the potential risks posed to their privacy. The current system limits the required trust of users to the system coordinating their data collection (referred to herein as “the system”). By the introduction of oversight (e.g., auditing by third parties into the data sharing) and privacy scoring (e.g., by ombudsmen determining the impact of sharing select predicates), as well as making privacy a quantity that third party companies can compete in terms of, the system improves privacy in fundamental ways.
The disclosed system further detects a running engine, whether by sound or by carbon monoxide detection, and automatically opens garage doors or starts exhaust fans. The disclosed system further determines the need for a new muffler or other car repair based on garage-centric sound profiles. It determines what a garage is likely used for, e.g., storing cars, motorcycles, or neither, based on sound profiles. The system further determines that a room is likely a garage based on temperature swings and the sound of garage door openers, and offers to connect the garage opener to user-controlled apps, using IR transmitters in outlets or switches to communicate with the garage door opener.
This, again, can be done using the computation of predicates, which serve as building blocks of functionality, and which will spur innovation and development by enabling third parties to provide services. Some of the generation of predicates will be managed by the system, and others by third party vendors that, illustratively, are operating on a platform controlled by the system; some of the predicates may be available for free to vendors, whereas other predicates may require subscriptions, licenses, or payment per unit of information. As functionality is exposed to third parties, this will lead to growth of services, and as data is monetized, end users will benefit from the onslaught of services driven by the opportunity.
The heuristic techniques work as a rule-based system, where the computation of predicates is based on sensor output, user registration data, already computed predicates, the time of the day, and profile data determined for the user(s) associated with the sensors of a space. For example, when determining the location of a user present in the space. A collection of nodes report measurements to a backend or other data collector. This is either raw sensor data or functions of such data, where the functions are locally computed at nodes or aggregators associated with the network. In our example, several nodes detect a sound, but three of them detect the sound much stronger than the others. Two of the nodes in the network detect a motion that is correlated with the sound, and two other nodes detect motions that are not correlated with the sound. Among the two nodes that detected the correlated motion, the one with the strongest sound signal is selected as being most strongly associated with the location of the present user, and that node is output as corresponding to the location of the presence of the user.
In a network where relative location of nodes has been established, e.g., from series of measurements as the above, a collection of sensor measurements are used to triangulate the location of the user, based on a selection of a small set of the nodes having been selected as being close to the user. If there are several users that are active in a space, the above algorithm generates multiple locations based on correlations between sounds and motion data. Other sensor data can also be used, e.g., an inertial sensor can be used to detect vibration associated with a person walking, and correlated with sound data. In another example, sound data is available, but no motion sensor data. This still enables the system to determine a likely location of a user, by correlating detected sounds with each other and identifying the location of the strongest signal or the signal with the greatest correlation with the others.
Correlation is illustratively performed both in the temporal and frequency spectrum, just as for other embodiments. The node with the strongest signal or correlation is selected as the location of the user. The system can also detect location by identifying mobile hardware, such as phones, based on changing location (in terms of node identity) of the signal strength associated with the mobile hardware, as identified, for example, by determining what node has the greatest signal strength for the reception of radio signal, wherein a pseudonym is extracted from the radio signal or two different radio signals are distinguished from each other based on the different signals they correspond to. The node with the strongest radio signal is selected as the location of the user corresponding to the mobile hardware.
In some embodiments, the system integrates multiple location determinations such as the ones described here and selects a location using a voting strategy, where each node that “wins” one of the described “competitions” receives one vote and the node with the maximum number of votes is selected as the location of the user. A more advanced version of this algorithm assigns three votes to the winner, two votes to the runner-up node, and one vote for the node with the third-highest value of the location assertion, whether based on the signal strength of one type of sensor or multiple types. The node with the largest number of votes wins. This is illustratively performed for each detected signal. For example, if there are two users in a space, and both emit sounds, make moves, interact with radios using hardware they transport, and exhibit other similar behaviors, then the signal collections are clustered based on the correlation of the inputs, after which the above determinations are made for each one of the clusters.
In addition to determining the approximate location of a user, a pseudonym is illustratively computed for each user. The pseudonym can be short-lived, and simply used to correlate one location detection to one that is made right afterwards. Such a pseudonym can be determined based on the FFT of the sound data detected, for example. This allows the system to “follow” a user as she moves around in the space, determining a series of locations. This is helpful to determine what the user is doing, whether the user is stressed, and how the locations are related to each other, among other things.
A better and more long-lived pseudonym is one that is based on a MAC address or other hardware identifier, or a session identifier associated with a WiFi connection or a software identifier based on the profile of the data being communicated using the detected device; for example, this profile can be used to distinguish watching a movie or listening to streamed music from being on a phone call or simply walking around without actively transmitting data, the only signals corresponding to automated maintenance connections between the mobile hardware and the system. In addition, the system uses available DNS data to determine sessions and to identify likely activities. To the extent that search data is available in an unencrypted format, e.g., in the URLs being requested, such data is also used both to identify sessions and to determine preferences, interests and activities.
All these kinds of data can be made available in the form of predicates. A predicate, in other words, is a type of data that could be a raw data type, but in almost all instances, is a derived data type, generated by combining two or more types of data (including predicates), and/or by aggregating data from one or more data sources, by applying a function, such as FFT, to one or more data sources, or a combination of such methods.
Once a location (corresponding to the “presence” predicate) and a pseudonym are computed, these two predicates may be used to compute other predicates, along with raw sensor data. Many of these derived predicates use multiple previous predicates in the form of a history. For example, the determination of what is likely to be a bedroom is performed based on detecting a long-term presence with a very weak signal (not much movement, not much sound, and with periodic lack of signal), illustratively during the same time of the day (or night) for a series of days. For example, the system may determine that a user enters an area at 10:35 pm, and that a user, presumably the same user, exits the area at 6:00 am, and that there is intermittent minor movement and sound in between. On a second day, the system, may detect the arrival of a user at 10:25 pm and the departure of a user at 5:52 am, with a low-strength sensor signal associated with the space in between. After a few days of largely consistent determinations of this type, this area, defined as the space covered by a small set of nodes and associated sensors, is determined tentatively to be a bedroom. As more classifications that are consistent with this classification are made, the certainty of the classification improves. This can be expressed as a score, a probability, or as a variance computed from different individual observations.
The classifications can be made available as predicates. As part of predicates, there can be an aspect describing an estimate of the accuracy of the classification, such as the score, probability or variance. This applies to most types of classifications, and associated predicates.
The energy profile associated with an outlet is indicative of the use of the outlet. By computing FFTs of the energy consumption, the system distinguishes common appliances (such as refrigerator, microwave, hair dryer, and dishwasher) from each other, and associates these with the outlet used. This is done by comparing the computed FFTs to stored FFTs obtained from known appliances and stored in a database associated with the backend. This database illustratively contains FFT profiles associated with multiple brands and models, and also, FFT profiles associated with common types of malfunction or needs for repair. In addition to FFTs, the system also stores temporal data indicating the periodicity with which power is drawn, e.g., every 30 minutes (as might be typical of a compressor for one freezer, during a time the freezer door is not opened) or every morning around 7:00 am (as may be associated with a toaster, a hair dryer, or an on-demand water heater).
This profile data is used to determine the likely function of the area where the node is located (e.g., kitchen or bathroom); the likely type and model of the appliance; the likely state of the appliance, including being recently introduced in the system, having recently changed in terms of its power consumption profile; and other features and characteristics. This is illustratively correlated with the pseudonym of the user associated with the appliance, as applicable, generating one or more predicates accordingly.
The system determines preferences of users associated with a space by identifying sound signals, illustratively based on their FFTs, as being correlated with genres or individual content pieces. For example, the system identifies likely music based on the FFT spectrum associated with the detected sound matching a general music FFT profile. This will also overlap with the spectrum of many movies. However, temporal aspects can be used to distinguish the two. For example, music is commonly broken into series of a length of a few minutes, where two series are commonly separated by another FFT profile, such as that of conversation or one of relative silence. Music commonly also changes in terms of spectrum every few minutes. In contrast, a movie does not exhibit these temporal aspects, but commonly has other indicators. For example, comedies can be detected by the sound profile associated with laugh track.
A more fine-grained determination can be made as well, wherein the system distinguishes likely jazz music from likely rock, and distinguishes a sitcom from a tennis match, based on FFT comparisons with representative segments of content stored on the backend, and based on temporal determinations such as the periodicity of detected sounds (whether laughter; a tennis ball smack; a drum; etc.) Furthermore, the detection and classification of the sound is also done in the context of multiple nodes, enabling an improved determination of the layout of the space. Furthermore, as the system identifies the likely location of the sound source, this is used as input to the determination of the use of a space, e.g., a TV room is where TV sound is commonly heard. Since people may have a TV in other rooms as well, each room can have multiple classifications, e.g., a room may be both a bedroom and a TV room, or both a kitchen and music listening room.
The system correlates the energy consumption patterns with the use of a TVs, stereos, etc., in order to determine the location of the associated equipment. This is useful for control of the system. For example, when the system determines that nobody is at home, some appliances such as TVs may have their power supply reduced or turned off by the system in order to reduce the consumption of phantom power. However, the system would not turn off the power supply to a router, a refrigerator or a yet-unclassified appliance. This demonstrates another type of benefit associated with an automated determination of functionality, whether of appliances or areas of use. The determinations described above can be shared with third party service providers, assuming the data owner grants permission.
The determination of the type of room a node is associated with is helpful for the system to interpret sensor output associated with this space and initiate actions. For example, determining that a space is a garage simplifies the classification of observations made after that. The sound of a motorcycle in a garage, for example, suggests that the resident may have a motorcycle; the same sound in a TV room does not imply the same conclusion. Similarly, the sound of a broken muffler in a garage is a meaningful observation in the sense that it could lead to the generation of a coupon for a replacement muffler; the same sound in the kitchen does not have the same meaning. The sound of cries for help are important if the TV is not on, but should be ignored if the TV is on and the system identifies the movie as one in which somebody cries for help. The motion and inertial sensor output associated with a fall is much more likely to correspond to a fall if it is observed in a room that is identified as the shower room, bathroom, or stairway, and following the sound of water being used. In contrast, the sound of impact and inertial sensor output suggesting a correlated crash to the floor is likely to be less risk in a children's playroom, as it may simply correspond to a child jumping off a bed. The detection of correlated inertial sensor output for all sensors is likely to correspond to an earthquake, or possibly, an explosion; however, the localized detection of such output is likely to mean that something or somebody fell to the floor.
Observing series of detections helps disambiguate high-risk events from their potential false positives. For example, detecting a series of impacts in a playroom is much less indicative of the risk of a person falling that the detection of one such impact. Detecting loud voices or cries before a fall is much more likely to correspond to an altercation than a detected fall without loud voices or cries. Similarly, the same detection in the context of a space with only one detected user is almost certain not to correspond to an altercation. However, a risk of falling is much more significant for a space with only one apparent occupant than a space with multiple occupants, just like the risk of falling is not of great concern if it is followed by normal motion, such as walking around than if it is followed by a user being stationary or near-stationary. The detection of cries for help is more significant in context of higher risk. Thus, to suppress false positives, the system illustratively determines the degree of risk, based on heuristics such as these, when evaluating potential indicators of a user needing help, such as cries for help.
The use of predicates can be customized to conveying select events and signals, with the permission of end users. Anonymized data can also be shared based on granted permissions, as described in this disclosure.
The raw sensor data is typically sensitive from a privacy perspective. For example, moment-by-moment motion data may be uncomfortable for many people to share with non-trusted parties, as is sound data potentially revealing conversations. However, general needs such as a need for a new muffler; general preferences such as a preference for R&B over classical music and tennis over football; risks such as the risk of burglary; as well as locations such as a bedroom being adjacent to a study, are not very sensitive in terms of their privacy implications. At the same time, advertisers and service providers find these non-sensitive predicates much more valuable than the raw sensor signals, as the predicates correspond directly to needs and opportunities. For the same reason, users are more likely to be willing to share such data, as it helps them obtain benefits such as predictive maintenance, get discount coupons for events and products of relevance, and get help and protection when needed.
The disclosed system creates a platform for sharing of non-sensitive predicate data, and the protection of sensitive sensor data and user registration data. The system helps selectively share such predicate data in the same way that the current smartphone architecture invites app creators to access some device data while protecting other types of data, thereby enabling a new economy based on secure and privacy-respecting sharing of curated data. External service providers such as burglar alarm services benefit from this, as do service providers such as contractors performing maintenance, and landlords interested in making sure that their property is protected.
In one embodiment, sensor data is locally processed, e.g., one or more nodes comparing the sensor outputs they observe to determine the location of a user. In another embodiment, sensor data is centrally processed, e.g., all sensor data being transmitted to a cloud computing resource or centralized server. In a third embodiment, a combination is performed, e.g., local nodes determining what signals to transmit based on one or more criteria, and based on the phase of the system, where the system may transmit more data in a discovery or configuration phase than what it would in a later “usage” phase.
The central processing corresponds to at least one of a cloud computing environment managed by the system; one or more servers under control by the system; one or more servers associated with a third party service provider; or a combination of these. In addition to the processing of sensor data, other types of data can also be processed in these contexts. For example, data from a data source such as a social networking account or an email mailbox can be processed as described above, whether along with sensor node data or in a context that does not include sensor node data. The disclosure may refer to a specific data source, such as a microphone or other sensor associated with a node, to illustrate a principle and describe functionality that can be achieved using the described principles. However, although the functionality may be very different when applying the same or similar principles to other types of data, the underlying principles disclosed herein can be similarly applied and will have similar benefits for other types of data. For example, the principle of creating predicates and aligning permissions with these is not specific to the context of sensor data, as a person of skill in the art will readily recognize and appreciate; and the examples herein are not in any sense intended to limit the scope of the invention to the illustrative embodiments herein.
In some embodiments, at least a portion of the processing is performed locally, on the system, as opposed to on a backend. This is to ensure adequate performance of the system even in the case of degraded or disrupted Internet connection. In the case when determinations, classifications or generations of predicates, as described in this disclosure, are performed on a local system, it is beneficial to transmit the inputs (or a description of these) that caused a decision to be made, along with the decision or a description of this, to the backend, once this is possible. This enables an audit trail that is important in case of catastrophic situations and system failures, as well as the opportunity for the system to fine-tune the functionality and respond to end-user complaints relating to perceived failures. This is due to the fact that the situations can be re-enacted, simulated and analyzed on the backend.
Thus, such information is illustratively transmitted, to the extent it is possible, and stored in a log associated with the end-user system. To the extent that the processing leading to the determination is made by the backend in the first place, the saving of the relevant inputs and the resulting output or decision are likewise stored in the log. In both of these contexts, an input is considered relevant if it is commonly or always a factor associated with the decision of the associated type, is used as an input to the rule, or is otherwise associated with the decision.
The local processing may generate predicates that are transmitted to a central server corresponding to the system or a third party service provider, and may also be used locally, as part of computation that influences the state of the local system, whether with the participation of the central system, one or more third parties, a combination thereof, or without such participation.
Additional illustrative embodiments will now be described with reference to FIGS. 1 through 8.
FIG. 1 shows a collection of sensors 100 and a set of predicates 120 derived from at least some of the data provided by sensors of the collection of sensors 100. The collection of sensors 100 illustratively corresponds to one or more network nodes, which may comprise, for example, an outlet, a switch, a breaker, or an appliance such as a router, a smart appliance, a portable user device, etc. A given network comprising multiple such network nodes may correspond to multiple distinct collections of sensors from which data is being obtained and analyzed. The analysis can be performed, for example, on the individual node generating the data, on a node or other device that connects to one or more nodes that generate the data, or on a backend server that is connected to the network of nodes. The connections can be made at least in part using built-in radio units, such as a radio unit that is both a sensor in that it detects radio signals, and a communication port from one node to another, forming a local network that is connected using the Internet or other network to a backend server.
The collection of sensors 100 comprises one or more sensor units, where example sensor units include a sound sensor 101, a camera 102, a motion sensor 103, a radio unit 104 that may use WiFi, Bluetooth, Bluetooth low energy (BLE), near-field communication (NFC), ZigBee, and other types of radio; and which may comprise one or more of such units mounted on one or more separate boards associated with the collection of sensors 100. Additional sensors of the collection of sensors 100 comprise a humidity sensor 105, a pressure sensor 106, an inertial sensor 107, a carbon monoxide (CO) sensor 108, a tamper sensor 109 that detects that one or more of the sensors associated with the collection of sensors 100 is physically manipulated, or that the conveyance of the signals from the collection of sensors 100 is physically manipulated. The collection of sensors 100 further comprises a temperature sensor 110 and an energy sensor 112. The energy sensor 112 is configured to detect energy consumption or other types of energy use. Also included in the collection of sensors 100 is a user GPS sensor 111. Such a sensor is illustratively accessible via an API, and in some embodiments comprises a device that has GPS functionality and is associated with a user of the system. An example user GPS sensor 111 is the GPS unit of a handheld device such as a phone, which is accessed by the system via an API to access the location of the user. Although single instances of these and other sensors are shown in the collection of sensors 100, there can be multiple instances of one or more of the sensors, as well as multiple collections of different sensors, in other embodiments.
The various sensors illustrated in the collection of sensors 100 are examples of what are also referred to herein as “sensor devices.” One or more of the different sensor types shown in the collection of sensors 100 can be configured to provide proximity information, and a given such sensor is an example of what is more generally referred to as a “proximity sensor.” Other types of proximity sensors can be used in other embodiments. Some sensors disclosed herein are implemented as respective IoT devices.
It should be noted that the term “predicate” as used herein is intended to be broadly construed, so as to encompass, for example, various types of information in various formats, including those illustrated in the set of predicates 120 of FIG. 1, that are each computed using outputs from multiple sensors. A predicate in some embodiments therefore comprises, for example, information computed as a function of outputs of two different sensors of different types, such as a proximity sensor and a radio sensor.
One or more instances of a presence predicate 121, each also referred to as a presence 121, may be computed as a function of outputs from sound sensor 101, camera 102, motion sensor 103, as well as other sensors belonging to the collection of sensors 100. For example, if a first instance of sound sensor 101 registers a louder sound than a second instance of sound sensor 101 of the same network, such an event is likely to correspond to the sound source being closer to the installation of the node corresponding to the first instance of sound sensor 101 than to that of the second instance of sound sensor 101. At the same time, if motion sensor 103 or camera 102 associated with the same node as the first instance of sound sensor 101 registers the presence or change of location of an entity, this is likely to correspond to a user being in the proximity of the associated node. If there is no output from the sound sensor 101 instance of a first node but there is output from motion sensor 103 or changing output from camera 102 associated with the first node, then this is also indicative of the presence of a user.
In addition, other sensors can be used to improve the estimate corresponding to the presence 121 of a user, e.g., the inertial sensor 107 can be used to detect heavy footsteps; this is particularly useful if the output of the inertial sensor 107 is correlated with the outputs of sound sensor 101 or with motion sensor 103, etc. Thus, by detecting strong correlated signals from multiple sensors associated with a node, a presence predicate is determined, where the signals correspond to the arrival of a person close to the node of at least one sensor of the collection of sensors 100.
By keeping state of the location of the user, and determining what collection of sensors, at any point in time, registers the strongest confidence that a person is present, based on the strength of the signals and their match to the signal profiles typically corresponding to the arrival or presence of a person, the system can maintain an understanding of the presence of users even if they are still and silent, and potentially not visible to a camera. This is because the presence of the user will be detected as the user arrives, but there is no signal from the collection of sensors 100 of the associated node or another node that the user in question has moved away. Having access to serial presences, in terms of the nodes that register presence, of one or more users, the system establishes a physical layout of the network of nodes and their associated collection of sensors 100. This is illustrated in the example arrangement of FIGS. 2 and 3.
One or more instances of a pseudonym predicate 125, each also referred to as a pseudonym 125, may be derived from the MAC address of a device physically associated with a user, as determined by the presence predicate described above and the strength of the input to radio unit 104. A particular user may be associated with multiple MAC addresses. The system determines the common patterns of movement of devices, as represented by MAC addresses or other hardware identifiers, and associates MAC addresses that commonly travel together as being associated with the same user. Additionally, the pseudonym associated with this user, which comprises a locally unique identifier, is illustratively also associated with the types of devices that the MAC addresses correspond to, the nodes where this user is typically active, and potential activities, preferences and other determinations associated with this node and its surroundings.
When a previously unseen MAC address is detected by the system in the form of transmissions observable with an instance of radio unit 104 of any of the nodes in a network, a profile is illustratively created, and information associated with this user is stored. This is beneficial for purposes as distinct as enabling guests to get partial access to the network (such as remote control access to light switches in the room the guest is located, WiFi access, and access to functionalities such as voice-operated alarm clocks for the space the guest is associated with) and for detection of anomalies associated with burglary. In addition, the information can also be used as later evidence of a burglary.
The detection of a previously observed pseudonym (whether in the form of a MAC address or other associated indicator) is helpful for purposes of automated configuration and personalization, while the detection of a pseudonym that is not commonly seen in the network associated with its detection is an indication of a potential anomaly, and could be used for purposes of security as well as for audit purposes, where one audit purpose is the collection of evidence of an intrusion. If the tampering of a node is detected via an instance of tamper sensor 109, this is illustratively used along with the detected anomaly to generate alarms and produce information about the likely intruder. Intruders can be tracked throughout the space, according to an understanding of the layout of the space (as illustrated in FIGS. 2 and 3), and optionally, locks can be automatically engaged, lights turned on or off, legitimate residents alerted, police alerted, and warnings played on the audio system, including on speakers associated with the nodes, separate appliances, etc.
The system can assign a likely identity to a pseudonym by using identity data provided by a user during a setup session such as a time when the user installs a remote control app on his or her mobile device, requests WiFi access, during initial system setup or configuration, as well as other methods. During such sessions, and while accessing any portal associated with the system, the user typically would use a wireless device in range of one or more of the nodes, and therefore, the traffic and the MAC address would be associated with the identity information from the user, illustratively resulting in a record of a given instance of an identity predicate 129, also referred to herein as an identity 129. This record may also include or be associated with demographic data provided by the user.
The system illustratively detects the presence of one and the same user, as identified by her pseudonym and/or identity, over several networks. This enables cross-system customization opportunities, and provides the system with unique insights related to the relationships between different installations. As the system detects the presence of a user with a known pseudonym and/or identity, the system associates detected events, preferences and other predicates derived from the sensor inputs, with the pseudonym or identity, storing this information in a database that can be maintained locally and associated with the network of installed nodes or at a central location with a server that maintains multiple installations of networks of nodes.
The system can determine music type by comparing the sound spectrum to spectrums of different genres of music, e.g., by comparing FFTs or performing image recognition of FFT plots. The system can also maintain identifying segments or FFTs of a set of common songs and match the sound associated with a space with these identifiers. This enables the automated generation of a music type predicate 122. This determination will illustratively be done in the context of the determination of either pseudonym or identity, allowing the system to improve the automated selection of music, or of advertisem*nts related to the musical preferences of the user, such as notifications or discount coupons associated with local performances, as well as other correlated preferences.
It is not always important for the system to determine that a person who is known to be currently present is also listening to the music, but close temporal association is sufficient. Therefore, if the system detects the likely presence of a user within a threshold amount of time, such as 10 minutes, and a musical event such as the detection of a particular genre, then the user pseudonym or identity, and the associated profiles, can be associated with the music type predicate 122. In the case where individual songs are detected, the system can go further than general genre detection, and also identify likely preferred artists, pieces, and the like. This is also information that will be stored with the music type profile of a user.
It is important to recognize that this approach is not invasive of a user's privacy: whereas the system will identify the sounds associated with a space, it will not store these sounds, but only descriptions associated with the determinations. That means that conversation that takes place at the same time as the identified music (or at any other time) will not be stored or interpreted, but will only be noise in the background of the classifier that determines the music type. If the conversation is more prominent than the music, the determination of music type predicate 122 may fail where it otherwise, in the absence of the conversation, would have succeeded. However, whether it succeeds or not, the conversation is not processed, and is not stored. To the extent that the sound sensors are used for detection of voice commands, of course, the system will attempt to identify the presence of such voice commands, whether occurring in silence or in the presence of music, at the same time as it may also be attempting to determine the type of music being played. To the extent that the system has a baseline truth for any observation, e.g., the system was used as a remote control to select the music, this constitutes a tag that allows for the use of additional machine learning, and the determination of the success rate of the heuristics based simply on comparison with FFTs or time segments of sound inputs.
Similar techniques are also used to identify sports programs using sports program predicate 123. For example, a soccer match has a sound profile that is very distinct from that of a tennis match, enabling the use of FFTs to compare the sound associated with a space and that of a known sport or other generator of sound profiles. In addition, the system can compare the sound profiles observed with known broadcasting taking place at the time of the observation, e.g. using standard correlation methods between the broadcast stream (slightly delayed to account for the transmission) and the observed sound sequences associated with one or more sensors. Such arrangements can make use of not only sound sensors, but also, for example, inertial sensors (which detect low-frequency sounds), as well as motion sensors to find correlations between user actions and the broadcast sound profile. For example, the user may cheer after a goal is made, as will the on-site audience of the soccer game.
The system therefore maintains sound profiles associated with multiple sports and other sound-generating events, and compares these to observed sound (and other) sensor data; as well as correlating current broadcast sound levels with observed sound levels. This also enables the detection of other TV shows being watched, using one or more instances of TV show predicate 124. This can either be by type (e.g., identifying a comedy by the tell-tale laugh track signature) or by unique program (by sound profile or by comparing time segments of sound data between an observed sample and a stored sample). It is not necessary to limit this to comparison with programming that is currently broadcast, but the same methods can be used to identify on-demand programming. It is beneficial, as described before, to associate music, sports and other programming preferences with pseudonyms, identities, times of the day, other ongoing activities being detected, etc. This can be used for prediction of preferences as well as for detection of anomalies, which signals that a user is maybe not feeling well or has different needs than previously established.
It is beneficial to associate apparent TV programs with a location, which enables the determination, based on signal strength and presence detection, of a room being a TV room 126. As rooms are identified, the data is associated with a set of nodes, which are determined to be in that room or close to that room based on signal strength and consistency of signal strengths for detected sounds and other sensor data. As this is done, the system automatically clusters nodes in terms of their relative distance. The system can also store information relating to the certainty of such a classification. In addition, the system illustratively determines the relative location of nodes that are not in the absolute physical area of each other, but may be in rooms that open up into each other. This is further illustrated in the example arrangement of FIGS. 2 and 3.
The system can determine that a room is likely a bedroom 127 based on the apparent presence of a user during night hours, where this user is inactive for long periods of time. However, a user who works at night and sleeps during the day will also be correctly identified as sleeping, and the room in which he or she sleeps will be labeled accordingly. Thus, the determination of bedroom 127 is illustratively based on an activity profile comprising both a sound and motion profile, and potentially also inertial sensor data indicating tossing and turning or snoring, along with the duration of time during which these inputs are rather consistently observed. This enables the likely determination of a room being a bedroom 127, but also helps identifying anomalies in sleep patterns, which can be used to automatically modify environmental configurations, such as the temperature in the room identified as the bedroom 127. By identifying sleep patterns of a user, the system can also modify, within some interval, the time at which the user is woken up by an alarm he or she has set, as the alarms will also be controlled by the system.
In case a user is suffering a medical high-risk situation, as determined by the system, medical help can be requested. This is only possible if the system is able to guarantee very low error rates, which is one of the reasons that it is valuable for the system to identify all events and preferences and then identify anomalies associated with high risk. This can be done only for users who are determined to be alone; only for users expressing fear (e.g., by voice command); or only in contexts of identified high risk for distress 133.
Detecting the location of a bedroom and the likely event of a user sleeping, which is illustratively done in a fully automated manner, improves the system's opportunities to protect users when they are the most vulnerable, namely, when they are asleep, and to guide them properly in the case of a high-risk situation. It also places higher requirements on privacy. Thus, the system can process data obtained from rooms determined to be a bedroom 127 in different ways from ways in which data is processed when obtained from nodes associated with other rooms. The disclosed system therefore learns as much as it can about its inhabitants in order to personalize their use of the space as well as to protect them, whether from security or safety risks, or from exposure of personal information that would hurt their privacy.
A room can also be determined to be a kitchen 128 by being used in a manner consistent with kitchen use. This includes being used soon after a user leaves the bedroom in the morning (for breakfast); for having sound profiles from sound sensor 101 indicative of a kitchen (the humming of a dishwasher or the gurgling of a coffee maker); heat profiles from temperature sensor 110 indicative of the use of a stove or oven; the energy consumption profile from an energy sensor 112 matching the operation of a microwave; a refrigerator; a dishwasher, etc.; the correlation between such inputs (e.g., sound and energy consumption, as described above); and possibly others. Additional sensors such as humidity sensor 105 are also used to determine cooking, and are correlated with heat detected by temperature sensor 110. A room that is identified as a likely kitchen 128 faces a different use and risk profile than other rooms, and the system adapts its responses to observed events in the context of knowing the likely room type. In addition, by observing activity in the established kitchen 128, the system can determine the extent to which users eat at home vs. eat out; the extent to which users cook during weekdays; the type of appliances typically used for food preparation; and so on. This informs the risk profile as well as indicates demographic inferences and preference inferences for the users associated with the space, which is helpful for improving the system in terms of customizing configurations, services and content for the users.
As described before, the association of a collection of sensors with a given identified room, as also illustrated in FIGS. 2 and 3, allows the system to better understand the actions of a user and minimize the risk of false positives for inferences being made. For example, detecting likely distress in the context of food being prepared suggests very different actions than detecting likely distress in the context of a user rushing through his or her morning schedule, but at a slightly later time than usual. In the first situation, the user may have cut or burnt himself or herself, and may need first aid guidance or an ambulance; in the latter case, the user may need a reminder not to leave the house with the stove turned on; alternatively an automated or remote method of disabling the stove.
The system can detect that a room 130 has a washer, dryer and/or a water heater (“w/d/h”) based on sound profiles determined using one or more instances of sound sensor 101 being matched to stored appliance-specific sound profiles; by energy consumption profiles determined using one or more instances of energy sensor 112 being matched to stored appliance-specific energy consumption profiles; and by correlating the temperatures in the room as determined by one or more instances of temperature sensor 110 to stored appliance-specific temperature profiles; and by correlating the profiles of the different types to each other. For example, when the sound profile is correlated with the energy consumption profiles and the temperature profile with a correlation factor exceeding a threshold associated with a stored appliance-specific threshold, then this is indicative of the associated set of sensors with such outputs being in a space that is labeled as being a room with these appliances, such as a washer and dryer or a water heater. Similar profiles can be detected for other appliances, such as HVAC appliances.
In addition, the sensor profiles associated with these appliances can be correlated to external system requests to control set appliances, e.g., to turn up the AC in one room or to increase the operational temperature of a water heater due to the temperature of delivered water falling according to a temperature sensor believed to be located in the bathroom detecting a temperature increase that is slowing down, or which is not consistent in its increase with the humidity in the room, which in turn is indicative of the amount of water being used.
A room is identified as a shower/bathroom 131 based on the system detecting increased humidity and/or sounds indicative, whether in time or frequency space, of showering, taking a bath, flushing, or other uses of water, as well as changes in temperature accompanying these other inputs, and of correlations between changes of all of these types of sensor output. It is beneficial to know that a room is a bathroom; for one thing, the presence of water in bathrooms increases the risk of accidents such as falls 134, and therefore, the system will be configured to identify, using instances of motion sensor 103 from nodes in a shower/bathroom 131 that a user is potentially falling. Whereas users can also fall in other rooms, the risk is greatest in rooms with water, and rooms associated with stairs, and therefore, the identification and labeling of such rooms is important. Stairs can be identified based on sound detected by sound sensor 101 and motion detected by motion sensor 103 for nearby nodes. Inertial sensor 107 can also be used to detect impact, which results from falls. Multiple instances of these and other sensors can also be used.
An entry room 132, also referred to herein as an entrance room, can be identified based on sounds detected by sound sensor 101 associated with the opening and closing of doors, the operation of locks, the presence of remote-controlled locks, changes in humidity detected by humidity sensor 105, especially when correlated with associated sounds detected by sound sensor 101 matching the opening and closing of doors. It will also be identified based on sudden temperature changes detected by temperature sensor 110, pressure changes detected by pressure sensor 106, and the correlation of these with each other and with other sensor outputs associated with opening and closing of entrance doors. Knowing that a room is an entry room 132 is important in the context of understanding, in an automated manner, the layout of a space, as illustrated in FIGS. 2 and 3. This is beneficial for burglar alarms.
If a presence 121 is established in an entry room 132 while the alarm is turned on, and the identity 129 or the pseudonym 125 is not detected or recognized as a resident, then this is an indication that the alarm should sound or an alert should be generated. However, if a presence 121 is established in an entry room 132 while the alarm is turned on, and the identity 129 or the pseudonym 125 is detected or recognized as a resident, then a better response is the generation of a notification to the resident, to which the resident based on configuration may have to respond with a confirmation for an alarm not to be generated. However, if a presence 121 is established in a room that is not an entry room 132 but which may only have windows, then this is an indication of burglary, and should result in an alarm whether the alarm is turned on or not.
This demonstrates an important way that the system can automate and improve the protection of users based on an automated understanding of the space, its layout, its users, and the meaning of various critical events. If an intruder is detected while a recognized user is present, based on a detection of the recognized user's pseudonym 125 or identity 129, then the system illustratively takes another action than if there is no indication that a user is present. For example, if no user is present, all locks may be locked and all lights may be turned off, along with law enforcement being notified; on the other hand, if a potential intruder is detected and a user is present, a prudent response includes notification of the user, comprising camera footage of the detected intruder and his or her determined location, based on a presence determination. A person skilled in the art will recognize that these are simply exemplary descriptions of functionality, and that they demonstrate the benefits of the disclosed structure, methods and approach. If an increased risk of burglary 136 is detected as being likely, then this is communicated to the backend, and based on the configuration of the system, conveyed to an operator.
The system further detects increased risk of fire 135 based on output of temperature sensor 110 indicating increased temperature, energy consumption changes detected by energy sensor 112 indicative of melting of cables or overheating of appliances, failure detection of appliances as described above, sound profiles based on outputs of sound sensor 101 indicative of fire, changes in humidity or pressure from respective humidity sensor 105 and pressure sensor 106 indicative of fire, as well as combinations of these, and especially correlations of these types of sensor data, from one or more nodes in the network. As for many other disclosed types of detection, it is beneficial for the system to identify such risks in the context of the labeling of the room and the likely presence of gas or other flammable substances where the likely fire is detected.
As for other risks, risk of fire 135 is illustratively reported to an automated backend system and also to a human operator, in addition to sounding local alarms and automatically generating notifications for emergency responders in proximity of the residence or business being observed. A pinpointing of the location of the fire is beneficial, as well as all location data of potential users in the space. This is also illustratively reported, and can be used by emergency responders to prioritize and guide their efforts.
The risk of an earthquake 137 is important to automatically detect, which is done using data from one or more instances of inertial sensor 107, one or more instances of sound sensor 101, and one or more instances of motion sensor 103. The motion sensor 103 or multiple instances thereof are primarily used to identify the likelihood that shaking detected by inertial sensor 107 is due to a person falling or dropping something. Whereas motion sensors generate local outputs if a person falls or drops something, inertial sensors generate the same principal output no matter where they are located in the house or the neighborhood when the cause of the shaking is an earthquake. When an earthquake is detected, its strength is illustratively compared to a threshold and it is determined whether the likely earthquake is sufficiently large; if it is, an action is illustratively taken. This action comprises turning off the electricity delivery to all nodes where there is a risk of a gas line rupturing, e.g., in a kitchen 128 or a room 130 with a washer, dryer and/or a water heater for example. If the detected earthquake is stronger than a threshold and takes place at a time when it is dark outside, other rooms are illustratively lit to facilitate the evacuation.
Illustratively, only outlets and electric nodes connected to light sources are enabled, and electricity to any appliances posing a risk in the context of an earthquake is turned off or reduced until reset by a user or an admin, or until a sufficient amount of time has passed, where this time can be a function of the detected severity of the earthquake 137. Notifications are illustratively generated in rooms with user presence, and potentially in all rooms; along with notifications being sent to user devices. It is beneficial for the system to maintain power to nodes that are used to power a router or other communications equipment, as well as to nodes used to generate a mesh network for communication within or between protected buildings. This maintains communication channels at times when this is important for users threatened by risk.
Additionally, the system is configured to identify and locate pet-related activity 138, e.g., based on motion sensor 103, sound sensor 101, or multiple instances of these and other sensors, and the pattern of movement associated with the derived presence 121 of the pets. In situations where pets are chipped, the pets can also be detected by identification using wireless radio of their implanted chips. Knowledge of the existence of a pet or better presence 121 or location of a pet is beneficial in emergency contexts, such as when there is a detected fire 135, and is illustratively communicated to emergency responders along with information, when possible to derive, relating to what type of pet it is. That can be determined both based on movement profiles (as indicated by serial readings of distributed presence 121 determination) and sound profiles determined using one or more instances of sound sensor 101.
The system is also configured to determine events such as party-related activity 139, which can be determined based on sound profiles and motion profiles using respective sound sensor 101 and motion sensor 103, and camera 102 output, the presence predicate 121 of multiple users, some of which will be detected using pseudonym 125 based or identity 129 based methods; additionally, the system can determine based on the detected music type predicate 122, sports program predicate 123 and TV show predicate 124 and the locations of presence according to the room labeling illustrated in FIG. 3, that there is a party. Parties may sometime increase certain risks, and accordingly, detecting a party is potentially beneficial. A person skilled in the art will recognize that these are simply exemplary embodiments of the potential of the disclosed system and its benefits, though, and will recognize that the described methods and architecture can be applied to a variety of related contexts, and that this results in similar benefits to those described in this disclosure.
FIG. 2 shows an illustrative example of a set of nodes associated with a space. The nodes in this embodiment include a set of nodes comprising Node 1, Node 2, Node 3, Node 4, Node 5, Node 6, Node 7 and Node 8, also denoted by reference numerals 201, 202, 203, 204, 205, 206, 207 and 208, respectively. It is to be appreciated that other embodiments can include more or fewer nodes. Here, two nodes are connected with a line if they are determined to be in close proximity of each other, as determined by at least one of very high correlation of sensor output signals and sensor output signal strength; and sequential activation in terms of the node with strongest signal output for a selected one or more types of signal types, indicating that a person or other source of sensor input passes from one to the other node neighborhood. This is a self-organizing network, and is determined automatically by the system, as described above and in further detail herein. It also benefits from a labeling of rooms in terms of their functionality, e.g., determining that one of the rooms is a kitchen, and some collection of nodes is associated with this rooms, e.g., by high correlation of sensor outputs or sensor output strengths, or correlated and simultaneous detection of motion or other local events.
In the example shown in FIG. 2, each node corresponds to one or more sensors, and illustratively several sensors of several types. The nodes may correspond to the same set of sensors or to different types of sensors, where some nodes may not have overlapping sensor types whereas others will. It is beneficial for most nodes to correspond to as many types of sensors as possible, as this gives a richer input to the system. It is also beneficial with a large number of nodes, and situations where these nodes are co-located or located in the same room or within radio distance and/or earshot from each other. Nodes having speakers can emit low volume sounds, inaudible frequencies, and other nodes, equipped with one or more instances of sound sensor 101 can detect the sounds emitted by nearby nodes. By modifying the volume of the sounds and generating sounds of different pitches, and from different nodes, the system can self-organize a network layout map like that shown in FIG. 2. Different pitches of sound have the property of penetrating different materials to different extents, since they are absorbed by furniture and other objects to different extents. Thus, using different pitches to determine the co-location of two nodes not only allows the system to determine likely presence in the same space or room, but also allows a determination of the extent of obstruction in the room, where the obstruction is of furniture and other articles. Since different materials absorb sounds and radio of different wavelengths to different extents, the system also determines the likely type of furniture, e.g., tables and chairs vs. couches; and the likely presence or absence of wall-to-wall carpets and area rugs. This is also informed by sounds made by users walking in the space, and the extent to which movements are audible.
A similar type of probing and analysis can also be done using radio transmitters/sensors of radio unit 104 with low or variable signal strength. The automated organization of the network map shown in FIG. 2 is also made based on observations of events not generated by any of the nodes, as previously described. Node 1 is determined to be adjacent to Node 2. Node 2 is also determined to be adjacent to Node 1, Node 5, Node 6, Node 3 and Node 8. Node 3 and Node 8 are determined to be likely in the same room or small area as they are both adjacent to Node 2 and Node 4, as well as to each other. Node 7 has not yet been found to be adjacent to any of the other nodes, and may be in a room that is rarely used. This may be a storage room, for example, or a rarely-visited utility room.
FIG. 3 shows the nodes of FIG. 2, but with an associated classification of room type, illustratively including rooms 301, 302, 304 and 305, in accordance with a physical layout of at least a portion of the area. Undetermined room 301 comprises Node 1. Room 305 comprising Node 5 is determined to be a bedroom with a likelihood exceeding a threshold that may be set to a particular value, such as 0.8. Room 302 is determined to be a living room, and comprises Node 2, Node 6, Node 3, and Node 8. Based on correlations between sound that is determined to be associated with a TV, and power consumption from Node 8, which is illustratively an outlet, the system determines that Node 8 is connected to the TV. This enables the user to remotely turn off the TV simply by instructing the service associated with the network to turn off the power to Node 8, or turn it to a sufficiently low voltage that the TV goes into hibernation and is only woken up when the system determines that a user enters room 302, based on presence determination, or when a user utilizes the remote to turn on the TV. Here, the remote may be implemented using any device with network connectivity, allowing the connection to the service provision using an app, a webpage, or associated method. Node 2 in the living room (room 302) is illustratively known by the system to be a switch, and is located next to the door to the living room, as also evidenced by its proximity to Node 5 in room 305, which is determined to be a bedroom. Node 6 is illustratively an outlet whose power consumption profile is associated with a refrigerator. Thus, room 302 is not only a living room, but also likely has a kitchen portion. Alternatively, the space may be a dorm room without a real kitchen facility, as evidenced by the fact that the occupant appears to go out for meals in a predictable manner, and there is no evidence of the space being used for food preparation. Room 304 comprises Node 4, which is illustratively a smart door lock with a camera.
Based on an adjacency matrix associated with the connections between nodes, the system knows that the entrance room is connected to the living room, which in turn is connected to the bedroom. Based on traditional configurations of homes, the system may tentatively assign room 301 to being a study. Based on the absence of connectivity between Node 7 which is illustratively a GFCI outlet, and other nodes, the system may tentatively assign Node 7 to being located outside the home corresponding to rooms 301, 302, 304 and 305; it may, for example be an outlet in the garage. Thus, even nodes that cannot be assigned a room designation or determined to be adjacent to other nodes can sometimes be given a probabilistic assignment of roles based on traditional layouts, the type of node, and other auxiliary information.
FIG. 4 shows how a user can connect to a node 403 in a local space comprising one or more nodes arranged in a network, using a device 401 such as a smartphone or a tablet, and from the node 403 to a service provider 405, which in turn is connected to an advertiser or advertising network 406. The connection from the user device 401 and node 403 is typically made using a central node 404 of the local space 402, where the central node 404 may be a router. The user connects using user device 401, using a webpage or an app, which causes a connection to be established with service provider 405. The user uses this connection to perform customization, to obtain access to access data or control nodes, to register an account, associate an account with a space, to enable service provision, or other related activity in which the user selects, creates or modifies an account. Illustratively, the user is also requested to pass an authentication, establish one or more credentials, or request access from another user, or a combination of such functions. As a connection is established, the system, controlled by the service provider is configured. In addition, a connection is illustratively made between the user device 401 and the advertiser or advertisem*nt provider 406. This enables automated determination of device identity, generation of a device identifier string, and detection of other activity related to creating an association.
This can be performed at advertiser or advertisem*nt provider 406 between an identity or pseudonym provided by service provider 405 to advertiser/advertisem*nt provider 406 or provided by advertiser/advertisem*nt provider 406 to service provider 405, where the device identity is related to the user device 401. In other words, this allows the advertiser/advertisem*nt provider 406 to recognize user device 401, and enables service provider 405 to convey predicates and other information related to user device 401 to advertiser/advertisem*nt provider 406, using the pseudonym established between the service provider 405 and the advertiser/advertisem*nt provider 406. The device identity can use cookies, cache cookies, user agent, flash cookies, unique identifier information stored in an app, and other related methods, as will be appreciated by a person skilled in the art. An app, furthermore, can generate connections to advertiser/advertisem*nt provider 406 using a webview interface, which corresponds to browser traffic.
The establishment of the connection between the user device 401 and the advertiser/advertisem*nt provider 406 can be initiated by the service provider 405. For example, the service provider 405 can redirect traffic between the user device 401 and the service provider 405 to the advertiser/advertisem*nt provider 406 using a redirect directive, which can be performed in an 0-sized iframe if a browser connection is used, or by opening a webview connection from the user device 401 to the advertiser/advertisem*nt provider 406 if an app is used. A person skilled in the art will appreciate that there are alternative methods of establishing such connections, and that those can alternatively be used instead.
The conveyance of the pseudonym can be performed using a direct connection between the service provider 405 and the advertiser/advertisem*nt provider 406, or can use the user device 401 as a proxy by redirecting traffic to a unique webpage associated with the advertiser/advertisem*nt provider 406, wherein a unique element of the unique webpage comprises a pseudonymous identifier. This is well understood in the art. User device 401 may also request access to a resource local to the local space 402, which causes two requests to be made, one which is associated with the response to the request, and which involves the service provider 405 or a node associated with local space 402; and the other which corresponds to a connection to advertiser/advertisem*nt provider 406. This allows advertiser/advertisem*nt provider 406 to create a profile for user device 401, if it does not already have one, where this profile is stored by advertiser/advertisem*nt provider 406 and associated with the local space 402.
To the extent that service provider 405 can determine that user device 401 corresponds to a user profile with another device that has already been registered with service provider 405 or advertiser/advertisem*nt provider 406, the service provider 405 illustratively communicates to advertiser/advertisem*nt provider 406 that that previous device, indexed by its associated pseudonym, corresponds to the currently observed user device 401, allowing the advertiser/advertisem*nt provider 406 to associate both of these devices with the same record. As predicates are obtained by advertiser/advertisem*nt provider 406 from service provider 405, these are associated with the record with which the relevant pseudonym is associated.
This conveyance is illustratively performed out of band from service provider 405 to advertiser/advertisem*nt provider 406, and not involving a user device such as user device 401 as a proxy. One alternative way to direct traffic from user device 401 to advertiser/advertisem*nt provider 406 is using a captive portal associated with local space 402 and managed by service provider 405. The captive portal may redirect traffic from the user device 401 to the advertiser/advertisem*nt provider 406, as described above, or using one of the variations known by a person skilled in the art.
The result of these operations is that the advertiser/advertisem*nt provider 406 will have information about end users including device identifiers of user device 401, and also including predicate data related to the associated user. The advertiser/advertisem*nt provider 406 may additionally obtain a feed of temporary relevant data, such as the current needs and interests of the user; the current or recent activities associated with the user; and demographic and other information associated with the user. This is illustratively related to a pseudonym 125 or an identity 129, as described above. For privacy reasons, service provider 405 illustratively does not provide user identity 129 to advertiser/advertisem*nt provider 406, nor any sensor data directly from the collection of sensors 100, but only related to selected predicates in the set of predicates 120 related to local space 402 being accessed by user device 401. These can include, for example, data related to their optional association with pseudonym 125 or an identity 129 and related records detailing predicates.
FIG. 5 shows a sensor collection group 501 corresponding to at least a portion of the collection of sensors 100, and the predicates 502 computed from data provided by these sensors. These predicates illustratively correspond to at least a subset of the set of predicates 120 computed from the outputs of the collection of sensors 100 using the heuristic techniques described herein. While FIG. 1 shows the generation of the set of predicates 120 from the single collection of sensors 100, it is beneficial in some embodiments to compute the predicates from multiple collections of sensors. The system also obtains optional user-provided inputs 503, e.g., using the registration feature, or using manual configuration, manual selection of preferences and settings, and similar. The user-provided inputs 503 include information both about recurring users and temporary users. The outputs from the sensor collection group 501, the computed predicates 502, and the user-provided inputs 503 are provided to a training unit 504 that generates or fine-tunes a machine learning or ML model 505. This ML model 505 later obtains inputs from the sensor collection group 501, the computed predicates 502, and generates classifications that are made part of the computed predicates 502.
Tagged data in the outputs of sensor collection group 501 are particularly valuable for the training unit 504 to generate or improve the ML model 505, as these provide the training unit 504 with a baseline truth. Additionally, to the extent that high-confidence predicates are generated in the computed predicates 502, these can be used as tagged data by training unit 504. Here, the confidence is illustratively computed using standard statistical methods, such as maximum likelihood, or determined to be high-confidence after having consistent classifications (e.g., of room names) for at least a threshold amount of time. A person skilled in the art will recognize that there are many alternative approaches of determining confidence. High confidence corresponds to the computed confidence, which can be a probability or a score or a collection of values in multiple dimensions, exceeding one or more thresholds that are set by the system. One such threshold is a probability of 95%, and another is a score of temporal consistency indicating that the classifications have been consistent over a period of 10 days, and yet another is that the classifications have been consistent for at least 50 observations of a user that led to a subsequent classification.
FIG. 6 illustrates a training phase 600 and an emergency alert phase 610. Training phase 600 comprises receiving a sensor output time series 601, also referred to herein as “time series data.” The time series data may comprise a series of informational elements, where an element comprises a time indicator indicating the time of creation of the element; a sensor descriptor indicating the type and location, or identity, generating an output value; and the output value. For example, the type of a sensor may be “sonar,” “microphone,” “presence detector,” “radio unit,” “inertial sensor,” etc. The location may correspond to an indication corresponding to an already determined location from layout determination 603. The identity may correspond to an at least locally unique value that corresponds to an identifier that may be generated at installation or manufacturing, and which is associated with a sensor. The output value comprises at least a portion of data generated by the sensor.
The time series data is received by a training unit. The training unit correlates received time series elements in step 602, and uses the correlated time series elements to determine a layout in step 603. In step 611, an emergency alert unit associated with the training unit associated with training phase 600 receives sensor data. In step 612, the emergency alert unit receives an indication of an emergency; this may comprise sensor output, such as from one or more inertial sensors, or it may be obtained from an auxiliary source. In step 613, the emergency alert unit generates a location prediction based at least in part on the layout determined in step 603 and at least in part on the sensor data received in step 611. The generated location prediction may be transmitted to auxiliary recipients, such as emergency responders. The location prediction optionally comprises information related to a classification associated with the sensor data of step 611.
FIG. 7 shows a training phase 700 in which a model is trained and a prediction phase 710 in which a prediction is made. Training phase 700 comprises a step 701 in which a model training unit receives data with explicit identifiers; an example of such data is data associated with a radio identifier such as a MAC address or a Bluetooth identifier, data associated with HTML cookies or data associated with an app. The term “explicit identifier” as used herein is intended to encompass these and other types of information that can be used to identify a particular device and/or a particular user.
The training phase 700 also comprises step 702 in which a tag is generated from one or more data elements comprising explicit identifiers. In step 703, the model training unit receives sensor data that does not comprise explicit identifiers. Examples of such sensor data is sensor data generated by a sonar sensor, a microphone, an inertial sensor, a proximity sensor or a motion sensor. In step 704, a machine learning model or other functionally related model is trained using the sensor data without explicit identifiers, from step 703, and the tags determined in step 702. The training unit stores a trained model in step 705. This trained model is conveyed to a prediction unit, which may be housed in the same hardware unit as the model training unit. In step 711, which is part of the prediction phase 710, the prediction unit receives sensor data without explicit identifiers. In step 712, the prediction unit determines a prediction of an identifier based at least in part on the model generated by the model training unit in training phase 700, and at least in part on the sensor data of step 711. This permits the use of sensor data without explicit identifiers to predict an explicit identifier. Such techniques can be used to identify various types of users, including humans, pets, appliances and household robots. The term “user” as utilized herein is therefore intended to be broadly construed, and can include hardware and/or software entities as well as humans associated with those entities.
FIG. 8 illustrates an intrusion detection technique. In step 801, a layout is determined, e.g., using the approach described in FIG. 6 or using alternative means such as obtaining a user-provided layout. In step 802, the intrusion detection unit receives an indication that there is no presence of a user in a target room. A target room may, for example, be a bathroom. In step 803, the intrusion detection unit receives an indication that there is also no presence of a user in a room adjacent to the target room. An example room adjacent to a target room may be a bedroom that has an entrance to the bathroom that is the target room. In step 804, the intrusion detection unit receives an indication of a user presence in the target room. Thus, this user did not enter the target room through the adjacent room, since there was a preceding indication that there was no presence in the adjacent room. Typically, this means that a person entered the target room through an open or broken window. In step 805, it is determined whether the detected user in the target room is registered, e.g., by determining if the user corresponds to a known identifier such as a radio identifier or a predicted identifier as described in FIG. 7.
In step 806, a comparison is made whether there is a likely match to a registered user. Here, user registration may correspond to an explicit user action or may be based on repeated historical observation of identifiers associated with the user. If there is a match, then the intrusion detection unit initiates a verification in step 808. Such verification may comprise sending the matched user a notification and requesting a response, such as a confirmation of presence, an authentication using biometric techniques, etc. If there is no match, the intrusion detection unit initiates an alert in step 807. This may comprise sounding an alarm, sending messages informing registered users of the potential intrusion, or the collection of additional data, such as video data that may normally not be collected due to privacy concerns.
The particular features and other advantages described above are examples presented in the context of illustrative embodiments, and therefore such features and advantages need not be present in other embodiments.
For example, references herein to “a system” or “the system” in conjunction with various distinct types of features or functionality should not be construed as a requirement that all such features or functionality be present within the same single system. Instead, different systems in different embodiments can include different combinations or other arrangements of the various disclosed features and functionality.
Also, references herein to particular features or other aspects as being “optional” refer to utilization in one or more particular embodiments, and should not be construed as an indication that any other features or aspects, such as features or aspects not explicitly referred to as optional, are required in any particular embodiments.
Illustrative embodiments include systems, methods, apparatus and computer program products comprising non-transitory storage media storing program code.
For example, in some embodiments a system for generating user predicates uses at least one node comprising a set of sensors, said set of sensors comprising at least two of a sound sensor, a motion sensor, a camera, a humidity sensor, a pressure sensor, a carbon monoxide sensor, an air-quality sensor, an energy consumption sensor, a temperature sensor, an inertial sensor and a radio unit.
The system in some embodiments generates a location estimate for a person based on output from at least two of the sensors in the above-noted set of sensors. The system also generates a room descriptor based on the location estimate and output from the at least two of the set of sensors. In addition, the system generates a user action determination based on at least two of the room descriptor, the location estimate and the at least two of the set of sensors. The user action illustratively comprises at least one of determination of a music genre, a sports type, an entertainment genre, a risk of burglary, a risk of fire, a risk of a fall, a risk of earthquake, a need for maintenance, a need for a product, a need for a service, and a need for a notification.
In some embodiments, the system is further configured to generate a map based on at least two space descriptors and at least a time-based series of correlated events, wherein it is determined whether a first space and a second space associated with the at least two space descriptors are connected to each other, and wherein a space corresponds to at least one of a room and an outdoor area.
The above-noted radio unit is illustratively used to generate a user pseudonym associated with a user for which a location estimate has been generated, and wherein the user action is associated with the user pseudonym. In some embodiments, the user identity is associated with the user pseudonym during a registration event.
The system in some embodiments is configured such that at least a portion of the location data is transmitted to first responders when the event is one of the risk of burglary, the risk of fire, the risk of a fall and the risk of earthquake.
Additionally or alternatively, power, gas, or similar utility or safety related functions are automatically turned off in response to detecting at least one of the risk of fire and the risk of earthquake.
The system in some embodiments generates a commercial offer in response to at least one of the music genre, the sports type, the entertainment genre, the need for maintenance, the need for a product and the need for a service.
Illustrative embodiments include a control system configured to receive, from a sensor network comprising at least one radio sensor and one proximity sensor, a first signal comprising at least an indication of a first radio detection and an indication of a first proximity detection. The control system illustratively generates a profile associated with an entity, stores in the profile data associated with at least one of the first radio detection and the first proximity detection, receives a second signal comprising at least an indication of a second radio detection and an indication of a second proximity detection, generates a predicate based at least on data associated with the profile and the second signal, and transmits the predicate to a registered entity.
In some embodiments, no additional information comprised in the first or second signal is transmitted to the registered entity.
The registered entity may comprise, for example, at least one of an advertiser, a third party service provider, and a user associated with the system. As another example, the registered entity may comprise an emergency responder.
A given one of the indications of radio detection illustratively comprise one of a MAC address, a Bluetooth Identifier or a unique identifier associated with a software application, although numerous other indications of radio detection can be used.
In some embodiments, the predicate indicates at least one of the presence of a registered user, the presence of a user other than the registered user, the absence of the registered user or the absence of a user other than the registered user.
Additionally or alternatively, the predicate can indicate an activity associated with an entity associated with at least one of the second radio detection and the second proximity detection.
As another example, the predicate can indicate a classification associated with an entity associated with at least one of the second radio detection and the second proximity detection, wherein the classification corresponds to at least one of a human, a robot and an appliance.
These and other particular features of illustrative embodiments are presented by way of example only, and should not be viewed as limiting in any way.
Further illustrative embodiments will now be described in detail with reference to FIGS. 9 through 17. The embodiments relate to privacy enhancement using derived data disclosure, and illustratively utilize predicates of the type described above, and possibly also involve determining preferences and events and generating associated outreach therefrom, in the manner previously described in conjunction with FIGS. 1-8.
Privacy, and the potential lack thereof, can be either the cornerstone or point of failure in Internet-based systems. Strong privacy protection will protect users and their data, but weak privacy protection can destroy the confidence of users in the system and cause massive damages to the health of Internet-based technologies and the parties that depend on these. Privacy failures can be categorized as unintentional (such as privacy abuse resulting from a data breach) or intentional. In an intentional privacy failure, the organization sharing data did so intentionally, although not necessarily with a clear understanding of the consequences of sharing. An example of this latter type is the data sharing from Facebook, of their user data, to Cambridge Analytica; this privacy failure had a large impact both on society and on people's trust in technology, not to speak of the reputation of Facebook.
While protection against both unintentional and intentional privacy failures is important, the former is fairly well understood, but there is an almost absolute lack of understanding of intentional privacy failures. One critical aspect in the context of protecting against intentional privacy failures is to convey to end users, in a meaningful, and controllable, manner, how their data will be (or might be) used, and allow them to determine whether this is acceptable. Doing this will not only serve as a helpful guideline in terms of how data is used but also enable end users to make meaningful decisions. In a society enabling users to make such decisions, the preferences of users will stop abusive (intentional) sharing of data, or, should it still take place, enable the assignment of liability, which will deter abusive sharing. Moreover, the meaningful conveyance of privacy protection will bolster trust among users, who will be able to understand what data of theirs is used, by whom, and, to some extent, how. This building of trust will strengthen organizations that help protect user privacy according to the user preferences.
However, traditional methods for conveyance of how data is shared has been proven to be complex and confusing, and typical users do not understand what data requests mean to them, and therefore increasingly simply click through permission requests, rendering these meaningless. For example, if an app asks for read/write access to a user's mailbox, and read access to the user's list of contacts, few typical users realize that this translates to the app being able to send email to the user's contact on behalf of the user himself/herself.
This problem is increasing in severity as the number of data sources and data resources grow, making it almost impossible for non-specialists to understand the possible consequences of granting access to a data source or resource. For example, sensor networks such as IoT systems for home or office automation increasingly comprise large number of sensors, such as camera, microphone, pressure sensors, motion sensors, temperature sensors, inertial sensors, and more. Most users may not even know that an inertial sensor detects sudden movements, making it very useful to identify earthquakes, and therefore, granting access to data from such a sensor requires a leap of faith. In contrast, people understand the use of a camera very well, and would not find it acceptable to share data from a camera in their bathroom (should there be a camera there in the first place.) However, if the camera is only used to detect the presence or absence of a person, and only that one bit of information is shared, then many people may relent and accept sharing of this data, provided the use is sufficiently beneficial for them. For example, combining an inertial sensor and the presence detector from a camera could help detect a fall in the shower and automatically call for help, if necessary. This may be a desirable functionality for many users, although few would know what an inertial sensor is, nor feel comfortable sharing camera data. If a user understood that an inertial sensor and the output from a camera could be used to detect falls in the shower, he or she may feel comfortable granting access to such data sources; however, this user would feel betrayed if it turns out that the camera was used to send photos of them in the shower to an external site, where these photos potentially could be seen and shared. This example demonstrates the complexity of granting access to data sources, and also, the risks and pitfalls this could entail.
The ability of typical users to understand and appreciate the use of their data is further complicated by the use of complex artificial intelligence methods. For example, consider a hypothetical machine learning technique that uses a large number of sensors, such as a camera, microphone, pressure sensors, motion sensors, temperature sensors, and inertial sensors, to determine whether a user has fallen in the shower. The use of the temperature sensor may be inexplicable to a typical user, and therefore, the inclusion of this data source in an access permission request will seem unreasonable. It may even be that experts do not know why the temperature sensor is useful to determine whether a user has fallen in the shower. In typical machine learning systems, a large amount of data types are provided during a training session, and, as a result of the training session, it is determined what data sources are useful. However, while typical machine learning methods will identify what sources are useful, they will not identify the reasons for this, making even experts wonder why a particular data source is needed. Therefore, as an increasing number of methods use artificial intelligence methods such as machine learning, the permission requests will become increasingly inexplicable. As this happens, users increasingly have to rely on faith that their data will not be abused, or avoid using systems that may be perfectly benevolent in spite of making inexplicable requests for data. This trend will increase end-user risk-taking, lower the extent to which service providers can be held responsible and lay the foundations for abusive behavior among data consumers.
The problem of data access is not limited to conveying permissions in a meaningful way, but also includes malicious behavior of the data recipient. The malicious behavior could be in the form of a data consuming application or process that is advertised to perform one task but which performs another, and less desirable task. Potentially, a data consuming process may have one stated use, but in addition, perform additional processing that is not desirable but which is not made public. Furthermore, there is a risk that an application or other data consuming process is advertised as performing one task, which it does, but then later to change its behavior and perform a second task. This second task would be different from the first task, and may use the same data but in another way, and in a manner that produces a result that is not desirable to the end user whose data it uses. This change of behavior may come about as the organization providing the data process or application is sold, control changes, circ*mstances change, or that the organization is infiltrated by a hostile party, e.g., by malware or a rogue employee, causing an undesirable change. Thus, even if users can trust that their data is used properly, according to how it is conveyed to be used, this can change. This is another source of concern to users, and is a risk in the context of privacy. Even if there is no malicious behavior, users may be concerned that there is, or could be, and this concern may cause a reduction of willingness among users to engage with service providers as well as fears that they may be abused.
Thus, it is a long-felt need to address the manner in which data access is managed. This need will mushroom with the development of sensor-rich networks, aggregation of data sources, and the use of big data methods for processing of information, including the use of artificial intelligence, but is not limited to these contexts. The development of methods to address these problems are vital for the ongoing success of the Internet and associated services.
Illustrative embodiments provide techniques for privacy enhancement using derived data disclosure. Some of the illustrative embodiments are advantageously configured to address and solve one or more of the above-noted problems of conventional approaches. The illustrative embodiments include but are not limited to systems, methods, apparatus, and computer program products.
In one embodiment, an apparatus comprises at least one processing device comprising a processor coupled to a memory. The processing device is configured to determine a representation characterizing data from one or more sensor devices of at least one sensor network, to determine a privacy impact indicator for the data, to provide the representation and its associated privacy impact indicator for presentation in a user interface of a user device, and to control access to information relating to at least portions of the data by one or more third party service providers based at least in part on user permission feedback relating to the representation and its associated privacy impact indicator as presented in the user interface of the user device. The “information relating to at least portions of the data” in some embodiments illustratively comprises predicates computed from at least portions of the data, results of functions applied to at least portions of the data, and/or at least portions of the data itself. Such related information as the term is used herein is therefore intended to be broadly construed.
The processing device in some embodiments implements at least one node of a plurality of interconnected nodes coupled to at least one of the one or more sensor networks. Additionally or alternatively, the processing device can implement at least one pre-processor coupled between at least one of the one or more sensor networks and one or more servers of the one or more third party service providers. Numerous other arrangements of one or more processing devices associated with one or more sensor networks, possibly involving nodes of one or more node networks, are possible in other embodiments.
In some embodiments, determining a representation characterizing data from one or more sensor devices of at least one sensor network comprises receiving data generated by one or more of the sensor devices, computing one or more predicates based at least in part on the received data, and determining the representation based at least in part on the one or more computed predicates. For example, computing one or more predicates illustratively comprises computing at least one first-order predicate comprising information indicative of at least one of presence, layout and activity within an area that includes at least a subset of the sensor devices of said at least one sensor network. Computing one or more predicates in some embodiments further comprises computing at least one second-order predicate based at least in part on at least one of the one or more first-order predicates.
The representation can comprise, for example, information specifying a particular visual icon for presentation in the user interface of the user device wherein the visual icon conveys information regarding a type of data from the one or more sensors without identifying the data itself. The representation can further comprise, also by way of example, a description of a service provided by a given one of the third party service providers, where the given third party service provider requires access to information relating to one or more particular portions of the data in order to provide the service.
Some embodiments are configured to produce data of conceptually meaningful types, such as “a person is in the room,” “a known person is detected,” “a person is sleeping,” “a dog is present but not barking,” etc. This is in contrast to “all data that can be used to determine the identity of a person,” but which can nonetheless be used to detect many relevant conditions in a variety of different contexts.
For example, in some embodiments, at least one permission is represented by an icon. Thus, “a person is sleeping” may be represented by an outline of a person with “ZZZ” rising from his/her mouth, while “a dog is present but not barking” could be represented by a dog icon where the dog's mouth is closed, and the tail is wagging.
One may not be able to represent all meaningful data as icons only. For example, a conditional permission “share all sound data for up to 30 seconds after a wake command has been given” may be difficult to describe in an image. Accordingly, some embodiments are configured such that at least one predicate can be visually represented or otherwise displayed in a user interface.
Not all predicates need to be displayed. For example, one may display: (a) only those predicates with privacy impacts above some specified threshold (that may be set by the user); (b) only those predicates that the user has not previously marked as “always approve;” and/or (c) one or more selected predicates at first, thereby allowing a user to indicate that he or she wants to see more predicates (that are potentially less intrusive).
References above and elsewhere herein to “determining a representation” are intended to be broadly construed, and can include, for example, generating the representation or particular components thereof, accessing existing stored components, and/or various combinations of these and other techniques. Accordingly, in some embodiments, predetermined representations may be stored for use with certain types of sensor data, and a particular representation may then be retrieved or otherwise accessed when used to characterize particular sensor data. These and other arrangements are considered examples of “determining a representation” as that phrase is broadly used herein.
In some embodiments, providing the representation and its associated privacy impact indicator for presentation in a user interface of a user device comprises transmitting to the user device information sufficient to allow the user device to obtain in a browser of the user interface at least one window that includes the representation and its associated privacy impact indicator. For example, providing the representation and privacy impact indicator in some embodiments involves sending a link to the user device, which allows the user device to retrieve a web page that includes the representation and privacy impact indicator. Alternatively, at least one of the representation and the privacy impact indicator may be sent directly to the user device. Numerous other arrangements are possible.
The privacy impact indicator may be configured to identify a particular level of privacy associated with the data, with the particular level being selected from a scale ranging from a relatively low level of privacy to a relatively high level of privacy.
In some embodiments, determining the representation, determining a privacy impact indicator for the data, and providing the representation and its associated privacy impact indicator further comprises determining a first representation characterizing data of a first type from one or more sensor devices of said at least one sensor network, determining a second representation characterizing data of a second type from one or more sensor devices of said at least one sensor network, determining a first privacy impact indicator for the data of the first type, determining a second privacy impact indicator different than the first privacy impact indicator for the data of the second type, and providing the first and second representations and their respective associated first and second privacy impact indicators for juxtaposed presentation in the user interface of the user device.
In some embodiments, controlling access to information relating to at least portions of the data by one or more third party service providers based at least in part on user permission feedback comprises receiving data generated by one or more of the sensor devices, computing one or more predicates based at least in part on the received data, and providing a given one of the third party service providers with access to at least one of the one or more predicates in a manner consistent with the user permission feedback.
Additionally or alternatively, controlling access to information relating to at least portions of the data by one or more third party service providers based at least in part on user permission feedback comprises receiving data generated by one or more of the sensor devices, applying at least one function to the received data to generate derived data, and disclosing at least portions of the derived data to a given one of the third party service providers in a manner consistent with the user permission feedback.
Examples of the at least one function applied to the received data to generate the derived data include an obfuscation function configured to obfuscate at least a portion of the received data from at least one of the sensor devices, and/or a combination function configured to combine at least a portion of the received data from one of the sensor devices with one or more other portions of the received data from one or more other ones of the sensor devices.
These and other aspects of illustrative embodiments will now be described in more detail with reference to FIGS. 9 through 17.
One aspect of the disclosed technology is a data pre-processing method that takes a collection of inputs, such as data from sensors, and generates what we will refer to as predicates, where these predicates relate to an action that can be conveyed to an end user.
For example, the input data to such a data pre-processor may be data from one or more inertial sensors, one or more humidity sensors, one or more cameras, and one or more microphones. The data pre-processor identifies from an inertial sensor associated with a bathroom that there is a movement, which may be a fall, but which may also be an earthquake or a child that is jumping. It also determines from another inertial sensor that is illustratively located in another room, but which may also be located in another building, whether there is a similar sensor output from that second inertial sensor as there was from the first inertial sensor. If so, that may be indicative of an earthquake as opposed to a fall in the shower. If not, it may correspond to a fall in the shower, but may also correspond to a child jumping in the bathroom. To distinguish these two from each other with a high probability, the system determines, using the humidity sensor in the bathroom whether the humidity has recently increased above a threshold amount, which is indicative of the use of the shower; or whether the humidity in the bathroom is significantly higher than the humidity in another room, which may be in another building that is not associated with the same user. This would also be indicative of the shower being used. This makes it more likely that there is a fall in the shower; however, it is not certain. For example, it is possible that there is a child in the bathroom that jumps while somebody draws a bath or runs a faucet or shower. The system now determines, based on microphone data if there was any vocalization, such as a yell or a scream, taking place largely at the same time as the detection of the movement using the inertial sensor. This is indicative of a person having fallen, although it is possible for a person, even under these circ*mstances to yell without falling or fall without yelling. Therefore, the system also uses the camera signal to identify a sudden movement correlated with the signal indicative of a potential fall, from the inertial sensor. Instead of or in addition to a camera, a sonar system and/or an electromagnetic indoor positioning system such as a WiFi time-of-flight mapping system can be used to make this determination. The system then combines determinations relating to the one or more inertial sensors, the one or more humidity sensors, the camera, and the mapping system, and makes a determination whether a person is likely to have fallen in the bathroom, whether in the shower or elsewhere. This result is referred to as a predicate.
A predicate can be provided by the system to one or more applications or data consuming processors. Consider a first app that automatically calls for help when a danger is detected. One of the indicators of danger is the indication that a person may have fallen in the bathroom. Another indicator of danger may be that of a home invasion, i.e., a person climbing in through a window; that may correspond to one or more other predicates. Such a predicate may also use the camera and sonar (to detect presence and movement); humidity (to detect an open window); but maybe not the inertial sensor.
Another app may also help detect dangers. These dangers may not be the same as the dangers detected by the first app. It may also be the same dangers being detected, but in a different way. For example, the second app may detect a home intrusion attempt by detecting a previously not seen MAC address (i.e., a phone or other mobile device) being present, along with an indicator that there is a window open in the room where the MAC address was detected, and an indication that the MAC address was not detected in other rooms between an entrance room and the room where its presence was detected.
The two example apps would both consume data that came from a home. However, they would use different signals to perform the processing. Neither of the apps would have access to raw sensor data underlying the predicate information they received. For example, neither app would have direct access to the inertial sensor, the sonar, the humidity sensor, the camera or the microphone. Therefore, the possibilities for abuse are strictly limited, which is a significant benefit. Furthermore, the granting of permissions is considerably simplified. Instead of asking a user for permissions to the inertial sensor, the sonar, the humidity sensor, the camera and the microphone, which correspond to the traditional approach, the system could ask the user permission for “transmitting data indicating a person falling in the bathroom, to application X”, which is easier to understand for end users than the traditional approach.
In the example above, the two competing apps performing detection of risks have different ways of detecting intrusion. The first app may request access to three predicates, namely a first predicate that corresponds to that a room does not have an entrance from the outside; a second predicate that corresponds to that the room has an open window; and a third predicate that corresponds to that there is a presence of a person in the room, but there was no person in adjacent rooms prior to the detection of the presence of the person in the room. The first predicate is static in that it will not change once it is computed (except in rare cases where the owners add another entrance, or never uses an entrance for a long time, thereby causing the system to determine that the room does not have an entrance.) This first predicate may be computed by identifying user movement for a long period of time, determining what rooms are adjacent to each other by correlating movement, and by determining what rooms users sometimes show up in or depart from without coming from or going to other rooms. The second predicate may use air pressure, sound, and humidity sensor signals to detect when a window (or door) is open. The third predicate may be implemented using sonar and by detecting sources of sound, such as footsteps. To get permission to access these predicates, the app needs the user to review and grant access permission before the system gives the app access to the data sources; these permission requests may be “Will you grant access to your floor plan to app X?”, “Is app X allowed to receive data about what rooms have open doors and/or windows?” and “Will you let app X know if anybody enters a room through a window?”. If the user permits app access to all this data, the app is able to perform the task, and is enabled. If the user does not agree to the access requests, then none of the requests are granted and the app does not gain access to the corresponding data. In some instances, a user may grant access to some types of predicates but not others, thereby allowing the corresponding app access to only the permitted predicates. This is meaningful when the granted predicates corresponds to a collection that is associated with a task.
Continuing with the example above, the second app performs a similar task using a different set of predicates, and correspondingly, results in a different set of permission requests. Some users may prefer the permission requests of one service provider to the permission requests of another, competing, service provider. Such a user will then select to receive service from the preferred service provider. While the end-user observed functionality of their services may be identical, and their error rates associated with the provision of these services may be the same, the permissions are different. If this causes a large number of users to prefer the service of one provider over the apparently functionally identical services of another provider, the first provider will be favored. Thus, the ability for users to understand the permissions not only leads to more content users who now understand what type of access to their data is possible, but moreover, it stimulates the emergence of competition based on beneficial privacy policies, as expressed using the permission requests.
The above examples show how the use of combination is useful for a system to generate predicates from sensor data and other privacy sensitive data. Another approach disclosed herein is the use of obfuscation of one or more signals, to generate a predicate from these one or more signals. An example of obfuscation follows.
An example app or data consumer wishes to identify a speaker, but without having to intrude on the privacy of the person. The system receives sensor data from one or more microphones and creates a predicate as an output. An example output comprises an FFT of a segment of voice data, such as a five-second segment, providing this as an output predicate. This is useful in the context of many applications; for example, voice recognition can be applied to the data without revealing what the person said; as another example, background sounds, such as music and sports, can be detected and identified. For an app to be provided access to this predicate, a user configuring the system may be given a permissions request such as “Company Y wished to access scrambled microphone data. It is not possible to detect what a person is saying based on scrambled data. Do you grant access?”
The access to FFT of microphone data is much less intrusive than access to the microphone data itself. Another example app wishes to receive commands from a user, similar to how Amazon's Echo receives commands. The commands typically have a “wake-up” word (such as “Alexa”) followed by a request, such as “Check the weather forecast for tomorrow.” After a pause, or after a set time period, it is assumed that the request portion of the command is completed. Therefore, an app like this does not need full access to microphone data, but only a selection of data (limited in time or by a pause), starting after the wake-up word has been detected. For the app to be provided with such data, the system can provide a user with the permissions request “Service Z1 wants to access your microphone for up to 20 seconds after a wake-up signal is received. Z1 will not be given access to the microphone at any other time.” A competing organization decides that it only needs a transcription of the detected voice data after a wake-up signal has been detected, as opposed to all the sound data. This is clearly more privacy preserving. The system may avoid making the distinction and use the same permissions request as for service Z1, or it may use a more specific request, such as “Service Z2 wants to receive spoken commands, given within 20 seconds of a wake-up signal. Z2 will not be given access to microphone data at any other time, and will never obtain access to background sounds.”
In one embodiment, each permissions request is associated with a score or a rating, where this may be in the form of a value, a color, or a diagram, and where the score or rating corresponds to the degree of privacy intrusion according to a rating system that can be provided by the system, an independent organization, or the user himself or herself. For example, the request corresponding to company Y above may have a rating of 7 out of 10, where greater numbers correspond to greater privacy; and where Z1 may have a rating of 4 and Z2 a rating of 5.3. The score or rating is illustratively graphically associated with the permissions request.
In one embodiment, the system displays the permissions requests in a structured manner based on the type of request. For example, the system may first present all requests that relate to collecting and sharing data that may identify a user (e.g. MAC address, voice data, etc.); then present all requests that relate to collecting and sharing data that will be used to generate a targeted advertisem*nt; followed by all requests that relate to collecting and sharing data used only for determinations that are pseudonymous and in which individual user data is aggregated, such as data sent to an app that determines the air pressure in a neighborhood, and uses this for weather forecasting. To the extent that some permissions relate to several such categories, the permission request may be listed, for a requesting app or data consumer, in all of these categories or only in the categories that are considered most privacy intrusive, whether by users in general or the user to whom the request is presented. An alternative display structure involves different types of sensor output, e.g., one category that relates to audible sound data; and another that relates to motion data; etc.
In one embodiment, apps and other data consuming processes are associated with one or more scores or ratings, and in an interface where the user is provided with information relating to multiple related apps or data consuming processes, the scores or ratings are conveyed to the user. For example, this can be done by displaying an icon representing each compared data recipient, such as the apps/data consumers, one or more scores/ratings, and a brief description of the functionality of the data recipient. In addition, traditional consumer-generated ratings relating to the different data recipients can be presented, where these can take the form of one or more scores, a count indicating the number of underlying and individual user reviews, and one or more reviews or scores provided by consumer organizations such as the Better Business Bureau. In one embodiment, the user in the process of deciding between offerings is able to cause the options to be sorted based on one or more indicators, such as sorting based on the average privacy impact associated with the one or more permission requests of the data recipient; based on the highest or lowest score/rating of the permission requests; based on ratings provided by consumer organizations; or based on an aggregate measure of two or more of these types of scores.
Since some apps and data consumers may request a large and complex set of predicates, it is beneficial for users to be able to state pre-set preferences, such as “I always accept permissions requests of this type,” “I always accept permission requests associated with a privacy impact lower than 6,” “Always accept permission to share data categorized as anonymous and containing no personally identifiable information” or “Only display the highest-scoring permission requests of each type.” The latter selection corresponds to only showing, for one data recipient, one permission request relating to collection of microphone data, namely the request that is the most intrusive according to the associated score/rating. The system may also display information relating to the use of the requested data, such as whether it will only be used in aggregate; whether the purpose of the use of the data is classified as a benefit to society (e.g., to identify trends in flu, determine what neighborhoods need financial support to improve home insulation, etc.) These and variant display techniques simplify the user experience and enable users to understand privacy impact, making them educated consumers.
In the examples above, the predicates are distinct from the final result of the determination by the app or data consuming process. In other words, an example app in the above would combine two or more signals to obtain a computational objective. The disclosure also supports uses where there is only one data source, as a person skilled in the art will recognize. This data source, in one embodiment, is a low-level predicate such as the raw signal of a sensor; in another embodiment, it is a mid-level predicate, such as a predicate determining the location of a user or the layout of a space. In yet another embodiment, it is a high-level predicate, such as a detector of a person having fallen. This may be computed, by the system, as a function of one or more other predicates, or of one or more raw signals. All of these apps may also use stored data, as will be appreciated by a person skilled in the art, where one type of stored data is historical records of predicates and/or sensor data, and another is user input, whether using voice, motion, keyboard, touch screen or other alternative input methods. The permissions will be modelled accordingly. In the latter case involving the fall detecting high-level predicate, the permission request may be “Do you agree to allow company X to receive data that identifies whether a person is likely to have fallen?” The more the permission request correspond to concrete phenomena or detections that users can relate to, the more meaningful.
The pre-processor stage in some embodiments may be implemented at least in part by hardware associated with the deployed sensor network; at least in part by hardware corresponding to backend servers associated with the organization manufacturing, selling or maintaining the sensor network system; and/or at least in part by a cloud-hosted processing environment. Additional or alternative pre-processing locations are possible, such as on a smartphone, on a laptop, on a tablet, on a router, on a firewall, by an ISP or privacy watchdog organization on premises used by them, and more.
A first aspect of the disclosed technology is a collection of nodes comprising a node network, where these nodes comprise sensors and means for communication, and where a node network is connected to a central utility that receives and processes data from the sensors associated with the network and/or exhibits control over the functionality of the nodes of the network. In an alternative embodiment, the collection of nodes can be replaced by at least one data source, such as a user's Facebook timeline; a LinkedIn user's set of changes to his or her profile, including messaging; an email user's mailbox; a collection of data resources associated with a phone, a tablet, a smart watch, a laptop, a desktop or other computational device, and the data sources and repositories associated with it. Other forms of data sources such as data sources associated with other parties are also useful. We will refer to the node network or other data source or repository collectively, and will interchangeably refer to these as the node network, the data source, or the data resource.
A second aspect is the central utility, which is illustratively, at least in part, in a location separate from the node network, and illustratively is connected to multiple node networks that are not part of the same installation or physical space. The central utility can be housed on one or more servers associated with a service provider, or in a cloud processing environment controlled by the service provider. For clarity and simplicity, we will refer to both the service provider as the trusted service provider, to distinguish it from third party service provider, which may not be trusted. We will also, interchangeably, refer to the central utility as the trusted service provider to reflect that fact that it is this party that provides instructions to the cloud service provider or server farm to implement the functionality performed by the central utility. Some of the functionality associated with this entity can be performed by a device associated with the data source, such as a node on the network, a router used to transmit data, a laptop, or a third party computational process that performs a task on behalf of the user; we will refer to the latter as the consumer ombudsman, in spite of the fact that the user may not be a consumer strictly speaking, but may be an employee, an organization, etc.
A third aspect of the disclosed technology is at least one third party service provider that wishes to provide a service to the users of the node network, with the assistance of the trusted service provider, or wishes to obtain data from the users of the node network. Two examples are a residential alarm company and an advertiser.
In some embodiments, example functionality of the disclosed technology includes a central authority or other trusted service provider being configured to receive data from one or more node networks and, from this data, to generate one or more predicates. Example data comprises output from sensors such as radio transmitters, microphones, motion sensors (including sonar, which uses a microphone, as well as other types of sonal sensors), thermometers, and sensors detecting volatile substances such as alcohol fumes, carbon monoxide, etc.
Example predicates that can be generated in illustrative embodiments herein comprise a description of a user, such as the user's likely gender; the preferences of a user, such as a preference to watch soccer games on TV; and/or the activities of the user, such as hosting parties, cooking or learning to play the guitar; as well as numerous others.
Additional predicates comprise whether any user is present at a location at a particular time. The predicates are less sensitive than the full data from which they are derived. Whereas some of the predicates, such as the location of a user at any one point in time, may be considered invasive, such predicates are justified in some contexts, such as in the context of a fire or other catastrophe, and may be acceptable to share on a limited basis, such as to organizations associated with emergency response.
Yet other predicates are not invasive to typical users, such as whether a user associated with the space is learning to play the guitar; however, it should be understood that some combinations of innocuous predicates may pose a concern to some users, if known to third parties, whereas such users would not be concerned with these third parties knowing any one of the predicates. An example of this general problem is described in L. Sweeney, “Simple Demographics Often Identify People Uniquely,” Carnegie Mellon University, Data Privacy Working Paper 3, Pittsburgh Pa., 2000. Analogously, one example of where datasets that appear to be anonymized could be de-anonymized was described in A. Narayanan and V. Shmatikov, “Robust De-anonymization of Large Sparse Datasets,” IEEE Symposium on Security and Privacy (SP 2008), May 18-22, 2008.
Accordingly, it is desirable in some embodiments to address situations in which colluding third parties may attempt to combine their inputs to gain insights beyond what is acceptable. These and other issues are addressed in illustrative embodiments through the use of pseudonyms, as will be described in more detail below.
A valuable aspect of providing predicate information to third parties is the creation of profiles associated with the predicate information, where these profiles are associated with identifiers that illustratively are pseudonyms unique both to the sensor networks or associated users, and to the third party service providers. This means that two third party service providers receiving predicate information related to two particular node networks, data sources or users would illustratively not be able to determine, if collaborating, that the two node networks, data sources or users are the same or related to each other.
In some contexts, however, it may be beneficial to use pseudonyms that are not unique to both the node network or user, and the third party service provider receiving the predicate data. For example, in some contexts it may be acceptable for two third party service providers to be able to determine whether they are observing and controlling the same node network. Furthermore, it may sometimes be beneficial for the pseudonyms not to be used as a method to anonymize the user. For example, a user of a network may wish to associate a service he or she is receiving with an email address he or she owns, and to enable the third party service provider to contact him or her directly, without the collaboration of the trusted party. In this context, the email address of the user may be used in place of the pseudonym.
At the other end of the spectrum on anonymity, some pseudonyms are unique not only to a node network or end user, or to the third party service provider obtaining information associated with related predicates, but also unique to a given session. Here, a session may be a period of time, such as three days, or a unique event, such as one transmission of predicate information sent from the trusted service provider to the third party service provider.
Another aspect of the disclosed technology is what we refer to herein as “capabilities.” The different types of pseudonyms described above illustratively correspond to different capabilities of the third party service providers obtaining predicates related to the pseudonyms. Another capability is the type of predicates that a given third party service provider gets access rights to, whether these access rights correspond to the right to receive information about the predicates or to initiate state changes associated with the node network. One example state change access right is the right to increase the temperature in a building, another is the right to set the volume for a speaker, and a third is the right to generate messages to the user. Typical users would want to understand what capabilities third party service providers have, illustratively before signing up for service with them.
Thus, a third party service can be described using their capabilities, or classes of capabilities, where one class corresponds to a common set of capabilities, and another class to another common set, and where each service provider can be described as requiring no more capabilities than the set of a given class. This simplifies things for users, who can decide, for example, that they are willing to provide “class C capabilities” to service providers they do not know, but “class E capabilities” to service providers that they trust. Different users may have different preferences, of course, and this approach simplifies the mental burden of assessing service providers and their access to data and control. Thus, permissions, described elsewhere in this disclosure, can be modeled both around access to predicates and around the granting of capabilities.
One aspect of the disclosed technology relates to reducing the sensitivity of data and associated predicates. For example, a user may be willing to let one third party service provider know whether he or she was in his or her home on a particular day, but not until a week afterwards. Similarly, a user may be willing to let a third party service provider obtain voice commands spoken by a user associated with a node network, but not utterances that are not voice commands or associated with such commands. As yet another example, a user may be willing to let a third party service provider know a frequency representation of the sounds in a home, from which the third party service provider could tell whether the home has a barking dog or not, while the user is not willing to let the third party service provider have access to a time-series representation of the sounds in the home, as this would amount to the third party service provider being able to determine what was being said.
These processing aspects of the data and the predicates also apply to information that is stored (whether by the trusted party or third party service provider), and to how sensor data is transmitted to the trusted service provider from the node network. We refer to these and related processing aspects as “modifiers.”
Another aspect of the disclosed technology is a method to use align the requests for permissions (of access to data and capabilities to control) with the predicates and capabilities, both of which may be low-level (such as the sensor data or other data from the data source), medium-level, or high-level (where the latter is a more conceptual level of the data and the control of the setting.) Thus, the use of predicates and capabilities as described herein prevent insights and control that are not wanted by the end user, and limit such insights and control to the desired aspects; it makes the conveyance of permission requests meaningful as it align these with concepts that are not abstract and for which users can understand the privacy impact, and furthermore, it limits the access, both to information and controls, to these permitted quantities by means of the system granting access to user information by means of an API or related technique that is aligned with the permissions.
In one embodiment, predicates are computed from sensor data by the trusted service provider. In an alternative embodiment, the predicates are computed from sensor data by one or more nodes that are part of the node network, or otherwise located on the premises associated with the node network. In other embodiments, at least one of the nodes that is part of a node network or located on the premises partially processes sensor data and transmits the partially processed sensor data to the trusted service provider, which then computes one or more predicates from the partially processed sensor data.
In these and other embodiments, the processing associated with computing predicate data from sensor data illustratively comprises, for example, at least a subset of selecting sensor data; aggregating sensor data; performing an FFT on sensor data; delaying the reporting of at least some sensor data or a function thereof; removing duplicates or strongly correlated signals from the sensor data; identifying a user and reporting sensor data associated with the user, but reporting a pseudonym instead of identifying data of the user; and determining, based on a user-controlled policy, what type of sensor data and/or partially processed sensor data to report to the trusted service provider. The determining can be based, for example, on the detection of a condition, such as whether an alarm system has been armed; whether a user is a recognized user or is associated with a recognized MAC address or other hardware identifier; the time of the day; the value of a computed predicate; or a combination of these or other conditions.
In some embodiments, the trusted service provider acts as an anonymizer between a node network and its associated users, and a third party service provider. For example, the third party service provider may be provided with certain types of node network usage data, such as how users move around, what type of products they use and what TV programs they watch, but is not provided with information useful to identify or locate the user or the node network, beyond some granular information such as a general area (e.g., a city, a county, a state or a country) or an age group of the user. In particular, the third party service provider would not be allowed to know the exact location of the node network or its users, as well as identifiers associated with the node network or its users, such as email addresses, HTML cookies, MAC addresses, etc.
This permits more detailed usage data to be shared by the trusted party to the third party service provider, since the actions of the users are protected by the anonymization or pseudonymization performed by the trusted party. This also enables the sharing of select sensor data that may be sensitive, as opposed to only select predicate data. The same type of anonymization can also be performed in a context where the third party service provider receives only predicate data and not sensor data, or a combination of such data types. Furthermore, the trusted party may aggregate data between different node networks before providing sensor data and/or predicate data associated with the aggregated sensor data. This is useful for carrying out general lifestyle surveys, characterizing various type of data relating to life in a home, in a manner analogous to Nielsen ratings for TV programs.
A user may grant permission to these by category, or may grant access to all anonymized data extraction that does not have capabilities, or grant access to all anonymized data extraction even when associated with capabilities. The granting of such permissions can be part of the installation process, but it is also possible that some types of permissions are part of an end user licensing agreement (EULA), and not possible to accept or deny explicitly. For example, it is necessary in many contexts for the system to extract at least some data to provide services, independently of whether third party applications are selected or not. A user, for example, wishing to remotely control his own system using his or her phone must implicitly allow the system to control the system. Similarly, the system may include functionality such as anonymized data use, whether aggregated or not, that users may not select not to permit.
Additional aspects of example systems will now be described. Although reference is made below and elsewhere herein to a particular system, it is to be appreciated that the various aspects to be described may be implemented in respective different systems. In addition, different combinations of the various aspects may be implemented in respective different systems. Numerous other arrangements are possible, so references herein to a given “system” should not be viewed as requiring particular features or functionality.
An example system in illustrative embodiments, also referred to herein as a “disclosed system,” is particularly designed to avoid abuse. One type of abuse of relevance in the context of sensor networks is intimate partner violence (IPV). With the development of digital technologies, such technologies have come to play a key role in IPV contexts, as abusers exploit them to exert control over their victims. IPV affects roughly one-third of all women and one-sixth of all men in the United States (see S. G. Smith et al., “The national intimate partner and sexual violence survey (NISVS): 2010-2012 state report,” 2017.) Thus, it is important to design systems in a responsible manner, to ensure that they are not being harnessed for abuse.
Third party application systems are a potential risk in this context, especially in systems where an abuser has the capabilities to install third party services without the knowledge of the victim. This is exacerbated by asymmetric technology awareness within households, and by complex user interfaces, lackluster audit capabilities, and the difficulty of automatically scanning the code or determining the functionality of services that have a substantial backend component. The fact that many applications may have dual-use purposes also complicates matters, as it means that most uses may be legitimate and desirable, but that the services also have undesirable uses. It is unlikely that much progress can be made by simply reviewing the access rights of applications, as two applications could potentially access the same type of data, while only one of the applications might have dual use or an undesirable and abusive capability.
The disclosed system in one or more embodiments addresses these and other problems, using multiple approaches. For example, one approach includes maintaining an installation and configuration log for the system, wherein all changes, the parties making the changes, and the time of the changes are documented. This enables a user to review recent installations or subscriptions. Similarly, the disclosed system incorporates a tracking method that identifies common constellations of applications and identifies those that do not match some notion of commonality, which could either be set by a user or kept at some threshold value. Third party applications that are identified as unusual could therefore be given extra scrutiny. For example, users could be alerted about the presence of such third party applications, or the user can request a roll-back of the installations/subscriptions or associated configurations. Simply reconfiguring unusual applications will address a large part of the problem, as this prevents exfiltration of data in the most common contexts.
In addition, the system illustratively correlates abuse complaints and other complaints with the constellations of installations and subscriptions, thereby singling out applications whose installation or subscription is commonly associated with abuse. Such applications may be banned from the marketplace of apps if the correlation is very strong, as this suggests an intent of abusive use, or users can be notified of the potential risks for applications with lower correlation, which suggests dual use.
The system illustratively maintains multiple thresholds for the determination of risk for different scenarios. One example scenario is where a user is excluded from a node network, there is a conflict report between registered users, when control is handed over from one user to another, or when one user is obtaining administrative (“admin”) rights for a system that is not a fresh installation without any applications. In such contexts, the new user(s) or admin(s) are notified of high-risk applications, where the alert may classify the application according to its correlation with abuse reports as well as whether the application is rare or new enough that there is not sufficient material for performing the correlation effort. Users can choose to remove or cause a reset or a reconfiguration of such applications.
An automated scoring system of this type helps maintain security and an abuse-free marketplace, as it allows the removal of undesirable applications by individual users, and the exclusion of applications that are commonly figuring in abusive scenarios. Users wishing to very quickly change the settings can choose to keep only applications that are sufficiently old and common that statistic information is available to the trusted party to evaluate their correlation with complaints from users, and which do not have a correlation to complaints above a threshold value. Herein, the type of complaint is illustratively taken into consideration in the determination. For example, a user complaint that the system does not select the right music or that the user does not wish to receive advertisem*nts is not relevant in a context of abuse, whereas complaints of unwanted access, attempts to evict users, initiation of access resolution mechanisms among permitted users, and others, are all signs of potential abuse.
In addition, the system identifies contexts that are associated with likely breakups, such as one out of two previously registered users disappearing from a node network for an extended time, apparent conflicts by two users, e.g., changing configurations back and forth, and other indications associated with a higher risk of abuse. These contexts and indications are then correlated with installed apps, especially those that are recently installed, and especially apps that are uncommon. Apps with a correlation exceeding a first threshold value, such as 0.4, are given additional scrutiny, included in notifications, etc.; while applications exceeding a second threshold value, such as 0.75 are blocked from the marketplace. By correlating applications to other complaints, other app problems can be automatically and statistically identified, as will be appreciated by a person skilled in the art.
As a more particular example, assume in the context of an illustrative embodiment that the third party service provider is a provider of radiant floors, and wishes to help provide better control to purchasers of their heating systems. Commonly, radiant heat systems comprise coils in the floor, where the coils either are electric coils or coils with water that can be heated by a water heater. However, due to the thermal mass of the floor above and around the coils, heat is retained in the floor and radiates into the room for a long time after it is transferred from the coils. As a result, if a thermostat detects that the temperature is too low in the morning, and starts heating the floors, by the time the sun has risen and started to heat the building and the air in it and the thermostat detects that the right temperature has been reached, the floors contain a large amount of heat. This heat will continue to radiate into the rooms well after the right temperature has been reached, thereby making the rooms too warm and potentially necessitating the resident to turn on the air conditioner to counter the temperature rise.
This cannot be addressed by setting the thermostat at a lower temperature as on a day when it is not as warm outside, the temperature will not rise at the same pace, and the house would end up being too cold. Additionally, different homes will have different heat swings than the one described here, due to the use of different materials. Furthermore, the habits of the residents may affect the response, e.g., one person may go in and out to the deck throughout the day (thereby affecting the indoor temperature) while another may not. Similarly, one person may use heat-generating appliances a lot on one day, but not at all on another; accordingly, the temperature response curve is not even the same from day to day for the same house, even if the weather is the same.
This problem can be dealt with by a central authority that comprises a machine learning system configured to detect the temperature in various locations inside and outside of the house, to obtain information about the thermostat settings, and to control the heating of the floor coils. Traditional automated control algorithms, additionally, can be used to identify the changes of temperature within the system and to regulate it using the control of the heating of the coils. Systems like this can be configured to determine the heat retaining capabilities of the floors, which can be done automatically based on the sensor inputs described above. Whereas this can be managed using a local feedback system, this is not optimal. It can also be performed by the trusted party; however, this party might be limited in terms of its business to providing systems and some form of control to users, and not specialized to control the temperature of the system using advanced methods. Different third parties may have different proprietary algorithms to address the problem, with various pros and cons.
Therefore, it is advantageous to be able to outsource the processing and service provision. However, if all customer sensor data were to be sent to the third party, that would be a severe privacy invasion. While the users have a trust relationship with the trusted party, they should not have to place trust in a larger than necessary number of organizations; therefore, it is beneficial that the disclosed technology removes the need for having to trust the third party.
In this embodiment, temperature sensor data (illustratively for all indoor nodes and also for at least one external node, if available) is transmitted to the third party, along with user preference data relating to the preferred temperature schedule. In one embodiment, no further data is transmitted, except for a temporally stable pseudonym that allows the third party to generate a model of the heat transmission and use this to predict what control to apply.
The sensor data does not need to be provided all the time, but could be transmitted every ten minutes, or in accordance with another specified transmission period. The third party transmits control information to the trusted party, tagged with data related to the temporally stable pseudonym, and this control data is transmitted by the trusted party to the node network. In one embodiment, the third party service provider may have sold or installed the physical heating system, and would know additional information, such as the type and thickness of the material above the coils, whether the coils are embedded in wooden panels or in concrete, and other related information. This type of information may allow the third party to fine-tune the model and speed up the conversion speed of the estimator. These techniques also help to identify a failed or failing system. It may also reveal real-life identities of the users, including their address. Since the temperature data could in some instances reveal whether the users are at home or not, e.g., as indicated by a repeated opening and closing of doors, this may be seen as a privacy drawback. A user who is concerned about this may prefer that only partial sensor data is provided, e.g., no temperature data from nodes close to entrance doors.
The trusted party can automatically identify what sensor data is indicative of potentially sensitive actions, e.g., by correlating temperature data with data indicating that doors are opened and closed, such as microphone data, and identifying sensor data that should be excluded. Such data would never be included in the feed to the third party. Thus, one advantage of the disclosed technology is an automatic determination of what sensor data is strongly correlated with potentially sensitive data, of which the system could have a list that is either provided by the user or which is from a template.
A third party provider of these services can either request permission to all the sensor data described above, or could, alternatively, request access to data indicating the current temperature and the predicted temperature in one hour, based on current and recent system events. If the system provides the machine learning functionality described above, the third party application provider can simply request access to predictive temperature data and permission to control the temperature by means of increasing or decreasing heating and cooling. The user can provide the desired temperature settings, both current and requested future temperatures, directly to the third party service provider, or can provide this information to the system, which then conveys it to the third party service provider. A third party service provider can also request access to the user's email box, allowing the third party provider to automatically identify from emails containing ticket purchase data, when a user will not be at home, thereby being able to create a suggestion or an automated policy regarding temperature changes for the affected time period. Similarly, a third party service provider can request access to a user's GPS coordinates, as determined either by a phone GPS, a vehicle GPS, the proximity to known networks with a registered device, and more; thereby enabling automated modifications to the temperature settings to address the expected needs of the user.
In the disclosed system, the trusted party collects sensor data and other user inputs and processes these to obtain derived data; then makes selected data and derived data available to third parties, and optionally receives control data from the third parties, where the control data or data derived from the control data is transmitted by the trusted party to the node network or at least one node associated with the node network, where the node network is at least in part the source of the collected sensor data. Optionally, the trusted party pseudonymizes the information transmitted to third parties.
The derived data illustratively comprises what we have referred to as predicates above. Apart from being a lesser impact on the privacy of users for derived or otherwise processed data to be shared, the structure disclosed herein also enables an alignment with constructs that align with user-understandable quantities and actions, as opposed to the much more abstract and vague descriptions that would correspond to raw data of sensors or other data sources. Therefore, the use of predicates help convey meaningful permission requests. Moreover, since the data access of third parties will be controlled by the system according to the permission requests granted by users, the privacy protection will be further enhanced.
The system can at any time reduce the access to future data from a user or installation to a third party vendor. The system can select to block communication of personal data (such as a MAC address of a phone) for selected users, whether these users are on systems that they control or are present at other installations that are also managed by the same system. At the same time, the system can grant limited access to circumvent such privacy features to law enforcement, whether this means tracking a select user or device within one or more installations, or responding to an inquiry whether or not a particular user was present in a particular location at a particular time. Thus, just like fine-grained controls can be applied to user data based on user preferences, exceptions can be made on a fine-grained basis, whether for purposes of law enforcement or other important inquiries.
Moreover, the system can permit anonymized movement data and other aggregated inquiries, whether by law enforcement, census bodies, and other organizations granted access to limited query functionality as that described above. This is implemented using the same kind of policies and controls as govern disclosure of user data to third parties.
Another example third party application is a preference derivation system that receives at least some data and derived data associated with a node network, from the trusted party, and determines at least one preference associated with the user(s) of the node network. Example preferences include “the user likes jazz,” “the male adult is often accompanied by a dog,” “people in this residence like to play ping pong,” and “user 1 of the residence sleeps poorly when the temperature is higher than 92° F.” The third party that derives such preference observations can do this without any knowledge of the location of the residence (or commercial space); without knowing the names or email addresses of the users; and without any other personally identifiable information. We refer to this third party as the preference determining third party.
Another third party service provider, such as an advertiser, a merchant selling dog toys, or service provider controlling the temperature of the residence (as described above), wishes to acquire data streams of relevance to its business, and pays the preference determining third party for relevant data, with the trusted party acting as a broker of information and payments. For example, a third party wishing to sell dog toys to users of a certain demographics, living in the US, may pay the trusted party $4 for each new referral. The trusted party knows the location of a residence, and has assessed likely demographics, whether by requesting information from users associated with the residence or by observing their behavior based on sensor data that it receives. The trusted party also knows what users it has sold data for relating to the third party wishing to sell dog toys. Therefore, periodically, the third party receives preference data relating to some pseudonyms, where the pseudonyms are illustratively specific to the user(s) as well as the preference determining third party. The third party then collects such preference data and determines what third parties are interested in the preference data, and determines whether the price the third party wishes to pay is sufficient.
Different preference determining third parties may have different pricing structures for preference data, based on the type of preference, the estimated accuracy, and other factors. If the price the third party wishes to pay is sufficient, the trusted party initiates a transaction in which preference data is sent to the third party, in this example, the third party wishing to sell dog toys. The trusted party then performs book-keeping both with respect to the buyer of the preference data and the seller of the preference data. The preference data that is sold is illustratively identified using a pseudonym that is unique to the user(s) to which it relates, as well as to the buyer of the data. Thus, the buyer of the preference data and the seller of the preference data cannot determine whether or not two pseudonyms relate to the same user(s). The third party wishing to sell dog toys may prepare an offer to be sent to the user, whether using the trusted party as an intermediary, or directly, in the case where the third party wishing to sell dog toys has been given a method of contacting the user(s).
Similarly, advertisers may build catalogues of likely preferences for different users, where the pseudonyms may be related to third party cookies or standard HTML cookies set by the third parties by a redirection method as the user(s) visit a webpage related to the trusted party. This allows later detection of a user, as he or she browses the web, where the detection is relative to a preference, such as “this user likes jazz.” There may be yet other pricing structures for data used in the context of advertising. In this example, such data is much more long lived than simply a referral of a user, and an associated opportunity to initiate the transmission of a dog toy discount coupon, by the trusted party for example.
The pricing structures may be set by the trusted party, or using market forces in which the trusted party determines the right price using experimentation, illustratively via an auction. Whereas most auctions involve goods or opportunities for which there can only be one auction winner, data such as preference data can be sold to multiple buyers, as long as the third party generating the preference data is willing to sell at that price, and the trusted party is satisfied with the margin it applies to the purchase.
In one version, this app consumes all the raw data that is required to generate the classifications; in another, it receives a data feed corresponding to the predicates associated with the classifications. The latter is less privacy intrusive. A third alternative is that the application receives all the raw data, but does not export this beyond a computational environment controlled by the system, and then exports the predicates. This alternative is less privacy intrusive than the first version of the functionality, since it does not export the raw data, but more intrusive than the second, since it is conceivable that the third party application encodes some fact beyond the predicates into the data it exports. Careful software analysis on behalf of a marketplace run by the system can help limit this conceivable risk. It is also possible that the derivation of the predicate data is performed by a preference-determining third party that is not controlled by the third party consuming the preferences; this further improves the privacy assurance associated with the third alternative version of the example, since it lowers the risk of leaks.
In certain versions of the above example, there are illustratively three different types of service providers: the trusted party; the preference determining third party; and the third party that consumes preferences. The trusted party acts as an intermediary, and also as a privacy representative for the end user. The preference determining third party buys, processes and sells data. In the example above, the prices may be set on a per-transaction basis, both for the purchase and sale of data; however, other monetization methods are also possible, such as the trusted party receiving a share of the profits, or the preference determining third party setting its own prices.
The preference determining third party is an example of a third party that buys and sells data. The data it buys may be sensor data from one or more node networks; predicates related to such sensor data, generated by the trusted party; and/or data purchased from other third parties that buy and sell data (whether directly, or using the trusted party as an intermediary). The other type of third party in this example is a service provider that provides a service to users, where the service may be to set the home temperature in the most appropriate manner; to provide discount coupons or other advertisem*nts selectively; or to perform other types of services.
This approach can be captured in a permissions request by clarification of the division of access between the two different third parties, where the data processing third party could be in a trust relationship with the data owner (i.e., user) that is different from the relationship between the predicate-consuming third party and the data owner. This is another aspect of illustrative privacy assurance methods herein; as it comes to the conveyance, division of abilities, and the ability to revoke access. The same approach can be used for capabilities, i.e., the ability to control the system of the end user.
The disclosed method supports permissions, whether of access to data (such as raw data and predicates) and of capabilities to control settings and actions. It does so on several levels. A user may have one agreement with the system. This may, for example, state that the system can access all the raw data from all the sensors and other connected data sources, and that it may sell aggregate and k-anonymous data, where k-anonymity corresponds to a privacy guarantee wherein the user cannot be distinguished from a number k of other users by scrutiny of the revealed/shared information.
It may alternatively state that some datum, such as image data showing a face, cannot be obtained by a central location of the system, but will be filtered out by the local installation, which is under the control of the system by having its software provided by the system. An agreement may state that the system may, further, use data to select coupons for relevant products, and convey these coupons to the user using one or more interfaces, such as postal mail coupons, in-app advertisem*nts and/or SMS based coupons.
Some users may be offered a discount in response to approving more disclosure of data, such as the right to present the user with advertisem*nts from third parties, based on data from the user that is analyzed by the system but not shared with the third parties. Another level of agreement related to data and capabilities relate to third party data processors, exemplified above. In one instance, these may be requesting permission to access the relevant data and to identify users interested in select activities, then identifying these activities to third party advertisers who can associate the activities (such as skiing, bowling or knitting) with the user and present advertisem*nts.
Another third party data processor may request to be allowed to convey the simple fact that the user in question is in an environment that is controlled by the system, as this may be associated with a desirable demographics. Yet another third party data processor does not use any user data other than to compute the predicates (as described above), for which the third party data processor charges any data consumers interested in the predicate data. Third party data consumers, in turn, may also have data access agreements and capability agreements with end users, as given examples of above. Although a large number of agreements may be seen as confusing at first, the structuring of these agreements around ratings, and the highlighting of concrete descriptions of selections of these, in addition to other aspects as described in this disclosure, helps the user manage his or her privacy.
Moreover, the user may install software that manages his or her privacy according to a set of pre-set principles that the user selects, where these principles may correspond to the types of data that can be shared and the circ*mstances under which it may be shared. For example, this can be done in some embodiments by placing the type of data in context of the associated privacy impact (according to the ratings, potentially provided by a consumer ombudsman) and the intended use of the data (e.g., for advertisem*nt, coupons, control, etc.), and whether the data is used for commercial purposes, a common good; the type of privacy (such as k-privacy for some set value of k such as 1000), etc.
A further example of a third party application is a billing service that measures the energy consumption associated with one or more network nodes and generates and attributes charges to one or more users. The trusted party determines the energy consumption, per outlet and/or node, for a registered node network, and reports this data to a third party. The energy consumption data may be delayed, e.g., by 48 hours, as a privacy protecting measure. This privacy protecting measure could be conveyed in the permissions request, e.g., “No data will be received in real-time. All data will be delayed by 48 hours.” A user may optionally select another time delay as well, in the permissions request. The user is provided with registration instructions, which illustratively involves at least one of a smartphone/tablet app and a webpage, where the user is asked to enter payment data. This payment data is collected directly by the third party in one embodiment, and collected and maintained by the trusted party in another embodiment. In the first embodiment, the third party receives pseudonym data associated with the node network; this can, for example, be conveyed in the path of a URL, or provided by the user at configuration time. Alternatively, the trusted party can maintain this data and perform billing of the user according to indications provided by the third party.
The third party associates a collection of network nodes with an account, where the account is associated with at least one of the pseudonym for the installation, a pseudonym of the registered user, payment data such as credit card data or a PayPal user name, and associated data related to user approval to charge. The third party obtains energy consumption data and determines the cost associated with the consumed energy. This may be the actual cost to the payer of the bill, but the cost may also be computed according to another formula, e.g., the cost of using the outlet is $10 for any 24 hour period, no matter how much it is used, or the cost to run the water heater, exceeding $5 per day, is to be billed, but not any amount lower than $5.
The charging policy may be provided by or selected by a service subscriber, such as the user, a party paying the bills to the energy company, or provided in other ways. The charges may also include overhead, e.g., the amount to be paid to the third party and the trusted party. In one embodiment, there is no overhead, but the user agrees to other commercial use of the data, e.g., the user may receive advertisem*nts or may have to qualify for an offer by providing demographic information or a discount coupon.
This aspect of the disclosure therefore enables the granular splitting of energy consumption costs, with applications to vacation rental properties, short term corporate rentals, dorm room energy billing, etc. A landlord may include a set amount of energy in the rent, e.g., up to one dollar a day, and the user/tenant would have to pay for any amount in excess. This protects landlords against abusive behavior, such as running the A/C with open doors and windows in the middle of the summer, or the mining of crypto currencies, unwanted charging of electrical vehicles or other highly energy-consuming activities that are not desirable.
In one embodiment, the trusted party acts as a payment network, taking payment from users and performing payments to other parties, including the third party, an owner/landlord, a party that subsidized the node network in exchange for a share of the profits, and/or the electricity company. Similarly, gas consumption is illustratively metered in an analogous way, or inferred from control data sent to gas consuming appliances. In addition, other utilities can be billed for using the disclosed system, including providers of telephone, Internet, TV, etc. The use of these resources can be determined by the node network in embodiments in which the node network is the intermediary used for control of the appliances. Other factors such as conveyance of signals can be used to detect usage.
The billing can be performed in a granular manner in which a user only pays for periods of time when he/she wishes to obtain the service, starting at the time of registration or other enablement, and ending at a preset time, when the user checks out or is detected (e.g., by GPS data associated with the application) as no longer using the services, or when another user registers to pay for the services.
The billing may involve selected outlets, switches, utilities, and usage patterns, but exclude others. A landlord in a location where winters are cold may demand that the thermostat never is set below 50° F., to avoid damage of the property due to freezing of pipes, and may accordingly pay for any heating that is necessary to bring the temperature to this threshold level, but not for energy associated with additional increases of the temperature. Accordingly, the system can determine the estimated costs for the landlord to pay based on the outdoor temperature; historical heating costs at various thermostat settings; user thermostat settings; user actions, such as keeping windows open in the winter; and additional or alternative factors.
The system illustratively also comprises a messaging system that sends notifications to interested parties. For example, a landlord may wish to know if the temperature or the temperature settings exceed or go below preset thresholds, or otherwise are anomalous; whereas the user may wish to know when the energy consumption rate exceeds a threshold value, or may wish to receive a notification every time the user has accrued another $10 bill, or any other amount, as set by the user. Some notifications are generated and transmitted by the third party, whereas others may be generated and transmitted by the trusted party. Notifications may either relate to exact consumption amounts or to estimates. Some notifications sent by the third party may be triggered by the trusted party, which is helpful if the third party receives sensor data and associated predicates in a delayed manner, or if two third parties are used for different utilities (e.g., gas and electricity) but the user has opted for a unified billing or notification.
In embodiments where the trusted party is engaged in payment processing and to ensure strict protection of credit card payment data, it is beneficial for the trusted party to interface with a third party that is a Payment Card Industry Data Security Standard (PCI DSS) compliant payment processor. The third party payment processor would obtain only the necessary session data associated with the business transaction in question and the session data associated with the trusted party system would not be exchanged.
This is another example of the method also described above, where a data-processing third party is distinct from a predicate-consuming third party; in the example relating to payment processing, the data exported by the data-processing third party is not predicates, though, but aggregate data corresponding to an energy quantity or a payment amount. Like the use of predicates, this aggregation of input data results in a privacy enhancement. This can be described in the permissions request simply by describing to the end user what different parties will obtain, and what they will output or export. Like for other types of data, users will find it easier to understand and give appropriate permission since the types of data correspond to conceptual quantities (e.g., “The payment processor will only receive information about the amount of electricity you consume on a day-to-day basis, and will not know when you used the electricity. Approve?”)
Yet another example of a third party application is one implemented in the context of a residential or commercial alarm system that comprises a node network and an associated trusted party, configured to detect motion and presence and to optionally identify users based on MAC addresses of user devices and other data. The third party application in this example implements a third party service that receives data derived from the sensor data of the node network and determines whether a situation corresponds to a risk of a home invasion or other covered danger, and communicates with one or more users such as residents and/or landlords, and which also may communicate with law enforcement or private security companies. Such an arrangement is beneficial as it enables protection of residences and commercial spaces without deployment of special-purpose sensors, such as window-break sensors. Moreover, the disclosed technology is expected to result in lower operational costs, higher precision, greater opportunities for customization, as well as other advantages.
Multiple competing third parties can use different algorithms to infer risk from data derived from sensor data and auxiliary data provided by the users and/or their appliances and devices, such as a GPS device associated with a user vehicle. The accuracy of advanced sensor networks is expected to be much better than that of traditional systems, which typically rely on a very small number of sensors, most of which are battery operated, and therefore severely constrained in terms of computational capabilities, amount of data communicated, and other functionality.
The third party service provider may also provide additional sensors, such as one or more video cameras, and collect data from these additional sensors. Pre-existing sensors associated with the third party can also be incorporated in the system. This is illustratively done by enabling them to communicate with the node network. Such an arrangement can reduce service costs by avoiding the use of cell phone modems for the communication. It also simplifies the user interface as it can be configured to incorporate all access under a single portal, in the form of an end-user portal for accessing the node network and its associated services. This is illustratively an application for a phone or tablet, dedicated software for a laptop or desktop, or a webpage that is accessible to registered end users.
As in the previous examples, there are many options for what sets of permissions can be requested for data and predicates. For example, one third party service provider may request access to the MAC addresses observed by the system, whereas another may request access to predicates relating to a predicate that has only three values: no detected MAC addresses, the appearance of a recognized MAC address, and the appearance of a MAC address that is not recognized, where a MAC address is said to be recognized if it is associated with the installation and is understood to correspond to a legitimate resident or user. Thus, in this example, the actual MAC addresses would not be revealed, but only the classification into one of the three classes described above. As will be understood by a person of skill in the art, this is only one of the types of data required for the application; for the other ones, there are similar options in terms of the requested permissions. The generation of predicates can be performed by the system or by one or more third party data processors, as described above.
External data sources can be accessed using pull and/or push technologies. For example, a resource can provide the trusted party with API access to data for a particular feed or it can transmit updates using a push mechanism. Multiple mechanisms can be used at the same time, to conform to the communication methods already used by the external data source. For example, one external data source can transmit data using standard emails while another may access an API to the trusted party to upload data. The trusted party may mimic the format of an existing application. For example, to get calendar update data, the trusted party may generate an interface matching that of a mailbox, allowing the sharing with the trusted party. This mailbox may correspond to an existing email account of the user whose calendar data is to be uploaded to the trusted party.
The user can sign up for the service by requesting that calendar data be shared with an application corresponding to the trusted party, which illustratively mimics a mailbox or a calendar. Other example external data sources include vehicles and their infotainment systems, phones and their GPS systems, smart appliances that permit uploading of data or connection to services, which can be mimicked by the trusted party to simplify compatibility.
Systems that do not produce data, but instead consume data, can also be connected, such as smart light bulbs. By connecting these to an account of a user, and by permitting access by the trusted party, the user can control such devices from the same portal used to control the other nodes in the network. External units that both generate and consume data can be connected in a manner similar to that described above, and their data may be processed by the system; by one or more third party data processors; one or more data consuming third party applications; or a combination of these.
In contexts where third parties or users have data, such as sensor data from sensors they control, manage or distribute, or otherwise receive a feed from, these third parties or users can provide the sensor data to the trusted party, along with a policy describing how the sensor data can be used, and associated with a profile, e.g., a pseudonym or identifier associated with the installation or user. Additional data feeds can be imported to the trusted party in the same way, e.g., email feeds, calendar feeds, and social networking feeds. These can be requested or initiated by an end user associated with the data or data source, and can be associated with a specified policy of use.
One example policy of use is that the data is not allowed to be used to compute any predicate that is provided to a third party. Another example policy is that the data provided to a third party cannot possibly be used to determine personally identifiable information (PII). Yet another policy is that the data must not be associated with the identity of the user, but only with a temporal pseudonym relating to the user's data. A further policy can be a delegated usage policy, meaning the data can be used as seen appropriate by the trusted party, given that the user has a trust relationship with that party.
Other examples of policies include policies that specify what other third parties are allowed to access the provided data or predicates derived from it. These and other policies are illustratively selected by the user, but in some embodiments such policy selection can involve interaction with a third party providing certain types of data. Such approaches are beneficial as they can expand the sensor data from that related directly to sensor hardware associated with the trusted party, allowing essentially any data feed of a standardized format.
In one embodiment, the processing performed by third party service providers, corresponding to respective third party applications, is performed on a cloud platform or other computational platform controlled by or associated with the trusted party. Such a cloud platform may comprise, for example, a cloud hosting environment associated with the trusted party. This enables some degree of oversight and control by the trusted party, and also simplifies and speeds up the processing of sensor data and associated predicate data.
One form of oversight is that the trusted party can scan submitted apps or code units to determine that they comply with security and privacy requirements, similar to how app marketplaces screen apps before they are made available to users. Another is that the trusted party can determine when the apps pass messages to other entities, whether operating in the same platform environment or an external environment, and, under some circ*mstances, determine the nature of the messages.
In an embodiment with third party service providers' applications hosted within the trusted party hosting environment, it is beneficial to have operating agreements to safeguard the health, security, and privacy of the enterprise, such as security and privacy attestation by the third party, and security and auditability of third party code via the trusted party. Additionally or alternatively, the trusted party could be provided with administrative rights to enable, disable and remove hosted applications that could present an operational, security, or privacy risk to the enterprise or its users. The trusted party hosting environment can provide environment segregation so as to limit any potential scope of a security risk through the use and implementation of security best practices, industry standards and required governance compliance.
When multiple third parties are connected in a chain, e.g., consume each other's data, it is particularly beneficial for at least a portion of the processing to be performed in one computation environment, as opposed to in a distributed fashion. In one embodiment, at least portions of the computation are performed in computational environments that are not under control by or associated with the trusted party. This is beneficial for third parties that cannot or do not wish to expose at least some of their processing to the trusted party, or that rely on processing that is intrinsically external to the processing environment associated with the node networks. For example, an advertising network may use data from the trusted party to determine preferences of end users, but the application of this data to the selection of advertisem*nts should illustratively be possible to perform on a computational platform controlled by the advertisem*nt network or a third party associated with the advertisem*nt network, as opposed to on the infrastructure controlled by the trusted party.
In this example, the functionality is partitioned into different aspects, some of which are performed external to the platform associated with the trusted party, whereas other parts are performed either on the platform associated with the trusted party or on another platform. There are processing benefits associated with performing at least some processing on the platform associated with the trusted party. Some of this processing is under the full control of the trusted party, i.e., the code is provided by the trusted party or associated entities, whereas other parts of the processing may be performed by one or more third party service providers on the platform associated with or controlled by the trusted party. A person skilled in the art will appreciate that a tight integration of processes, in which code associated with the trusted party and one or more third parties is executing on the same processing platform, results in benefits of many types, including but not limited to faster processing, lower communication costs, greater robustness, and better oversight.
In another example, there is a third party application that relates to a fire alarm. This application requests the following permissions from a user interested in installing it:
1. Access to information indicating high risk for fire.
2. Control over lights and speakers in case of a detected fire.
3. Full access to microphone data in case of a detected fire.
4. (Optional.) Access to location information including address.
5. (Optional.) Control over in-home emergency alert notifications.
Here, the first request item corresponds to a high-level predicate that reveals only a high risk for fire, but which may be based on temperature sensors, energy consumption sensors, motion sensors, microphones, and more. The second request item is a capability, namely the control over lights and speakers. This example capability is a conditional capability, where the condition is evaluated by the system, and access to control is granted only when the condition is satisfied. The third request, which in combination with the second item turns a home at risk into a giant intercom, allowing manual and automated messaging and guidance, is a conditional permission. The fourth item is optional, which means that the user can agree to request items 1-3 only and install the application without the associated functionality of request item 4, or the user can agree to all request items and also obtain functionality associated with request item 4. Request item 4 allows the third party application to notify emergency crews and direct them to the location of the fire. However, since item 4 is not conditional on a detected fire at the premises of the installation, it also allows the service provider corresponding to the application to identify a high risk of fire in the neighborhood, in spite of this not having spread to the location of the installation in question, and to notify the users of risk. This can be done using the capability corresponding with request item 5, which is not a conditional capability, and which allows the application provider to notify the user of danger. Request item 5 does not provide full access to the speakers, as the second request item does, but only access to transmitting pre-recorded and system-approved alerts. Like item 4, item 5 is conditional, and a user that chooses not to allow the capability of item 5 also chooses not to receive the alarm that is generated in response to a risk that is not related to the detection of fire on the premises (i.e., request item 1), but another risk. For example, item 5 may relate to a fire in the neighborhood, as mentioned above, but could also relate to other risks, such as a tsunami. This is not specified in this particular permissions request, but another application provider providing fire and tsunami alerts may have permissions that are different from the permissions described above.
Traditional permission requests are abstract and broad; for example, a user may be asked to grant permission to a service to see his or her contacts as well as all his or her email. It is hard for typical users to know what the actual privacy impact is from such a permission, and many people do not realize the types of abusive-feeling behavior they are agreeing to, as a result.
The privacy associated with the disclosed methods benefit from user-understandable quantification of data disclosure, where the data disclosure is typically limited to more concrete and much more limited quantities. For example, a service provider may request permissions to microphone data, but only after the system has detected an attention phrase (such as “Ok Google” or “Alexa”) or corresponding action (such as pressing and holding down a button for two seconds), and only for the duration of a verbal request (e.g., ending as the user stops speaking or after a maximum period, such as 30 seconds). This is much more limited in that unconditional microphone access is harder to abuse, and can be expressed as a simple and concrete permissions request, e.g., “Do you want to grant access to your microphone for up to 30 seconds after the system detects a request from you?” Here, the system would listen to the microphone data all the time, but would not process or share data until the attention phrase (or action) is detected. While the concreteness is beneficial, it is not necessary, however. For example, consider access to a combination of sensor data and other potential data, and refer to this particular combination as “the gamma data.” A user can be asked if he wishes to grant access to the gamma data to a third party. Whereas gamma data does not mean anything to users today, and may be too complex and abstract to explain in a meaningful way to all but experts, many users would be willing to agree to granting full access to gamma data to an entity they do not have reason to mistrust, provided a person or organization they trust would support the sharing of gamma data. This way, complex or abstract collections of data can be given a name, experts can analyze the privacy implications and publish opinions, and rating agencies and consumer ombudsmen can publish privacy impact numbers (e.g., “Sharing gamma data has a privacy impact of 4.3 according to your collection of trusted authorities.”)
Turning now to the generation of predicates from sensor data and other data sources, this can be done in a variety of ways, and result in a variety of predicates, which can be used to convey permission requests, enable permissions, and then, to perform processing in third party applications.
Additional illustrative embodiments will now be described with reference to FIGS. 9 through 11.
FIG. 9 shows an example user interface, including a window 901 showing a service provider or app logo 902, a service provider or app name 903, a service or app description 904, a first privacy rating 905, an image 906 representing the type of sensor data associated with the first privacy rating 905, and a first permission description 907 associated with the first privacy rating 905. Further, the figure shows a second privacy rating 908, an image 909 representing the type of sensor data associated with the second privacy rating 908, and a second permission description 910 associated with the second privacy rating 908. The user is informed in dialog box 911 that there are additional permission requests of similar types but of lesser privacy impact, where additional details can be obtained by following the instructions. If the user approves of the service provider/app represented by logo 902, name 903, and description 904, and wishes to grant access to the data corresponding to the permission requests described in descriptors 907 and 910, then the user clicks on clickable button 912, which initiates the access provision to the user data according to the description associated with the information displayed in window 901. The user may also decide not to grant access by clicking on button 913.
The data corresponding to descriptor 907 and 910 correspond to predicates P1 1121 and P2 1122 of FIG. 11, where a predicate is one of a combination of data from nodes A1 1101, A2 1102, B1 1111 and B2 1112 of FIG. 11; an obfuscation of data from nodes A1 1101, A2 1102, B1 1111 and B2 1112 of FIG. 11; or another function of data from nodes A1 1101, A2 1102, B1 1111 and B2 1112 of FIG. 11. Some predicates, such as predicate P2 1122, correspond directly to at least some of the output of a node such as node A2 1102, without any combination or obfuscation performed.
FIG. 10 shows an example user interface, including window 1001 showing the logos 1002 and 1012 of two service providers, the names 1003 and 1013 of the two service providers, service descriptions 1007 and 1017 of the two service providers, and privacy indicator icons 1004 and 1014 indicating the type of data request that corresponds to the highest privacy impact, where the ratings 1005 and 1015 are the corresponding privacy impacts of the requests with highest privacy impact. Ratings 1006 and 1016 show the user-provided ratings for the two services relating to their experience with the provided services. Indicator 1018 explains the meaning of icons 1004 and 1014 and associated ratings 1005 and 1015; indicator 1018 is clickable, allowing a user to select another criteria of comparing the services. Clickable indicator 1019 allows a user to apply filters, such as only showing services with a user rating exceeding a threshold value such as 3 out of 5, only showing services with privacy impacts below 7 out of 10, etc. By reviewing the information shown in window 1001, the user will see that the first application, corresponding to logo 1002, has a more favorable user rating 1006 than the second application corresponding to logo 1012, whose user rating is shown in box 1016. The user can also see that the privacy rating 1005 of the first application is more favorable than the privacy rating 1015 of the second application. Additional information, not shown in the figure, can indicate purchase price, whether the user will receive coupons, whether the user will receive advertisem*nts, and whether there are in-app purchases, which correspond to additional features that the user can select. Example additional features include features such as automated logging of alerts, notification by email, removal of advertisem*nts, etc.
FIG. 11 shows installation A 1100 and installation B 1110, which are two independent sensor network installations, comprising node A1 1101, node A2 1102, node B1 1111 and node B2 1112. Each node comprises at least one sensor. Installation A 1100 transmits data related to at least one of node A1 and node A2 to pre-processor 1120, and installation B 1110 transmits data related to at least one of node B1 and node B2 to pre-processor 1120. Pre-processor 1120 generates predicate P1 1121, predicate P2 1122 and predicate Pn 1123 and transmits predicate P1 1121 and predicate P2 1122 to service X 1130, and predicate P2 1122 to service Y 1140. Predicate P1 1121 is a function of data obtained from installation A 1100, and is provided to service X 1130 since a user associated with installation A 1100 has granted permission for service X 1130 to receive access to predicate P1 1121. If the user retracts this permission, service X 1130 will no longer receive access to predicate P1 1121. Access to predicates by service X 1130 can be performed either using a pull or a push architecture, as can the access to data from Installation A 1100 by pre-processor 1120.
Additional illustrative embodiments will now be described in further detail with reference to FIGS. 12 through 17.
These embodiments are advantageously configured to provide an architecture aligning terms of service with permissions.
As indicated previously herein, user privacy is, increasingly, a source of contention. There is an often-tense relationship between data originators (such as end users) and data consumers (such as advertisers and service providers.) Another aspect is the difficulty with which privacy can be quantified and understood, and therefore compared. This difficulty aggravates the tension by encouraging predatory behavior among data consumers.
At the heart of the problem is the way in which data, permissions to the data, and algorithmic objectives are handled. With the increased deployment of IoT installations, with their associated massive growth of data, this problem is bound to get worse unless addressed in a thoughtful manner. Illustrative embodiments herein provide a foundation for a structural change to improve privacy. Approaches disclosed herein constitute an important alternative to increased regulation as well as an opportunity for big-data companies to improve their image in the eyes of the public.
Privacy is notoriously difficult, both to enforce and for users to understand. From the perspective of typical end users, EULAs are hard to comprehend due to their use of legalese and rarely read simply due to their excessive length. However, brevity by itself is not the answer to the problem, as demonstrated by the permissions users have to grant third parties for these to get data access. While these fit comfortably in a small pop-up window, they are much too broad to be meaningful—for example, typical permissions may grant access to all emails, all data related to contacts, and all location information—without any explanation of how the data will be used or why it is needed.
As a result of privacy agreements—whether in the form of EULAs or permission requests—being hard to comprehend for typical end users, it is also hard for these users to evaluate whether a company's use of their data is reasonable. Similarly, it is near-impossible for users to compare the privacy impacts of two competing products. This leads to poor choices, often motivated by fear or fatalism, and to widespread disappointment with the seemingly predatory behavior of service providers. This perception of abuse is often well founded, and is fueled by the over-broad access to data as much as it is by a lack of appreciation among companies of the importance of offering meaningful privacy assurances.
Most companies argue that being quick to the market is more important than providing solid privacy protections, since being late might mean not getting any customers—and without customers, any privacy expenditure would be a wasted effort. Instead, companies protect themselves against the risk they are exposing themselves to—namely a risk of being named in a lawsuit—by broad and demanding EULAs, which closes the loop and fuels the rise of the problem.
This is a worrisome situation, and it is our belief that without strong efforts to break this negative cycle, things may soon spiral out of control. This is not only for the reasons outlined above, but also due to the rapid growth of data collection in the context of IoT, including wearable technology and home and office automation technologies. With dramatically greater varieties and quantities of data, much of which is sensitive, there is a tangible risk of abuse—especially as corporations and political entities increasingly wake up to the competitive advantages they can attain by profiling users—and an environment in which very little holds them back.
In the following, we briefly ground our effort in stated end-user preferences and formalize these in terms of a set of technical goals. This is followed by an architectural overview; and then by a description of user interfaces and permissions. We then explore one way to bootstrap privacy impact estimates, followed by a review of expected consequences of the deployment of our general system.
With regard to technical goals, we will now outline how illustrative embodiments herein address the current privacy situation, starting with an overview of end-user priorities and continuing with a list of goals.
As part of the effort leading up to the design of the technical constructions of some embodiments herein, we performed surveys among users on Amazon Mechanical Turk (MTurk), a popular crowd-sourcing platform. In one representative survey, we asked 254 MTurk workers the following two questions relating to the use of end-user data:
Question 1. *Company* determines the temperature in homes using their thermostat and compares that to energy consumption. This way, users can be notified if their insulation appears to be poor, causing them to pay unnecessarily high gas and electric bills. Is that reasonable? (Choose all that apply.)
The options subjects could choose from (ordered here in terms of order of decreasing commonality) were:
1. This is ok as long as everybody knows this before they buy a *Company* thermostat. (Selected by 52%.)
2. This helps the conservation of energy, which is good for society. (Selected by 52%)
3. This could be useful, especially if *Company* could also find local deals for insulation and suggest reputable contractors. (Selected by 43%.)
4. Most people know already whether their insulation is good or not, so this is pretty useless. (Selected by 8%.)
5. This is abusive. (Selected by 6%.)
Question 2. Google uses search queries to detect and predict flu outbreaks. When many people in one zip code search for “headache,” for example, that is indicative of a rise of infections. Is this reasonable? (Check all that apply.)
The options subjects could choose from were:
1. The data is anonymized and is not even very sensitive, so this is not a problem. (Selected by 58%.)
2. This is a public service and should be applauded. (Selected by 31%.)
3. Google serves targeted advertisem*nts based on search requests, too. This is not acceptable. (Selected by 26%.)
4. Google should not be allowed to use any search data without explicit user permission. (Selected by 25%.)
5. I do not use Google exactly because of things like this. (Selected by 4%.)
Out of the respondents, 41% were women and 59% men; 1% were 18-20 years old; 36% were 21-30 years old; 38% were 31-40 years old; 13% were 41-50 years old, and 11% were 51 years or older. 1% preferred not to disclose their age, and 54 were MTurk Masters. All were U.S. residents.
While the number of subjects in the survey is rather small, the findings are informative. They tell us, first of all, that informed consent is critical (“ok as long as everybody knows this before they buy” and “should not be allowed . . . without explicit user permission”.) A second valuable insight is that use of personal data is generally acceptable if the resulting outputs do not reveal personal data (“data is anonymized and is not even very sensitive”)—especially if there is a societal benefit (“which is good for society” and “a public service . . . should be applauded”.) Another interesting observation is the preference for coupons over advertisem*nts, suggesting the perceptional value of direct end user benefits in comparison to indirect benefits.
At the heart of the problem is the notion of permissions, and aligning these both with end-user needs and functional requirements. To make this possible, illustrative embodiments herein restructure permissions to fit the following example categories:
1. Conceptual Permissions. As far as possible, permissions should correspond to accessible mental concepts. For example, users will readily understand the privacy impact of a third party being able to determine whether two people trying to locate each other in a busy mall are headed towards each other or not—this corresponds to a clear mental concept. On the other hand, it is much harder to understand the privacy impact of this third party being able to identify the location and social network of users having opted in. This is not only because the latter set of permissions is broader, but it also enables many unspoken combinations of the data. As the number of data sources increase, this burden of understanding the impact snowballs; in contrast, with conceptual permissions, it does not.
2. Purpose-based Permissions. It is beneficial to tie permissions to functional purposes. When end users are able match the contents of a permission request to their understanding of the delivered functionality of an application, this is helpful. For example, if a dictation application asks for access to microphone data, the reason will be clear to most end users; however, if the same application asks for access to microphone and location data, this is not clear.
3. Conditional Permissions. A conditional permission corresponds to the right to access data when a particular event takes place. For example, many end users are likely to find it unreasonable for a fire alarm application to have access to microphone data, and an ability to control speakers; however, if these rights are tied to a situation in which a fire is detected and the fire department called, most end users are likely to find the access request much more reasonable.
4. Differentiable Permissions. Permissions that can be differentiated allow a comparison of at least two distinct sets of permissions, side by side, by a typical end user, allowing the end user to determine the relative desirability of the two or more sets. Permissions that are differentiable support a race, among data consumers, to a low-impact use of data by allowing for direct comparison. Thus, a data consumer with a broad permissions request (and, accordingly, privacy-invasive abilities) would be seen as less desirable than a data consumer with a less broad/invasive request.
Referring now to FIG. 12, an information processing system 1200 configured in accordance with an example architecture supporting improved privacy is shown. The system 1200 in accordance with the architecture illustratively comprises an interface 1202, a processing environment 1204, and a set of third parties 1206.
The interface 1202 illustratively comprises a set of nodes 1210, each of at least a subset of which has one or more sensors and/or may be controlled using techniques disclosed herein. For example, an outlet or a switch may correspond to a node, and may include sonar/microphone, wireless radio units, temperature sensors, humidity sensors, particle sensors and/or energy consumption sensors; and permit remote switching and dimming. The interface 1202 generates observations 1212 which are transmitted to the processing environment 1204, and receives actuation control signals 1214 from the processing environment 1204.
Data observed by the interface 1202 includes observations 1212 transmitted to edge and/or central processing components 1216 of the processing environment 1204. Accordingly, the processing environment 1204 can be part edge-computing, part central computing (which, in turn, can be implemented both using dedicated servers and cloud computing approaches). The processing environment 1204 receives observations 1212 (such as sonar and wireless data) and creates interpretations of the data (such as an indication that a known user enters a room); these interpretations are examples of what are more generally referred to herein as predicates 1218. The system 1200 can then actuate based on the predicates 1218—for example, it may determine that if person A (who likes it warm) is in a room, but person B (who likes it cold) is not, then the temperature should be set higher than if person B is present but person A is not. The processing environment 1204 can also grant access to the computed predicates to third parties 1206, which can then provide actuation requests to control components 1220 and 1222 of the processing environment 1204 for the system 1200 to act on.
The third parties 1206 include respective third party processing components 1224 and 1226, and respective output components 1228 and 1230. A given one of the third parties 1206 receives data from the processing environment 1204 and uses this to perform a service. Traditionally, the data the third party receives is broad—such as all data of some types, generated by the user. This is a contributor to the privacy problem as the capabilities that come with very broad data access are poorly defined, from the perspective of the end user and the trusted party, neither of which can know for sure how the data is used. In illustrative embodiments herein, third party providers such as third parties 1206 would receive predicates instead of raw data. Since these predicates are much narrower than the underlying raw data, this results in immediate privacy benefits. Given the importance of third party processing in extending new functionality to end users, and the fact that many third parties are much less known than the trusted parties themselves, limiting data access of third parties has a profound effect on privacy by itself, irrespective of the re-architected permissions of illustrative embodiments, which will be described in more detail below.
The nodes 1210 of the interface 1202 generate observations 1212 for delivery to the processing environment 1204, and receive actuation commands and other actuation control signals 1214 from the processing environment 1204. The processing environment 1204 computes predicates from the observations 1212. These are used to perform services, whether by the processing environment 1204 or by one or more of the third parties 1206. The services result in actuation and in information output—such as automated calls to emergency response, selections of advertisem*nts or coupons, and generation of anonymized user statistics. The processing environment 1204 and third parties 1206 maintain state, e.g., for purposes of personalization. Privacy is defined relative to the third parties.
For the sake of simplicity, we assume herein that the two processing environments—the edge computing environment and the central processing environment—are controlled by one and the same party, which we refer to as the trusted party. The choice of this name is made to reflect that in typical environments, the end users have to trust the party that controls the software governing the collection of data and actuation of the end-user system: if the user were not to trust this party, she would have to write the software herself. However, there may be systems in which the edge computing and the central processing are not controlled by the same party; while that further adds requirements and constraints, it does not result in a solution that departs from the principles described herein.
Thus, the architecture we describe in the FIG. 12 embodiment does not protect user data against the trusted system; doing so is a near-impossible problem given that typical users do not have the technical skills to determine what the trusted system truly does, even if provided with source code. Instead, we focus on the somewhat more tractable problem of protecting user data, observed by the trusted system, against third parties. The trusted system may comprise an IoT installation, a laptop, an operating system or a search engine, or combinations of these and other components, while the third party may comprise an advertiser, a content provider, a burglary alarm system, or some other form of service provider.
The FIG. 12 architecture supports sharing of data with third parties, while also limiting the exposure of end-user data to these third parties. One benefit of this example approach is that sharing predicates instead of raw data reduces the privacy impact, since the predicates are aggregations of data from which the raw data cannot be computed. The move to sharing of predicates with third parties, instead of raw data, can be seen as a shifting of computational responsibilities from third parties to the trusted party (and typically, to the central processing environment).
Sharing predicates instead of raw data is only meaningful as long as the predicates are useful. It is believed that many, if not most, third party applications can operate using predicates instead of raw data. Accordingly, we will now describe a set of predicates and describe how these can be used in a broad collection of third party applications.
For sake of concreteness, we will focus in the example context of home automation, but note that the very same set of principles apply to commercial settings. Moreover, we believe that these methods are also applicable to “traditional” privacy contexts, such as services related to email events, web browsing events, and more, but do not focus on these contexts in the FIG. 12 embodiment.
Consider the following example first-order predicates, which are predicates derived directly from raw sensor data:
1. Presence. The presence predicate identifies that a person is present at a given location, where a granularity aspect may be used to signal the location within a room; as an entire room; or as a larger space. The trusted system may determine the presence predicate using sonar, wireless radio, and interactions with controllers.
2. Layout. The layout predicate identifies distinct portions of a space; assigns them names based on their observed uses; and details how they are connected to each other. The latter can be determined by identifying movements of users and their mobile devices over time, thereby determining what rooms are connected to each other and what rooms have entrances from the outside. For example, the presence—identified based on energy consumption patterns—of a microwave oven suggests that a room is a kitchen, while the presence of a sleeping person during the night suggests a room is a bedroom. By combining multiple heuristic assessments of this type, a good description of the space use can be established.
An example layout is shown in FIG. 13. This figure illustrates an inferred layout 1300 of a space comprising four separate areas, denoted 1301, 1302, 1303 and 1304, also referred to by respective numbers 1, 2, 3 and 4. The letters correspond to sensor nodes, whose relative locations are determined based on correlating signals and time series of signals. In this example layout, area 1 has four sensor nodes, two of which (A and B) detect a person with stronger signal than the others (C and D) as he/she comes in from the outside. As a person goes from room 1 to room 2, the signal is always strongest with node D right before the person is detected by sensor nodes E and F, at which time the signal from A and B is weak. This pattern of detection allows for an approximate identification of the size of room 1, as well as the relative locations of the four nodes in the room. The system knows that there is a door between room 1 and 2 and one between room 2 and 3, as sounds are sometimes detected within all rooms, and sometimes only within one of these three rooms. Room 4 does not have any sensors, but is distinguished from the outside by the fact that a user who leaves room 2 without entering rooms 1 or 3 always returns to room 2 within a short time—the same is not true for a person leaving to the outside from room 1. A user in room 3 always leaves this rooms to enter room 2; therefore, the system knows that room 3 has no exit to the outside.
3. Activity. The activity predicate describes what is happening in a space, such as, sleeping, watching TV, cooking, playing music, etc. It can be derived from movement patterns, sounds and energy consumption patterns; the activity predicates may be grouped in categories and subcategories, where the latter could, for example, identify the type of TV program—for example, “watching soccer,” “watching football,” “watching news”, “watching cartoons”—all of which can easily be determined based on analysis of the frequency distribution of detected sounds.
4. Pseudonym. The pseudonym predicate is a long-lived label (such as “user 3,” “adult,” or “recognized party”) that can be used for purposes of configuration while hiding the underlying user identity from the third party. Accordingly, the system may offer different types of pseudonym predicates to third parties.
In some implementations of system 1200, the pseudonym predicate can be determined by the trusted system by mapping a long-lived wireless identifier (such as a MAC address or a Bluetooth device identifier) to a local pseudonym or descriptor. By making mobile devices consumer centerpieces in terms of control and automated personalization, automated user identification and tracking will become within reach even for entry-level systems, and based on this, generation of pseudonyms can be performed.
Although automatically determining age or gender may be beyond the reach for typical consumer systems, it is easy to produce such predicates in some embodiments herein simply by combining registration data (e.g., of user accounts associated with various mobile devices) with the device detection. While imperfect, these classifications are likely to be correct most of the time, and therefore useful.
The example predicates above can be used in a variety of ways. For example, the presence predicate combined with the pseudonym predicate can be used to provide customization—of light intensity and temperature, for example. Combining these two predicates with an activity-related predicate, a third party can identify an opportune time to send a discount coupon for a meal delivery service, or generate an advertisem*nt that appeals to users who watch soccer on TV. In addition to the example predicates described above, related to movement and activity, the system 1200 can generate first-order predicates relating to safety (e.g., risk for fire); health (e.g., risk for mold), and many others.
The system 1200 is also configured to generate second-order predicates. Second-order predicates are predicates that are derived from other predicates. For example, if the system 1200 has generated predicates for presence pseudonyms, it can determine if an arriving person is recognized. If so, it generates a second-order predicate that we may call known-person-arrives, while if an arrival is detected without the arriving person being detected, it generates a second-order predicate that we may call stranger-arrives. Such a predicate is very useful for a third party that implements a service for home intrusion detection; it could, for example, be used both to automatically disarm an armed system (when a recognized person arrives) and to sound an alarm if the system is not manually disarmed within some set time.
Another example of a second-order predicate is a predicate that corresponds to the presence of a person in a room without an entrance/exit to the outside, without this person having first been detected in an adjacent room. This predicate, which utilizes the layout predicate in addition to a time series of the presence predicate, can be used to identify another possible intrusion—namely a person potentially climbing in through a window. This may result in a third party service generating an alarm—even if the system was not already armed.
Yet another example of a second-order predicate is one that uses size estimates corresponding to the layout with temperature data and energy consumption data—from multiple different homes or offices—and identifies locations where there may be a need for better insulation, based on identifying unusually high energy costs. Without disclosing the underlying data sources, the trusted system can share predicates relating to locations with need for improved insulation—for example, with resellers of insulation products—or generate coupons that home owners can use to get discounts for selected insulation products.
On a more general level, second-order predicates may also capture demographic insights, preference classifications, and other similarly complex notions.
Second-order predicates have a lesser impact on privacy than the predicates they are derived from, just like predicates have a lesser impact on privacy than the raw data from which they are derived. This is because as data (whether raw data or first-order predicates) are combined or aggregated, information is irretrievably lost; as data is processed in this manner, it becomes increasingly conceptualized.
However, the fact that different predicates have different privacy impacts is an important observation, as this invites a quantification in which the trusted system—or a third party, such as a consumer representative—generates privacy impact ratings of different predicates, allowing an easy comparison of third party applications in terms of the sensitivity of the data they require access to. This will be described in greater detail below, in aspects of illustrative embodiments relating to connecting predicates and permissions.
One benefit of the use of predicates is the sharing of aggregated data instead of raw data, as described previously. Another and more significant benefit, which we will detail herein, is to make permissions comprehensible—and therefore actionable.
With regard to conceptual permissions, while raw data does not lend itself well to be understood—at least not by typical originators of the data—predicates have the potential of being conceptual. This means that permissions, based on conceptual quantities—we refer to this as conceptual permissions—are useful for end users having to determine whether to grant access or not. Examples of permission requests based on these and other techniques disclosed herein are shown in FIGS. 14 and 15.
FIG. 14 shows an example of a conceptual permission request 1400, illustratively generated by the system 1200 and presented on a display screen or other user interface of a user device such as a computer or mobile telephone. The icons represent the types of predicates being requested, and the texts to the right of the icons describe the requests and their estimated impact on the user's privacy according to some metrics. The permissions are conceptual in that they allow a user to understand what type of data is being requested, and they are purpose-based in that the permissions help the user align the requests with the functionality of the application. The third and fourth request, in addition, are conditional, in that the data/control is accessible only in conjunction with the occurrence of specified events (e.g., detection of a sufficient level of risk). The use of juxtaposed privacy impact ratings permit differentiation between the various permissions that may be granted.
FIG. 15 shows a second example permission request 1500. The third party service may have the same general functionality as the third party service for which permissions are requested in the example of FIG. 14. However, the permission requests shown in FIG. 15 are broader, as reflected by the higher privacy impact ratings. The use of ratings like this allows consumers to compare services in terms of their privacy impact, which helps encourage the use of low-impact access requests by providing a competitive advantage to services with lesser privacy impact, all other things being equal.
It is not just data originators, but also application writers, that can understand the world in terms of conceptual quantities (e.g., the location or activity of a person; a pseudonym; or an indicator of danger) as opposed to raw data. Therefore, a shift towards conceptual quantifications of data using illustrative embodiments disclosed herein is likely to aid the development of third party services, as opposed to hold it back. The obvious exception to this are processes that use machine learning, applied to vast data quantities, to understand the world. A development of conceptual predicates limits the need for such processing among third parties, and in fact shifts the need for machine learning methods to the party computing the predicates. Therefore, the methods may not change, but the exposure of the data does: instead of anybody needing access to sensitive data, mostly only the initial data processor—the trusted party—does.
In some embodiments, third parties can still request access to raw data; however, this might be looked upon with skepticism by data originators.
To the extent that the trusted party does not offer suitable predicates, this opens up opportunities among third party developers, given appropriate support for third party computation of predicates. These third party predicates—like any other predicates—would be available to service providers for a fee. Illustrative embodiments can utilize a variety of different trust models and techniques for third party predicate generation.
In some embodiments, purpose-based permissions as used. Loosely speaking, this corresponds to end users understanding why the third party needs access to the data. This may be best described using an example; while it is likely to be clear to users why the third party permission request 1400 in FIG. 14 includes the individual requests (such as “data indicating intrusion”), it may not be equally evident why the third party permission request 1500 in FIG. 15 includes a request for the user's cell phone id and location. Although this information can be useful to the third party application, for example, to determine whether a person entering is a legitimate resident, this connection may not be very clear to the user. Thus, the permission requests in FIG. 14 are more purpose-based than those in FIG. 15. The notion of purpose-based is not the same as conceptual, but rather, describes the apparent alignment between the request and the stated purpose of the application.
A conditional permission is one that only results in access (whether to data or control) if some event takes place, where this event is determinable by the trusted party. FIG. 14 shows two conditional permissions—one to a predicate (“Full microphone access, if risk is detected”), another to a control capability (“Control of speakers, lights if risk detected”). Conditional permissions have a lower privacy impact than permissions that are not conditional, and are therefore preferable to end users.
Permissions are differentiable if they allow a user to compare two sets of permission requests in terms of their privacy impact. Including privacy impact ratings, as shown in the examples of FIGS. 14 and 15, makes the associated permissions differentiable. While these two figures illustrate the concept, alternative techniques can be used to provide permission request differentiation, such as, for example, combining the multiple permission requests to form a combined privacy impact score.
We note that the decision between two or more applications is not limited to a comparison of privacy impacts alone, but can additionally or alternatively include other relevant factors, such as price, functionality and familiarity. Nevertheless, allowing a comparison with respect to privacy, in the manner described in conjunction with illustrative embodiments herein, will help consumers make educated decisions, whatever priorities they have.
In the examples above, we described the use of differentiable permissions as a method of enabling comparisons between competing services. The associated privacy impact ratings can be generated by an expert panel, e.g., involving representatives of the trusted system, privacy researchers and consumer representatives, or using any of a wide variety of other techniques. For example, another approach, which is particularly useful to bootstrap the system, is to simply rely on consumer assessments.
With reference now to FIGS. 16 and 17, we illustrate the feasibility of this latter approach by describing the results of one small survey, involving 177 Amazon MTurk workers, all of which are residents of the U.S. In the survey, respondents were asked to rate the privacy impact of five different data uses. The vertical axes in FIGS. 16 and 17 denote percentage of respondents, and the horizontal axes denote privacy impact.
FIG. 16 shows the cumulative distributions for two related questions related to hypothetical permissions needed by Alexa, where subjects rated the privacy impact (from least at 1 and greatest at 10). The dashed graph corresponds to Alexa's actual design, wherein voice data is transmitted to Amazon only after a wakeup command (“Alexa”) is given; the median privacy impact assessment is 5. The solid graph corresponds to an example design in which all microphone data is transmitted and interpreted; the corresponding median privacy impact is 10.
More particularly, FIG. 16 illustrates the sensitivity of microphone data; in the corresponding questions, survey takers were asked two questions relating to the design—and a hypothetical design—of Alexa. In one design, all microphone data is transmitted to a central authority (Amazon), and in another, only microphone data after a detected wake-up command is transmitted. The survey takers very clearly identified the former as having a significant undesirable privacy impact (10 out of 10), whereas the approach in which voice data is only conditionally transmitted had a perceived privacy impact of only 5 out of 10. Both of these measurements correspond to the median assessment of the respondents.
FIG. 17 shows the cumulative distributions for two questions related to a hypothetical service that turns off the A/C in a home when everybody has left. The dashed graph corresponds to an application that uses local motion sensors to determine when to turn off the A/C; this corresponds to a median privacy impact of 5 out of 10. The solid graph corresponds to an alternative implementation in which the locations of users are instead determined using the GPS coordinates of their cell phones. This is (correctly) seen as being more intrusive, with a median privacy impact of 8.
Accordingly, FIG. 17 quantifies the sensitivity of location data, by illustrating the impact on privacy of determining presence in a home either using motion detectors (dashed graph with privacy impact of 5 out of 10) or using GPS (solid graph with privacy impact of 8 out of 10).
The embodiments of FIGS. 16 and 17 involve determining the perceived privacy impact of a collection of predicates, in example contexts that make sense from purpose-based perspectives. It should be noted that the severity of perceived threats do not necessarily correspond to the actual severity of these threats since people typically exaggerate the risk of spectacular threats in comparison to more mundane threats.
In these and other embodiments, the sharing of predicates instead of raw data limits the exposure of potentially sensitive data by aggregating large quantities of data into less revealing predicates. Additionally, by tailoring the types of information that is shared enables the trusted system to control the types of inferences third parties can make based on end-user data, which limits the amount of trust that has to be placed into these third parties. Moreover, by connecting the predicates to permissions, end users can review the exact disclosures made before agreeing to these. This enables the end user to be responsible for her actions, and by doing that, helps reduce concerns of potential abuse.
Conceptual permissions help end users understand and evaluate permissions requests. Instead of having to blindly trust third parties, or having to understand the technical consequences of the shared data, conceptual permissions—and their associated predicates—help the end users understand the consequences of data sharing. Conceptual permissions make it possible for a typical user to understand the impact of disclosing the corresponding type of information.
Purpose-based permissions help users understand whether a set of permissions is reasonable in the context of the stated use of the data. For example, a typical user would understand that a service provider offering users who are looking for each other have a good reason to require GPS data of users, but another service provider whose goal it is to help turn off the lights in empty rooms would not likely be understood to need such information. Therefore, by making sure that the conceptual permissions align with the expected functionality of a service, concerned users are given the opportunity to assure themselves that requests are reasonable. In other words, purpose-based permissions make it possible for a typical user to evaluate whether it is reasonable to disclose the corresponding type of information.
Conditional permissions, where the release of data would be triggered by an event detected by the trusted system, permit a reduction of exposure. Conditional releases exist today in the context of releases from edge computing environments to central processing environments, enabling voice data to be released only after a wake-up command is detected. Extending this notion to releases to third parties and to access to control capabilities, as described herein, will help further improve privacy assurances.
The use of differentiable permissions will allow end users to compare two or more related service offerings with respect to their privacy impact. Such arrangements enable educated decisions among end users—even more so than the use of conceptual permissions, in fact—and in so doing, help create competition in terms of privacy. Once users can understand the privacy impacts of their choices—and compare their options to each other—this will fuel efforts to limit the exposure of end-user data in order to earn the business of the newly enabled users.
Additional illustrative embodiments will now be described in further detail with reference to FIGS. 18 through 20. These embodiments are illustratively configured to implement various aspects of privacy enhancement using derived data disclosure as described elsewhere herein, and can also implement additional complementary functionality, such as, for example, third party application enablement as described in the above-cited U.S. patent application Ser. No. 16/676,978.
FIG. 18 shows an example information processing system in an illustrative embodiment. A trusted party 1800 receives data from node network 1801, node network 1802, third party 1803, third party 1804, external party 1805 and external party 1807. The data transmitted in 1811 from node network 1801 and the data transmitted in 1812 from node network 1802 comprises sensor data and data derived from sensor data. In addition, users associated with node network 1801 and node network 1802 can provide the trusted party 1800 with configuration data and additional data (not shown). Trusted party 1800 also receives data 1813 from third party 1803, data 1814 from third party 1804, data 1815 from external party 1805, and data 1817 from external party 1807, and generates, from this received data, combined data (not shown), whether by combining data of the various sources, deriving new data based on the data from the various sources, or both.
The combined data, or at least part of it, is transmitted in 1823 to third party 1803, in 1824 to third party 1804, in 1825 to external party 1805, and in 1826 to external party 1806. In addition, the combined data, or at least part of it, is sent in the form of control data 1821 to node network 1801 and control data 1822 to node network 1802. Here, control data 1821 and 1822 configures node network 1801 and node network 1802. At least some of the data 1813 transmitted by third party 1803 is transmitted 1824 to third party 1804. A third party is primarily a source of processing of data, but may also contribute data it obtains from sources other than the trusted party 1800. An external party such as external party 1807 is primarily a source of sensor data or related data, or (as external party 1806) a recipient of control data 1826 used to configure a system associated with external party 1806, but both a recipient of control data 1825 and source of sensor data 1815 as in the example of external party 1805. External parties also optionally perform some processing, but are primarily used as sources or destinations of data.
Node networks 1801 and 1802 are typically under the full control of the trusted party 1800, which may, for example, update firmware associated with individual nodes of node network 1801 and node network 1802. External parties 1805, 1806 and 1807 are typically not under the full control of trusted party 1800, although in some embodiments, trusted party 1800 is provided with access rights to fully control such external parties. Third party 1803 illustratively comprises one or more software processes running on a platform controlled by trusted party 1800, where such a platform could, in one example, be a cloud platform, and in another, a processing environment on the same premises as the processing environment of trusted party 1800. On the other hand, third party 1804 illustratively comprises one or more software processes that are not running on a platform controlled by the trusted party 1800 or on the same premises as the process associated with trusted party 1800. Thus, the messaging corresponding to flows 1813 and 1823 are performed by passing data using APIs, using buffers, or using messaging within one computer system, whereas the flows 1814 and 1824 correspond to transmissions of data over a network, such as the Internet, from one processing environment to another. The external parties correspond at least in part to a physical system that is associated with an installation of sensors or output devices, such as loudspeakers, screens, or actuators.
FIG. 19 shows an example processing flow from the perspective of the trusted party 1800. In step 1901, the trusted party 1800 receives sensor data from node network 1801, where the sensor data comprises at least one of raw data from at least one node of the associated node network and a function of raw data from at least one node of the associated node network. Example raw sensor data comprises motion sensor data. An example of a function of raw data is an FFT of microphone data from a selected time period. In step 1902, trusted party 1800 optionally computes a predicate related to the sensor data received in step 1901. An example predicate is that a user associated with the node network 1801 is in the kitchen, and another example predicate is that a user associated with the node network often listens to jazz music at a loud volume.
In step 1903, trusted party 1800 transmits at least some of the sensor data received in step 1901 and some of the predicate data computed in step 1902 to a first third party, which may be either third party 1803 or third party 1804. In step 1904, the trusted party 1800 receives data from the first third party 1803 or 1804. In optional step 1905, trusted party 1800 transmits data to a second third party, where said data comprises at least in part sensor data received in step 1901, predicate data computed in step 1902, and data from first third party received in 1904. In step 1906, trusted party 1800 receives data from second third party, assuming optional step 1905 was performed. The dashed line in the figure indicates that steps 1905 and 1906 may be bypassed, with the flow instead proceeding from step 1904 into step 1907 as indicated.
In step 1907, trusted party 1800 processes data, where said processed data comprises sensor data received in step 1901, predicate data computed in step 1902, data received from first third party in step 1904, and optional data received from second third party in step 1906 if steps 1905 and 1906 are not bypassed. The result of the processing is illustratively the generation of at least one control signal. In addition, the trusted party 1800 optionally stores at least some of the processed data, for example, in a database or other repository. In step 1908, trusted party 1800 sends the generated control signal to node network 1801.
The particular processing operations shown in the diagrams of FIGS. 18 and 19 above are presented by way of example only, and should not be construed as limiting in any way. For example, the ordering of the steps in FIG. 19 can be varied, and additional or alternative steps can be used in place of the particular steps shown. It is also possible that certain steps can be performed at least in part concurrently with one another instead of in the particular order illustrated in the flow diagrams.
Referring now to FIG. 20, an example information processing system 2000 of an illustrative embodiment is shown. The system 2000 in this embodiment is configured with functionality for privacy enhancement using derived data disclosure to third party service providers. For example, the system 2000 is illustratively configured to execute processes of the type previously described in conjunction with FIGS. 18 and 19. At least some of the smart nodes illustratively comprise respective sensor devices of one or more sensor networks, although additional or alternative types of smart nodes can be used in other embodiments. A given such smart node in some instances includes a limited user interface, but other types of smart nodes can include more complex user interfaces.
The system 2000 in the present embodiment more particularly comprises a central authority 2002 coupled via one or more communication networks 2004 to a plurality of distinct sensor networks 2005-1, . . . 2005-N. Each such sensor network is assumed to comprise a set of sensor devices of the type described elsewhere herein. A given set of sensor devices is also referred to herein as a “collection” of sensor devices. The sensor networks 2005-1, . . . 2005-N interact with respective sets of user devices 2006-1, . . . 2006-N. The sets of user devices 2006 are not necessarily disjoint, as some of the user devices can interact with multiple ones of the sensor networks 2005. The user devices 2006 can comprise, for example, desktop, laptop or tablet computers, mobile telephones, smart watches, gaming systems or other types of devices associated with one or more system users and capable of communicating with central authority 2002 over the one or more communication networks 2004.
Some of the user devices 2006 comprise respective instances of what are generally referred to herein as “mobile devices” and those can include, for example, laptop or tablet computers, mobile telephones or smartphones, smart watches, etc. Other types of mobile devices can be used in other embodiments, and the term “mobile device” as used herein is therefore intended to be broadly construed.
The dashed lines between the user devices 2006 and the sensor networks 2005 are intended to encompass a wide variety of different types of interaction that can be supported in a given implementation. Such interactions are not limited to machine-to-machine communications using wired or wireless communication protocols, but more generally encompass visual, aural, haptic and/or other types of interaction. For example, a sensor device can present an image to a user via a display, and/or play an audio file to a user via a speaker, in conjunction with the implementation of one or more processes relating to privacy enhancement involving third party service providers as described elsewhere herein.
The central authority 2002 illustratively comprises a processing platform that includes one or more processing devices. For example, the central authority 2002 can be implemented as one or more computers, servers or other backend devices that are configured to interact with the sensor networks 2005 and the user devices 2006. In other embodiments, multiple separate central authorities can be provided, each having responsibility for configuration and management of a different subset of the sensor networks 2005. The central authority 2002 can include additional or alternative components, such as, for example, a database or other repository of the type referred to elsewhere herein. The term “central authority” as used herein is intended to be broadly construed, so as to encompass a computer, server or backend device coupled to a network, and should not be viewed as requiring any particular geographical relationship relative to the particular sensor networks 2005 for which it is responsible. It is illustratively referred to as “central” in this embodiment in that it provides a common control point for configuration and management of the sensor networks 2005.
Similarly, the term “backend server” as used herein is intended to be broadly construed, so as to encompass a server that is reachable over a network connection from a user device, and performs various backend-type functions relating configuration and management of nodes of one or more of the sensor networks 2005, such as performing authentication and verification protocols 2012 in conjunction with establishing associations between particular sensor devices or other types of nodes and a user account.
The “user” associated with a particular user account may be an individual human being, such as a homeowner, tenant or installer, but the term “user” should be understood to include other types of entities, including by way of example, hardware and/or software entities, such as robots and other automated entities configured with machine learning functionality or other forms of artificial intelligence. A given such user illustratively has at least one communication address (e.g., IP address, email address, mobile telephone number, etc.) or other contact information recorded with the central authority 2002 in association with the user account. Such a user may but need not have previously gone through a registration process with the central authority 2002. Instead, by way of example, contact information for that user may have been provided by another user that has gone through a registration process. Other techniques can be used to register users as that term is broadly used herein.
One or more of the sensor networks 2005 may each comprise a plurality of sensor devices deployed within an interior space and/or an exterior space of building or other structure associated with a residence or business. Such sensor devices are considered examples of what are more generally referred to herein as “nodes” of a set of nodes. The “nodes” as that term is broadly used herein can include smart nodes with limited user interfaces, smart nodes with more complex user interfaces, smart nodes with no user interfaces, and other types of nodes, as well as various combinations thereof. Numerous other arrangements are possible.
As indicated previously, illustrative embodiments advantageously provide enhanced privacy in these and other situations involving third party service providers. It is to be appreciated, however, that other types of sensor systems, and more generally other sets of nodes, can be utilized in other embodiments. Accordingly, embodiments disclosed herein should not be viewed as being limited in any way to use with particular types of sensor systems.
The central authority 2002 in the present embodiment comprises a plurality of backend servers 2010 that implement protocols 2012 for authentication, verification and other security-related functions within the system 2000. For example, the protocols 2012 include one or more authentication and/or verification protocols carried out in conjunction with privacy enhancement using derived data disclosure as described herein.
The backend servers 2010 and their associated protocols 2012 of the central authority 2002 are illustratively implemented at least in part in the form of software running on one or more processing devices, each of which includes a processor 2015 coupled to a memory 2016 and a network interface (“Nwk I/F”) 2018. Example implementations of such components are described in more detail below.
The system 2000 further comprises third party application platforms 2020 coupled to the one or more communication networks 2004. Each such third party application platform comprises one or more third party applications 2022. Although shown as separate from the central authority 2002 in the figure, at least a subset of the third party application platforms 2020 and their respective third party applications 2022 may be encompassed by the central authority 2002. For example, such platforms and applications can be provided in some embodiments using a subset of the backend servers 2010 of the central authority 2002. Moreover, at least portions of one or more of the third party applications 2022 can execute at least in part on one of the user devices 2006, or on other system processing devices.
As indicated previously, terms such as “third party application” and “app” as used herein are intended to be broadly construed, and in some embodiments can include, for example, an application that includes multiple distinct software components operating on different processing devices or processing platforms.
A given one of the third party applications 2022 can comprise, for example, one or more software programs of a third party service provider. Such a third party application illustratively executes in a trusted processing environment that is under the control of a trusted party such as central authority 2002. Accordingly, one or more of the third party application platforms 2020 may be under the control of the central authority 2002, although such third party application platforms are shown as separate from the central authority 2002 in the figure. It is also possible that a third party application may execute in a non-trusted processing environment that is not under the control of the trusted party. Thus, one or more of the third party application platforms 2020 may not be under the control of the central authority 2002. Various combinations of trusted and non-trusted processing environments may be used to implement different ones of the third party application platforms 2020 in illustrative embodiments.
In operation, at least one processing device of the system 2000, such as a processing device that implements a given one of the backend servers 2010 of the central authority 2002, is configured to receive sensor data from one or more sensor devices of at least one of the sensor networks 2005. As indicated previously, such sensor devices are examples of what are more generally referred to herein as “nodes” and the corresponding sensor networks are examples of what are more generally referred to herein as “node networks.” It is therefore to be appreciated that other types of nodes and node networks can be used in other embodiments.
The above-noted processing device is further configured to compute predicate data based at least in part on the received sensor data, to transmit at least a portion of the received sensor data and the computed predicate data to one of the third party applications 2022, to receive additional data from the third party application 2022 responsive to the transmitted portion of the received sensor data and the computed predicate data, to generate a control signal based at least in part on the received sensor data, the computed predicate data, and the received additional data from the third party application, and to transmit the control signal to at least one of the sensor networks 2005, such as to one of the sensor devices of the sensor network or to another type of node of another type of node network of the system 2000. It is also possible that a control signal may additionally or alternatively be transmitted to one of the user devices 2006.
A given “control signal” as the term is broadly used herein can comprise, for example, one or more commands, such as, for example, a single command directed to a particular node, multiple commands directed to a particular node, or different sets of one or more commands directed to respective nodes, in one or more node networks.
In some embodiments, the central authority 2002 comprises or is otherwise associated with a trusted party responsible for configuration and management of one or more of the sensor networks 2005 or other node networks of the system 2000. The above-noted processing device illustratively implements at least a portion of the central authority 2002, such as one or more of the backend servers 2010. Multiple processing devices can also be used, in place of a single processing device.
Such a processing device or set of processing devices in computing predicate data based at least in part on the received sensor data illustratively computes the predicate data in accordance with one or more user-specified policies relating to access by the third party application to information including or derived from the sensor data. Examples of such policies are provided elsewhere herein. The computed predicate data illustratively comprises only information that a user has permitted the third party application to access.
Additionally or alternatively, computing predicate data based at least in part on the received sensor data illustratively comprises associating the computed predicate data with a pseudonym that prevents the third party application from determining an identifier of a source of the corresponding sensor data.
The computation of the predicate data in some embodiments further involves computing the predicate data in accordance with one or more compliance policies, such as policies designed to protect PII of system users. These include data privacy policies such as the General Data Protection Regulation (GDPR) of the European Union, as well as a wide variety of other types of governmental policies.
In some embodiments, computing predicate data based at least in part on the received sensor data more particularly comprises computing, from received sensor data that includes one or more explicit identifiers, anonymized predicate data that does not include the one or more explicit identifiers. The anonymized predicate data in such embodiments may be indicative of at least one of an activity associated with a user within an area in which the sensor data was collected, and a classification of the user within the area.
As noted above, the central authority 2002 in some embodiments implements a trusted processing environment comprising one or more of the third party application platforms 2020 configured to execute one or more of the third party applications 2022. A given such trusted processing environment is illustratively configured to obtain sensitive data, and to allow third party applications created by third party service providers to execute in the trusted processing environment. Inputs are illustratively provided to the third party applications in the form of predicates of the type described herein, with the predicates being computed as functions of the sensitive data, and with the predicates illustratively corresponding to information for which the user has expressly granted access to the third party service provider.
A trusted processing environment in some embodiments is additionally or alternatively configured to enforce best practices and industry standards, such as making sure that the computation meets government compliance.
As a more particular example, a trusted processing environment provided by a trusted party such as central authority 2002 in illustrative embodiments herein can be implemented in the form of a scalable cloud environment utilizing virtual machines that can access specified APIs according to their approved permissions. For example, the trusted party may be configured to define the APIs, to determine what virtual machines to create and what third party application software to run in them, and to control how the virtual machines access data and send information to outside entities. The trusted party can run multiple instances of a virtual machine with the same content, and provide it with different inputs, some of which may not be real sensor data, but are instead synthetically generated for testing of the system. The trusted party can also determine whether the third party application software running in a given instance of the virtual machine complies with a set of rules.
In some embodiments, the third party applications can execute in whole or in part outside of a trusted processing environment. For non-trusted processing environments of this type, the predicates are illustratively associated with a pseudonym that may change over time. The third party service provider does not know the mapping from the pseudonym to an identifier of the source of the sensitive data.
Additional illustrative embodiments will now be described in further detail with reference to FIGS. 21 and 22. These embodiments are illustratively configured to implement various aspects of device location services. Such embodiments in some implementations utilize one or more aspects of privacy enhancement using derived data disclosure as described elsewhere herein, and can additionally or alternatively implement other types of complementary functionality, such as functionality relating to determining preferences and events and generating associated outreach therefrom.
Many services in a wide array of different contexts can benefit from knowing the approximate location of a user. For example, it is beneficial for emergency response services receiving an emergency call from a home to quickly be able to locate the person in distress. In case of a fire, this may not be straightforward, and knowing the likely location ahead of time may be the difference between life and death.
In some embodiments, users may benefit from location services on a daily basis, e.g., when a residential IoT system uses radio signals associated with a user to automatically turn on lights in rooms and corridors the user is about to enter, and turn off lights in areas that the user has left. As another example, by determining time series of device locations within a residence or commercial space, it is possible to determine how rooms are connected with each other, what areas are highest use, and to determine, based on usage information (e.g., based on energy consumption patterns) what rooms have what purposes. Such a graph can be used to identify what rooms are entrance/exit rooms, for example. If ever a user is detected (whether based on radio signals or sonar) in a room that is not an entrance/exit, but there was no activity in adjacent rooms prior to the detection, then this is a clear sign of risk, as it is indicative of a burglar. This is a type of detection mechanism that does not rely on detection of breaking glass, and therefore, enables the detection of somebody climbing in through an open window, floor, or ceiling.
There are many more benefits associated with determining the location of a user and detecting when the same user moves to another adjacent space. However, existing technology is not well suited to support this type of functionality. For example, consider a wearable device using BLE, where the device has been paired with a phone (that may not always be in the presence of the wearable device) and then placed in “non-discoverable” mode. Bluetooth equipped sensors in a space would not be able to detect the wearable device, and would also not be able to track it as it moves around. Typical Bluetooth devices operating in slave mode, as this example device would, would also only be able to be associated with one master at a time, which would typically be the phone. This would prevent Bluetooth nodes in the environment from pairing with the wearable device, as doing so would cause the phone to no longer be paired. Moreover, if one such Bluetooth node were to pair with the wearable device, this would automatically prevent another Bluetooth node in the environment from performing the pairing. Therefore, this is a practical constraint that prevents detection and tracking of this device.
Illustrative embodiments disclosed herein provide a set of methods to overcome these and other related problems, allowing an infrastructure based system to track the movement of users carrying devices with radios, in spite of existing technology constraints and usage limitations. Solving this problem has significant benefits as it relates to safety (e.g., location by emergency response), security (e.g., detection of previously undetectable home intrusion attempts), energy savings (e.g., automated light controls), convenience (e.g., automated inference of likely intentions, based on an understanding of the context and historical usage data) and personalization (e.g., the distinction of different users and application of preferred contextual configurations.) As will be appreciated by a person skilled in the art, there are many other benefits and associated services that are enabled by illustrative embodiments disclosed herein, which have not heretofore been possible.
Some embodiments are configured to identify and track devices with wireless radios, such as Bluetooth radios and WiFi radios. Many of the techniques also apply to other types of radios, such as NFC and ZigBee radios, as the disclosed approaches relate to common features of radio communication. For different radio protocols, the exact constraints differ, and therefore, the exact configurations of the tracking methods also differ. The most constrained setting is that of Bluetooth, including BLE, since this family of protocols (containing regular Bluetooth, BLE, and different versions of these) include features to defend users against abusive tracking. We describe a set of methods that enable a user, wishing to do so, to allow tracking of his or her device by a designated system, such as their home IoT infrastructure, their office IoT infrastructure, or a commercial communication infrastructure such as what may be deployed in a movie theatre complex, a mall or a hospital.
Since other communication standards, such as WiFi, are less constrained, not all of the techniques described in the context of Bluetooth are necessary for a tracking solution to work for less constrained technologies. While there are differences between Bluetooth and BLE, such as Bluetooth using 79 channels and BLE using only 37, there are also significant similarities. Some embodiments herein provide approaches that apply to both Bluetooth and BLE. Thus, a person of skill in the art will recognize that the disclosed techniques apply to both Bluetooth and BLE with minor modifications such as the number of channels. Moreover, a person of skill in the art will recognize that several of the embodiments also apply to wireless technologies that do not use multiple bands.
Some embodiments also address the fact that Bluetooth radios can be either slaves or masters, or have both of these roles. This impacts the manner in which the devices communicate, and accordingly, the manner in which they can be tracked. Moreover, it affects the constraints associated with the technologies. For example, typical Bluetooth master devices can communicate with up to eight slave devices, although there is no theoretical limit. However, Bluetooth slaves can only communicate with one device: the associated master. If a person wishes for his or her device to communicate with another device, then using traditional Bluetooth technology, the slave device needs to be paired with this other device, which will cause it no longer to be paired with its old master. This may be disruptive to the user, e.g., in the case of a wearable device that is a slave and which transmits health-related data to the user's phone, which acts as its associated master device. By breaking the bonding between these two devices in order to track the wearable device, the wearable device no longer will deliver data to the phone, which would be undesirable to the user. This and other problems are overcome by the solutions we disclose in illustrative embodiments herein.
One aspect of the disclosed technology is a method to determine the hopping sequence of a device to be tracked. If this device is acting as a master, then the hopping sequence is determined by the Bluetooth Device Address (BD_ADDR) of the device itself, along with the clock of this device. If the device to be tracked is acting as a slave, then its hopping sequence is determined by the BD_ADDR and clock of the master it is associated with. If a slave is part of multiple piconets, it can have multiple masters, and accordingly, be associated with multiple hopping sequences. A given “piconet” as that term is broadly used herein refers to a type of ad hoc wireless network that links multiple wireless devices using a protocol such as Bluetooth. For example, in the Bluetooth context, a piconet comprises two or more devices occupying the same physical channel, including a master device and one or more slave devices. Other types of ad hoc wireless networks can be used in other embodiments.
Bluetooth uses techniques commonly referred to as frequency-hopping spread spectrum (FHSS) or Adaptive Frequency Hopping (AFH) in order to avoid interference problems. In accordance with such techniques, transmission of radio signals involves rapidly changing the carrier frequency among many distinct frequencies occupying a large spectral band. The changes are controlled by a code known to both transmitter and receiver. For example, in conjunction with forming a communication link between two devices, the devices are illustratively synchronized to change channels together many times a second (e.g., 1,600 times per second). The pattern of channels used is called the “hopping sequence” and is unique for each link. Bluetooth implements frequency hopping in the Industrial, Scientific and Medical (ISM) 2.4 GHz band, from 2400 to 2483.5 MHz, using 79 radio frequency channels in this band, starting at 2402 MHz and continuing every 1 MHz. Other types of frequency hopping techniques involving other bands and hopping sequences can be used in other embodiments.
In some embodiments, it is beneficial to determine the hopping sequence of a device to be tracked, as that allows for communications to or from that device to be effectively intercepted by network nodes. The signal strength of the intercepted signals determines the approximate location of the device to be tracked. Where the hopping sequence is not known, though, the device can still be tracked, using alternative approaches described below.
In Bluetooth, the master communicates on even numbered slots and slaves on odd numbered slots. To avoid collisions, slaves can only respond if requested to by the master. By listening in to the communication, a node can determine from the traffic in the even numbered slots what is being sent by the master. Additionally, the headers of these packets will contain an identifier associated with the master. The odd-numbered slots will have traffic from slaves, and the headers will contain an identifier associated with the master, and a number that corresponds to what number the master has assigned the slave that is communicating. While that can be altered, e.g., by the master parking a slave to engage another slave, the number identifying the slave is mostly static. Thus, by identifying from the headers of the slave-originated traffic what number is used, the slave can be identified with a high accuracy. The master-originated packets, similarly, will contain a number that identifies what slave the packet is for. Most packets sent by a master to a slave will only be sent if the master and the slave are in the same area, and sent as a result of a back-and-forth communication. By keeping track of the BD_ADDR, which identifies the master, the assigned number described above, which identifies the slave with high accuracy, and other header information (such as header information associated with a scan performed by a device), the system can determine the approximate location with reasonable accuracy, using a multiplicity of nodes that have the information required to perform this tracking. In addition, the system can determine direction and distance, e.g., using triangulation for a collection of signals and their associated signal strength.
In one embodiment, the system initiates pairing with a user device such as a phone or a tablet. The system is represented by one node of its network, which is within communication distance with the user device, and initiates pairing with the node in the role of master device, and the user device in the role of slave device. This step utilizes standard Bluetooth pairing. Then, the node in the master role issues a request to the user device to reverse the roles, such that the node becomes the slave, and the user device becomes the master. This is also utilizing the standard process for reversing roles, which is part of the Bluetooth pairing protocol. In conjunction with reversal of the roles, the user device sends its BD_ADDR and an indication of its time (e.g., the clock status of the user device) to the node, and the two devices generate a hopping sequence from these two information elements. In addition, the two devices agree on a code. There are multiple ways of doing that. The BD_ADDR and the code are used to compute a link key, which is later used for encrypting the data portion of packets between the two entities.
Alternatively, the node pairs with the user device with the node starting the pairing in slave mode, with no need to switch roles. In this case, the user device sends its BD_ADDR and clock data to the node, where this is later used to determine the hopping sequence. The devices also establish a code in one of several available ways and generate a link key from the master's BD_ADDR and the code. This node-starting-as-slave mode is also utilizing standard Bluetooth pairing methods.
As part of the disclosed technology, the node will now know the hopping sequence for communication between the node and the device it paired with.
The ability for a node to follow along the hopping sequence to track a device acting as a master or a device acting as a slave is beneficial. However, often communication is protected by link encryption. By being able to obtain plaintext payloads, i.e., the messages of the packets, the system can perform additional analysis. The link key used in a piconet is determined from the BD_ADDR, the clock of the master, and additional information exchanged between the devices during a setup phase. In one embodiment of a system as disclosed herein, a node is pairing with a device to be tracked, and is thereby obtaining information allowing it to establish the hopping sequence. In addition, this node also establishes a pairing key with the device to be tracked. This can be done using the standard Bluetooth protocol. After this phase has completed, the node will have a copy of the link key.
In some implementations of Bluetooth, this will be stored in /var/lib/Bluetooth/[BD_ADDR]/linkkeys, and can be accessed by the node from this directory. Additional information is available in other directories in some implementations. For example, HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Services\BTHPORT\Parameters\Key contains MAC addresses (e.g., BD_ADDR) of all Bluetooth adapters, and under each adapter, all the paired devices' MAC addresses with their link keys. On some implementations of Bluetooth, other storage hierarchies may be used.
As a node in an IoT network extracts information relating to hopping sequences and/or keys from its storage system, it transmits this to other nodes in the network, which then store the information in their corresponding directories, thereby effectively cloning the first node and enabling the other nodes to listen for traffic on the related piconets. As multiple nodes identify such traffic, and preferably the signal strength of the traffic, when available, these nodes can determine the approximate location of devices. The system can choose to exclude (e.g., ignore) devices that are not mobile, and which may not be interesting to track. It can also perform tracking of such devices much less frequently, such as once a day instead of once a second. This enables the IoT infrastructure to detect the movement of devices that are not typically used as mobile devices, such as, for example, a Bluetooth enabled refrigerator or a desktop computer with a Bluetooth radio.
While traditional Bluetooth pairs devices according to the specifications of Bluetooth (e.g., Bluetooth Core Specification v5.1) there is no support for enabling devices other than the master or a slave to listen in to traffic. In particular, there is no support for enabling a multiplicity on nodes, such as IoT nodes, to identify this traffic without each one of these having been paired with the master. Accordingly, a second aspect of the disclosed technology is a method to enable this functionality. This is done by determining information useful for tracking at a first node, and then replicating this information to a multiplicity of nodes. For passive listening in to signals, that allows the multiplicity of nodes to emulate the first node in that they can all track the device in question. In order to allow active interaction, additional mechanisms, described below, are preferably used.
In contexts where no encryption is used or where the system has established the link key of a piconet for which it is tracking one or more devices, there are advantages to an approach that allows the system to actively probe for a device. It can do so by impersonating a master, sending a packet to one or more slave nodes in the piconet, causing such slave nodes to respond. A variety of messages can be used; preferably, the system would not send any message that changes the state of the recipient device beyond simply responding to the message. Sending a ping message that requires a response is helpful. By doing this, the system causes slave nodes in their surroundings to respond, effectively causing them to reveal their approximate location. If more than one out of a multiplicity of nodes is to perform this type of action, they should preferably coordinate with each other to avoid interference and to promote triangulation efforts. As they transmit a packet, they transmit it during a slot that is used by the master of the piconet, on a hopping sequence associated with the piconet. To prompt the response by a node acting as a master, the system can transmit a packet, on behalf of an observed slave node. However, since slave nodes do not typically initiate communication with a master, this has to be done in a way that mimics legitimate Bluetooth behavior, e.g., transmitting a packet announcing a node's presence (when this node is the impersonated node).
Instead of impersonating devices on the network, though, it is beneficial to impersonate the node that first paired with the master, when applicable. To the extent that the node wishing to generate the ping message to the master device is the same as the node that performed the pairing, this would not be an actual impersonation. In the case where it is another node on the same IoT network, it is more a matter of representing the original node, using the keys and other state data needed to communicate. None of these approaches cause interference on the network beyond what standard Bluetooth communication on another piconet would have. Preferably, as a user configures an IoT system, the user would grant permission for the system to attempt to impersonate nodes and/or devices to devices in its area, along with an assurance from the user that he or she has the right to grant this permission.
In one embodiment, instead of pairing a node of the IoT network to a device in the traditional way, the node observes traffic from the device to be tracked and infers the portion of the BD_ADDR of the device that is transmitted in plaintext. The node can then brute-force the keyspace of available remaining information to determine the hopping sequence of the device to be tracked. The approach of attempting to infer the BD_ADDR of a master of a piconet, and to utilize an inferred BD_ADDR to determine hopping sequence information, can be used as an alternative to pairing between one or more nodes of an IoT system and one or more devices to be tracked. Preferably, as a user configures an IoT system, the user would grant permission for the system to attempt to determine information related to devices in its area, along with an assurance from the user that he or she has the right to grant this permission.
A master in a Bluetooth piconet can park a slave, thereby moving it from an active engagement mode to a passive mode. In one embodiment, the system acts as a multiplicity of cloned slave nodes paired to a master and refuses to acknowledge the parking of it. If this request is not acknowledged, in some implementations, it is not considered completed, thereby maintaining an active role for the slave associated with the system, which in turn permits the system to communicate with the master. A person skilled in the art will recognize, though, that independently of whether the nodes of the IoT system are parked or not, they will be able to follow along the hopping sequence and thereby track devices as described above.
In another embodiment, the system tracks devices not by using their associated MAC addresses/BD_ADDR, hopping sequences, or the like, but rather, by observing the friendly names transmitted over the available channels, and in particular channels used to search for devices to connect to. For example, a user may have configured a phone to have a friendly name of “Alice's iPhone,” and a tablet to transmit the friendly name “Bob's Samsung pad.” Whereas these are not unique identifiers, and it is in theory perfectly possible for a user to change the friendly names over time or even have multiple devices with the same friendly name, neither of these are very common uses. Therefore, tracking is not guaranteed to be correct, but will be correct in a large number of cases. For example, in some embodiments disclosed herein, a layout of a space is determined and a graph is generated to reflect how various nodes are located relative to each other; whether there are likely to be doors in between some nodes; and so on. This determination uses statistic and historic data of movement patterns, and is enabled even if some devices were to change identifiers periodically. A person of skill in the art will recognize that this method is not limited to Bluetooth devices.
Another type of information useful for tracking is information indicating a device type or model, a MAC address that can be mapped to device type and model or simply used by itself as an identifier, and additional identifying information collected from headers, relating to the communication protocols, clock data, user agent information, etc. This, by itself or in combination with other data useful for tracking, can be used to identify a device and perform tracking, or to generate a log of use for purposes of audits, usage statistics, time card information, home invasion evidence, and other purposes.
To the extent that two radios, such as one Bluetooth radio and one WiFi radio, and their associated identifiers, appear to be co-located for several observations, one possibility is that these two radios are associated with one and the same device; this is then also of importance for the tracking. Another alternative comprises two radios that are not likely to correspond to one and the same device, e.g., by being of the same type (both Bluetooth, for example) or by being associated with headers indicating different devices; or by not always being co-located. This is also an important system observation as it permits an automated attribution of a device to a likely user. For example, a phone that is commonly co-located with a piece of wearable technology such as an Oura ring, can be determined to likely correspond to the same end user. Therefore, if there is a profile established for the identifiers associated with the phone, the same profile can be associated with the piece of wearable technology such as the Oura ring.
In cases where a mistaken association is made, or a device changes owners, the system will be able to identify new co-locations that contradict previous co-locations in terms of the associated profiles and resolve the profile associations accordingly. To the extent that this re-association cannot be done in an automatic manner, the system can generate a request to an admin or user asking for a user-driven classification. This can be done for all observed devices or only those that are associated with a registered user in the system. If it is done relative to other users, this sometimes is practical, as the registered users can assign names with observed users and generate customized logs identifying when various users enter and leave the system. It can also permit the generation of alerts, such as a notification sent to a user if he or she appears to have left home without his or her phone, wallet (containing a wireless radio, e.g., in a BLE enabled credit card), wearable computer and/or tablet computer, where this notification may be sent as a push notification to an application, an SMS or other text message, or an email. In one embodiment, notifications are sent to an automotive application associated with the user.
The system preferably maintains one or more profiles of the devices it is tracking on individual IoT nodes or accessible by these, as well as in a central location. The storage also preferably maintains additional information associated with the identifiers. A record may be generated and associated with one or more identifiers (such as Bluetooth identifiers used for tracking, WiFi identifiers used for tracking, etc.); and also information related to the user(s) of these devices, where such information may be pointers to profiles containing user preferences (such as what lights to turn on and off when the user is detected, security profiles relating to access control and alarm systems) and additional user information.
The system can compute predicates such as pseudonyms (e.g., unique or semi-unique identifiers that are not linked to the individual user in a way that can be used for identification, and which may change over time); demographic information (such as gender); and information used for notifications, alerts, and the like. Exemplary uses of these and other types of information are described elsewhere herein.
In one embodiment, the system actively identifies wireless devices that are new to the system and saves these, as well as previously identified devices. If, for example, a user were to enter a house when its alarm is armed, it is of relevance whether the user's devices had been seen before by the system. If the user is a resident or otherwise authorized to enter the house, according to some policy, then the detection of the user's device can be used to initiate a disarming action, e.g., by sending a notification to the user's phone, allowing the user to prove his identity by inputting a credential, providing a biometric or answering a challenge question; and then cause the alarm system to be disarmed. In contrast, if a non-authorized user device is recognized, this information is potential evidence, should the entry be unauthorized. A legitimate user can be notified of the entry and optionally be informed of the device identifier or associated information, should it be available.
In one embodiment, one or more devices associated with detectable radios are registered by the system. This is done by detecting the devices, as described above, and the automated generation of one or more profiles that are stored by the system. Devices can be associated with a user, e.g., by being associated with an application running on the device, on which the user provides at least some information that associates the device with a user profile (as opposed to a device profile). The user profile, in one example embodiment, contains a portion of user-provided data, such as name, a nickname, a credit card associated with service provision, and one or more addresses associated with the user. User profiles can also comprise health data, obtained from the system sensors; preferences, whether user-stated or inferred; usage statistics; and additional or alternative information.
In some instances, a device is not associated with a user profile, but with a space. For example, a router has a radio, and is associated with a home or business, but not with a specific user in the home or business. Similarly, a smart refrigerator may be used by multiple users in a space, and therefore, is associated with the space rather than with a user. However, information related to users accessing the refrigerator may be stored in a profile, whether associated with the user profile, with the device profile, both, or in a separate log. A user can access a list of device descriptors that identify the types of the devices, which the system can classify based on their IP address, their energy consumption profiles, their usage statistics (e.g., a fan is more likely to be used when it is warm than when it is not), and user-provided classifications. A user can attach digital “leashes” to registered devices, and select associated actions. Example actions include notifying the user if the leashed device is no longer detected (whether it is physically removed or its radio is otherwise not detectable); if the leashed device is moved (e.g., is detected by another set of nodes than those that normally detect it); if the leashed device appears on another system controlled by the same provider; and/or if the leashed device changes its operational profile, e.g., appears to malfunction.
The system can also help locate leashed devices, e.g., by a user selecting that he or she wishes to locate the leashed device, and the system in such situations can help the user find the leashed device by signaling its location. One way of signaling the location is to provide the user with a video feed or a photograph from a node in the direct neighborhood of the leashed device; another is to generate an audio signal from the closest IoT node, and/or to generate audio signals whose strength increases the closer the nodes emitting the signals are to the leashed device, where the system determines the distance by detecting signal strength from the leashed device, as observed by the various IoT nodes. In cases where a leashed device is stolen, the system can help identify the theft and notify the user, as well as helping to locate the stolen leashed device by providing a trail of locations associated with other IoT networks operated by the same provider, where these other networks detect the same leashed device.
In the case where the device identification is heuristic, e.g., based on a constellation of detected parameters as opposed to the detection of a unique identifier, the system can provide an indication of certainty of the detection, based on the relative likelihood of a given information configuration, which can be estimated based on system-observed distributions of heuristic identifier components. In some embodiments, tracking data is provided only to authorized users, where law enforcement may be authorized, for example, to track stolen goods, while end users may not; while tracking on the premises (e.g., locating a device within a property) is possible only to users associated with the same premises, and not to law enforcement. In one embodiment, a user can only track a leashed device if he or she is already associated with the device, e.g., is the primary user of the leashed device. The system enables these and similar policies, as will be appreciated by a person of skill in the art.
The disclosed techniques allow an association of users and devices. One way of doing that is by users “claiming” devices by installing apps on them and registering, or by providing a user name and access credential which registers the device to the user by associating it to a known user profile. The system also enables inferred associations between users and devices. For example, while the system identifies devices based on electromagnetic wave radio characteristics and data, the identification of persons may be a compounding of RF device data with acoustic information such as voice recognition, video or camera gait and/or face recognition, biometrics, ultrasonic spatial sensing, optical sensing and additional or alternative information, as will be appreciated by a person skilled in the art.
In one embodiment, a first node pairs with a device, with the first node acting as a slave, thereby joining the piconet of the device, which acts as a master. Current Bluetooth (and BLE) only permits each device acting as a master to operate one piconet. This piconet is associated with a hopping sequence that is a function of the BD_ADDR of the master, and the clock of the master. During the pairing, the slave is provided with information regarding the BD_ADDR and the clock of the master, allowing it to follow the same hopping sequence. Messages are changed in the bands associated with the hopping sequence, where every other slot (e.g., each even numbered slot) is reserved for the master and remaining slots (e.g., odd numbered slots) are reserved for at least one slave that has been pinged by the master.
As the first node becomes part of the piconet, it will be able to access all traffic on the piconet, in principle, and (for legacy Bluetooth devices) all traffic that is intended for the first node or a party with its information (e.g., slave identifier and knowledge of the hopping sequence). By transmitting information useful to determine the hopping sequence to a second node, the second node can act as the first node and follow the hopping sequence, obtaining traffic from the master intended for the first node. The slave preferably also knows its slave identifier. However, as this information is stored outside of the radio, in a space that is not authenticated, the second device can use different slave identifiers along with the BD_ADDR and clock of the master, thereby obtaining traffic sent to any node on the piconet. Similarly, a slave can store information indicating that it is the master, thereby also being able to output communicated data from a slave to the master.
A node with a modified Bluetooth implementation that outputs any traffic on the piconet does not need to switch roles in this manner in order to be able to access traffic. Nodes do not have to access the traffic in order to determine the proximity of devices with radios. A given such node may instead simply determine that it receives traffic for several consecutive slots along a hopping sequence, indicating that the channel is busy. It may be more beneficial, though, in some embodiments to be able to identify the transmitting node by its slave number or its identification as being the master (implicitly done by communication on the slots reserved for the master.) Slaves can initiate responses from their master by responding to messages requesting notification of slaves on the piconet; these are typically sent in a periodic manner by the master to determine what services are available, among other reasons.
In another embodiment, a first node pairs with a device, the first device acting as a master. The first node can then ping the slave (or any slave for that matter) on the piconet determined by the master's BD_ADDR and clock. This information determines the hopping sequence. Similar to the embodiment described above, the BD_ADDR and clock of the first node are shared with a second node, which can then mimic the first node and which is also able to receive traffic sent from a slave on the piconet. In other to avoid interference between co-located nodes using the same master BD_ADDR, the nodes can use a round-robin scheme or other scheduling to determine what node can communicate as the master, where the communication could involve a wake-up request for any slave on the piconet or other communication requiring a response from slaves. Thus, as responses are received by nodes (whether they transmitted the request or a peer node did), the receiving nodes will record the communication. A master device will always output identifiers of the slaves believed to currently be part of the network.
In yet another embodiment, a first node pairs with a device acting as a master to learn its associated piconet, and thereby becomes enabled to look for traffic on this piconet. For example, the first node, or another node given the information of the first node, can transmit a signal, posing as the master, requesting what slaves are present, associated with the piconet. The node can emulate multiple masters, one by one, scanning through a server of piconets to identify the presence of slaves by requesting responses. The node can, alternatively, scan multiple piconets simply listening for traffic, whether from a master or a slave, and output information regarding the traffic it observes.
Whereas some of the actions can be taken using legacy Bluetooth implementations, some of these approaches require modifications as will be understood by a person of skill in the art. For example, posing as another master is not supported by traditional Bluetooth implementations. To avoid potential interference that this may cause, it is beneficial for such “impostors” to first listen in to the channel to make sure that the probabilities for interference are minimized. This can be done by only acting as an imposter on a channel that has been silent for a threshold amount of time, with interference from other piconet traffic not being counted as activity on the channel in question.
In one embodiment, information about the detected node is output. Additionally, signal strength information is also output, allowing a triangulation by a party with access to multiple outputs. This party may be one of the nodes on a network or a central authority that receives multiple signals from one or more networks and identifies movement patterns.
The system can operate in several different modes, and can shift between these modes and variations of these modes and combinations of these modes. In a first mode, the system passively scans for the presences of recognized devices. One approach for doing this is for each node, when its radio is not busy for other purposes, to scan between the bands of each of the piconets that it is tracking. It can determine, for each time unit, one band that would be used by one of its piconets, if there were traffic on that piconet; if there is traffic, it determines from the headers what device is transmitting, and to what other device. This determines that the transmitting device is present, and makes it plausible that the receiving device is, too. To verify whether the latter is true, the node can determine on what band the receiving device would respond, if present, and listen to that band. Once one or more devices have been identified, nearby nodes can be notified, so that those nodes can scan for these devices to rapidly detect movement. On the other hand, far-away nodes can exclude these devices from the scan; if these devices are the only known devices on the associated piconet, this piconet's hopping sequence can be ignored for a time by far-away nodes. Once a node has determined the presence of one or mode devices, it is beneficial to search for other devices.
In a second mode, nodes actively scan for devices. This can be done by sending out requests, as described above, and awaiting responses. This can be done to locate devices whose locations are not known; to confirm the believed location of devices; to track down a device after a request has been made by a user to determine its location; and/or for other purposes.
Another mode of operation is to scan for devices in response to an environmental change, such as a movement identified by sonar and/or another type of sensor. Responsive to such a change, the system actively searches for radio activity in an attempt to assign the detected motion (or other environmental change) to a user, based on the user's association with the radio. Either the passive or active mode described above can be used in this context; alternatively, the nodes in proximity of the environmental change may scan for unknown devices, e.g., not with a focus on hopping sequences of known piconets. It is beneficial to use bands that are used by devices searching for other devices. It is also beneficial to scan arbitrary bands, e.g., pseudo-randomly chosen, or only certain bands (e.g., band number 5) for a duration of time, to give two concrete examples. Radio activity is detected as the nodes perform such operations. This may correspond to a known device (e.g., identified by matching the identifiers to stored identifiers, corresponding to a band of a known hopping sequence); an unknown device searching for other devices (e.g., using one or more bands associated with discovery); an unknown device communicating (e.g., this can be detected by identifying the presence of traffic, and optionally, extract identifiers such as device names or manufacturer-associated characteristics); or a known device not communicating (e.g., by transmitting a request for response).
In some embodiments, a first node of a node network determines identifying information sufficient for tracking a user device that generates radio communications detectable by at least the first node. The first node replicates at least a portion of the identifying information from the first node to one or more additional nodes of the node network. The first and additional nodes track the user device utilizing their respective portions of the identifying information. In the case of the first node, this includes the identifying information that it determined, and in the case of the one or more additional nodes, this includes at least portions of that identifying information that were replicated by the first node to each of the one or more additional nodes. Examples of first and additional nodes and a user device are shown in FIGS. 21 and 22, and described in more detail below.
The term “identifying information sufficient for tracking a user device” as used herein is intended to be broadly construed, and may include, for example, information sufficient to determine a hopping sequence of the user device, such as at least an identifier of the user device and clock information of the user device, from which a hopping sequence may be illustratively determined in some embodiments, such as those in which the user device is operating as a master device relative to one or more other slave devices. A given such slave device can include the first node, or another device that is paired with the user device.
In some embodiments, the portion of the identifying information replicated from the first node to one or more additional nodes of the node network includes all of the identifying information, which may include, as indicated above, at least an identifier of the user device and clock information of the user device, from which a hopping sequence may be determined. For example, as described above, a hopping sequence in a Bluetooth embodiment can be determined from the BD_ADDR of a master device and a clock value, clock status or other clock information of the master device. Such a determination is illustratively performed utilizing well-known features of the Bluetooth protocol, as will be readily apparent to those skilled in the art. Similar techniques can be used in other types of wireless networks to determine hopping sequences of particular user devices.
The replicated portion of the identifying information in some embodiments may alternatively comprise information specifying the hopping sequence used by the user device, as determined by the first node from the identifier of the user device and clock information of the user device. Other types of identifying information suitable for tracking the user device can be replicated from the first node to the one or more additional nodes, so as to allow the one or more additional nodes to track the user device in substantially the same manner that the first node is able to track the user device.
Each of the first and additional nodes, as well as the user device, illustratively comprises a separate processing device that includes a processor and a memory, as described in more detail elsewhere herein. For example, the nodes in some embodiments comprise respective IoT nodes of one or more node networks, although other types of nodes may be used, and such nodes may comprise or be otherwise associated with one or more sensor devices. Other arrangements of one or more processing devices may be used in implementing functionality for device location services in other embodiments.
In some embodiments, a device location service more particularly involves operations such as determining a location of the user device based at least in part on the above-described tracking, and providing at least one service based at least in part on the location of the user device. The provided service in some embodiments comprises a location service for simply locating the user device, but a wide variety of other types of location services can be provided based at least in part on a current location of the user device. The term “device location service” as used herein is therefore intended to be broadly construed, and can encompass any type of service that is provided in a system of the type described herein at least in part based on location of a user device, such as privacy-enhanced services provided using derived data disclosure as described elsewhere herein, or services that are provided at least in part by determining preferences and events and generating associated outreach therefrom.
In some embodiments, tracking the user device utilizing the identifying information in the first node and the one or more additional nodes comprises performing the tracking without requiring each of the first and additional nodes to pair with the user device. For example, the first node may pair with the user device in a master-slave pairing arrangement, or other type of pairing arrangement, while none of the one or more additional nodes pairs with the user device, but instead receive replicated identifying information from the first node, sufficient to allow each of the one or more additional nodes to track the user device, just as the first node could track the user device using the determined identifying information.
In determining identifying information sufficient for tracking the user device, the first node illustratively identifies the user device as having not been paired with any of the nodes of the node network, and extracts at least a portion of the identifying information sufficient for tracking the user device from the radio communications generated by the user device as part of its communication with another device with which it is paired. This is an example of an arrangement in which the first node does not itself pair with the user device, but instead determines the identifying information based on radio communications between the user device and another device, where such radio communications are detectable by the first node. The first node therefore need not be paired with or otherwise in communication with the user device in order to determine the identifying information sufficient for tracking the user device, although in some embodiments it may pair with the user device.
In some embodiments, in determining identifying information sufficient for tracking a user device, the first node extracts an identifier from the radio communications of the user device, extracts clock information from the radio communications of the user device, and determines a hopping sequence of the user device based at least in part on the master identifier and the clock information.
As indicated above, the first node and the user device in some embodiments are paired with one another in a master-slave pairing arrangement, with one of the first node and the user device operating as a master device and the other of the first node and the user device operating as a slave device. The extracted identifier in such an arrangement illustratively comprises one of a master device identifier and a slave device identifier.
The hopping sequence illustratively identifies a piconet or other particular ad hoc wireless network with which the user device is associated, and tracking the user device utilizing the identifying information in the first node and the one or more additional nodes illustratively comprises monitoring communications between the user device and one or more other devices of the particular ad hoc wireless network.
In replicating at least a portion of the identifying information from the first node to one or more additional nodes of the node network, the first node illustratively replicates at least information characterizing the hopping sequence from the first node to the one or more additional nodes, and tracking the user device comprises tracking the user device based at least in part on the hopping sequence. The information characterizing the hopping sequence can comprise, for example, the hopping sequence itself, or information that is utilizable to determine the hopping sequence, such as the identifier of the user device and clock information of the user device, as described previously.
In some embodiments, the first node enters a pairing arrangement with the user device, in which the first node operates as a master device and the user device operates as a slave device, and then subsequently requests the user device to reverse master-slave roles with the first node. In conjunction with the reversal of the master-slave roles, the first node receives from the user device a master device identifier of the user device and clock information of the user device, and determines a hopping sequence from the master device identifier and the clock information. As indicated above, the master device identifier in the Bluetooth context comprises the BD_ADDR of the user device, although other types of master device identifiers can be used in other embodiments.
In determining identifying information sufficient for tracking the user device, the first node in some embodiments illustratively generates, in conjunction with a pairing of the first node and the user device, a link key shared by the first node and the user device. The link key is replicated from the first node to the one or more additional nodes of the node network as at least a portion of the identifying information sufficient for tracking the user device.
In some embodiments, a given one of the first and additional nodes detects one or more radio communications generated by the user device, generates at least one signal strength measure associated with the one or more detected radio communications, and provides information characterizing the one or more detected radio communications and the at least one signal strength measure to at least one of (i) another one of the first and additional nodes, and (ii) a central authority, in conjunction with the tracking of the user device. The central authority illustratively comprises one or more computers, servers or other processing devices providing central storage and control functionality for the tracking of the user device as well as for implementation of one or more associated device location services. An example of such a central authority will be described below in conjunction with FIG. 22.
In some embodiments, at least one of the first and additional nodes associates at least one policy with the user device in at least one of the first and additional nodes, and controls generation of an output, in conjunction with the tracking of the user device, in accordance with the policy associated with the user device. Examples of such policies were described in more detail elsewhere herein.
It is to be appreciated that additional or alternative features can be present in other embodiments.
For example, in some embodiments disclosed herein, a user pairs a first device to a first node, and then the first device is automatically paired to peer nodes of the first node, without requiring any further user action. These and other embodiments disclosed herein are useful for home and office automation, including automated personalization and configuration, as well as in other applications.
Additionally or alternatively, some embodiments are configured such that a user pairs a first device to a second device, and pairs the first device to a first node. The second device is then automatically paired to the first node and to peer nodes of the first node, without requiring any further user action. Again, these and other embodiments are useful for home and office automation, including automated personalization and configuration, as well as in other applications.
In some embodiments, a system as disclosed herein is configured to identify the presence of a device that has not been paired with any of the nodes on a network, and to extract some identifying information based on the radio traffic of the device. This can also be useful for home and office automation, including automated detection and logging of potential intruders, as well as other applications.
In one or more of the above-noted embodiments, it may be the case that a given system has no control over the software, firmware or hardware of any of the devices, but has at least some control over the aspects of the nodes. Numerous other arrangements and associated operating scenarios are possible.
Referring now more specifically to FIGS. 21 and 22, illustrative embodiments of information processing systems comprising multiple nodes and other processing devices are shown. The particular arrangements of devices and their operation as illustrated in these figures should not be construed as limiting in any way.
FIG. 21 shows a first node 2101, a second node 2111 and a central storage and control 2121. First node 2101 comprises a radio module 2102, such as a Bluetooth, BLE, WiFi, NFC or ZigBee module, or a combination of these; profile storage 2103 wherein keys and identifiers are stored; and an extractor 2104 that reads information from profile storage 2103 and transmits it over connection 2110 to the second node 2111 or over connection 2120a to the central storage and control 2121. It also comprises an implanter 2105 that receives information from the second node 2111 or the central storage and control 2121 and implants this information in profile storage 2103. These and other references herein to “implanting” of information are intended to be broadly construed, so as to encompass, for example, persistent storage of the associated information in at least one non-volatile memory or other storage device of at least one processing device that implements the corresponding node or other system component. In addition, the first node 2101 comprises a controller 2106 that uses keys and profiles stored in profile storage 2103 to monitor traffic observable by the first node 2101, or uses keys and profiles stored in profile storage 2103 to cause first node 2101 to transmit traffic.
The second node 2111 comprises a radio module 2112, such as a Bluetooth, BLE, WiFi, NFC or ZigBee module, or a combination of these; profile storage 2113 wherein keys and identifiers are stored; and an extractor 2114 that reads information from profile storage 2113 and transmits it over connection 2110 to the first node 2101 or over connection 2120b to the central storage and control 2121. It also comprises an implanter 2115 that receives information from the first node 2101 or the central storage and control 2121 and implants this information in profile storage 2113. In addition, the second node 2111 comprises a controller 2116 that uses keys and profiles stored in profile storage 2113 to monitor traffic observable by the first node 2101, or uses keys and profiles stored in profile storage 2113 to cause second node 2111 to transmit traffic.
The figure also shows central storage and control 2121 that comprises profile storage 2123 wherein keys and identifiers are stored; an extractor 2124 that reads information from profile storage 2123 and transmits it over connection 2120a to the first node 2101 or over connection 2120b to the second node 2111. It also comprises an implanter 2125 that receives information from the first node 2101 or the second node 2111 and implants this information in profile storage 2123, and a controller 2126. The controller 2126 directs the operation of the central control and storage 2121, including its interactions with the first and second nodes 2101 and 2111 via the respective connections 2120a and 2120b.
Radio module 2102 operates using information stored in profile storage 2103, but does not need to know whether this data was originated at the first node 2101, second node 2111 or central storage and control 2121. Similarly, radio module 2112 operates using information stored in profile storage 2113, but does not need to know whether this data was originated at the first node 2101, second node 2111 or central storage and control 2121.
FIG. 22 shows a device 2201 with a radio module 2202 and profile storage 2203. Device 2201 is pairing with the first node 2101 over connection 2210a, resulting in the storage of information in profile storage 2103 of first node 2101 and profile storage 2203 of device 2201. The information typically corresponds to a BD_ADDR being stored in both profile storage 2103 and profile storage 2203, where the BD_ADDR corresponds to the master, which may be either first node 2101 or device 2201. In addition, an optional link key may be generated and stored, where this is derived using the BD_ADDR of the master as part of the input. In addition, information about the clock of the master is stored in at least the profile storage of the slave unit, which is the device 2201 if first node 2101 is the master, and otherwise the first node 2101. The hopping sequence is derived from the stored BD_ADDR and the clock of the master. At least part of the information stored in the profile storage 2103 of the first device 2101 is transmitted to the second node 2111, as described above in conjunction with FIG. 1.
In addition, if the first node 2101 is the master and the device 2201 is the slave, then information about the clock of first node 2101 is transmitted to the second node 2111. The second node 2111 uses the information in profile storage 2113 to generate a hopping sequence associated with the BD_ADDR of the pairing between first node 2101 and device 2201. In addition, the second node 2111 optionally uses a link key to decrypt traffic observed on the channel comprised by the hopping sequence. The second node 2111 thereby identifies traffic from device 2201 over communication channel 2210b, and optionally injects traffic on this channel as well. To communicate with the device 2201, the second node 2111 preferably takes the same role as the first node 2101, e.g., slave or master. In order to simply listen in, this is not necessary, but beneficial in order to avoid having to force the radio module 2112 to export the observed data to the controller 2116.
One or both of the controllers 2106 and 2116 each determines what traffic is observable to its corresponding node and generates as an output one or more identifiers associated with the device 2201, where this identifier comprises at least one of an identifier associated with the channel (such as the BD_ADDR used for pairing); an identifier used to identify the slave of a piconet; a MAC address (which commonly coincides with the BD_ADDR but may in some implementations be a different number); and/or other identifiers, such as a nickname of the node, and data that characterizes the observed device such as the device manufacturer. The identifier that is output is used to determine proximity of the device 2201 and one or more nodes such as first node 2101 and second node 2111. In addition, signal strength associated with the observed communication can be output, and can be used for purposes of triangulation. At least one node, such as the first node 2101, the second node 2111 and/or the central storage and control 2121, receives one or more signals corresponding to such outputs and makes a determination as to location of device 2201, and possibly other devices not explicitly shown in the figure.
Node 2111 may associate a policy with device 2201, specifying the circ*mstances with which to generate or communicate an output, where example policies include generating and communicating an output only when a change is seen from a previous time period (e.g., when a first-absent device 2201 is detected, or when a first-present device 2201 is no longer detected); generating and communicating an output at periodic intervals, such as every ten seconds; and/or generating and communicating an output whenever an environmental change is observed, where example environmental changes include the detection of a sound above a threshold level; the detection of a motion using sonar above a threshold level; the generation of an output from a nearby node; etc. In one embodiment, at least some nodes each periodically transmits a message for each of the piconets it is aware of, requesting any slave on the piconet to respond. Note that this can be done whether the node is the slave or the master, as well as independently of the node performed the pairing with the device is in the first place, or simply received information from another node that did.
Further illustrative embodiments will now be described in detail with reference to FIGS. 23 through 30. These embodiments relate to privacy control and enhancements for distributed networks, and can additionally or alternatively include other privacy-related features and functionality of the type described elsewhere herein. Accordingly, techniques described below in conjunction with FIGS. 23 through 30 can be implemented in other information processing systems disclosed herein.
With the rapidly increasing deployment of sensor networks, whether for residential or commercial use, and the connection of such networks to centralized authorities that collect information from the sensor networks, there is a great need for technology to control and enhance the privacy of users of such systems. However, conventional technologies fail to adequately address the privacy needs of users. One reason for this is that privacy is a very broad concept, and that typically, it is not possible to use one technical solution to address a problem other than one for which it was developed and used, as described elsewhere herein. With differing contexts and different needs of the various legitimate parties, privacy comes to mean different things. Whereas some concepts, such as encryption, are useful within many of these contexts, encryption does not solve the problems by itself, and complex solutions involving different constructions are needed. One reason why practical developments in privacy has been very slow for many years, causing great concern in society, is the difficulty of addressing new problems relating to privacy, even given knowledge of previous problems and their solutions. This, again, is because privacy is a very broad concept and different privacy problems, corresponding to different contexts, cannot be addressed in the same way, or even, using the same underlying building blocks or technical principles.
One important problem to address is that of consent to be tracked, where tracking means the provision of a personalized service or action based on detection of one or more features associated with a user. The service may be provided to the tracked user, another entity, or a combination of these; wherein the other entity may be another user, a service provider, law enforcement, etc. The tracked user is commonly a person associated with the node network, but can also be a person not associated with the node network, as well as pets, robots and property that typically is not mobile or moved but which can be moved.
At the heart of providing privacy to users is the notion of identifying what rules to apply to what users, based on the potential recognition, by the system and of the users. Some users wish to receive services that require tracking and communication of data related to their location, actions, and/or context, but not all users wish for this to happen. Therefore, in order to selectively apply rules related to services it is necessary to identify the user(s), based on observations, and to determine whether an observed user wishes to be tracked based on an opt-in component wherein users or their representatives identify their privacy requirements, or an alternative configuration method in which privacy preferences are expressed. At the same time, it is important for the system to distinguish contexts and perform selective tracking independently of the user privacy preferences. For example, consider the following five examples of detected users:
1. A resident who has opted in to being tracked in order to receive services that require tracking;
2. A guest who has not opted in and who does not wish to be tracked;
3. A burglar who does not wish to be identified, but who does not have any rights to privacy as a result of having unlawfully entered;
4. A person who does not wish to be tracked but who is trapped in an emergency situation, such as being trapped in a room while portions of the house have been destroyed in an earthquake;
5. A dog, which may be assumed to have no rights to privacy, but who must not affect the privacy of others as it enters a room in which there are people wishing for their privacy not to be negatively affected.
In this example, the rules applied to the different users (including the dog as such a user, and the burglar as an unwitting or unwilling user) differ.
The first user has opted in to some services, but not to others. Accordingly, the rules applied to this user are limited by what type of services or disclosures of information the user has agreed to. In one example, the user may have agreed to being tracked as he or she moves around, in order for the lights to be turned on and off as the user moves around, but no other type of tracking. In another example, the user may have agreed to being tracked as he or she moves around, and to be given coupons corresponding to his or her inferred needs.
The second user does not wish to be tracked, nor has he or she expressed any wish for personalization. Whereas the system could identify the user based on the user carrying a phone with an active Bluetooth radio, or based on sonar, for example, the system does not use this information for any personalization, nor the generation of any coupons, to mention two example services.
The burglar has also not opted in for any services, but is different from the guest. The system may be able to distinguish them a priori with a high likelihood, as the guest may ring the doorbell and be let in by a resident; may have been at the home many times before; may be at the property in the company of a recognized resident; may have entered using an access code provided by an admin user; or otherwise exhibit a behavior consistent with being a guest as opposed to a burglar. The burglar, in contrast, may enter in an unusual manner, e.g., not through the main entrance; may not be met by a resident (whether the resident is at home or not); may enter a large number of rooms within a short period of time; enter the home when the alarm is armed, but without disarming it; or other manners associated with a burglar as opposed to a guest. However, a guest or a burglar may behave in a way that cannot be distinguished, a priori, by the system as being one as opposed to the other with sufficiently high certainty. The system may notify a user and/or law enforcement or private security in response to detecting a burglar, and generate a log of features captured from the believed burglar and useful as evidence, where some such features (such as camera footage) may be provided to an admin user. This would not be done when the user is assessed to be a guest. A third action may be taken if the system cannot determine with high probability of correctness that it is one or the other. For example, the system may then generate a log of features but not expose this to the admin user, except if the admin user files a police report and provides an operator associated with the system management with a record showing that the police report was filed, in which case the operator or an associated person may review the logs and determine whether it is likely that they contain any evidence useful to determine the identity of the potential burglar.
The fourth case corresponds to a person who does not wish to be tracked, but who is in an emergency. The system identifies the location of the person and notifies emergency responders. This is done whether the user has expressed a wish to be tracked or not. Whereas it is also beneficial for first responders to be notified about the whereabouts of a pet, they will want to know whether a trapped user is a person, a dog (that may be dangerous), a cat (that may be scared); and also, to the extent possible, whether the person appears to be a child or an adult, to the extent that the system can determine such information. Therefore, it is beneficial for the system to build profiles for users who do not wish to be tracked; and track such users; but not desirable for the system to provide services or share data with service providers unless the case is an exception, such as an emergency.
The final example relates to a dog. Some pet owners may wish to know the whereabouts of their pets; know whether they are at peace or not; be able to watch camera footage of the pet, to the extent that is available, and so on. However, this ability to observe the pet must not negatively affect the privacy rights of another user, such as a person in the space. Therefore, the system will determine the whereabouts of both pets and of other users in the space, and determine, based on the privacy settings associated with potential human users, what rules to apply to captured sensor data. These rules correspond to different services, each one of which corresponds to one or more conditional actions that are performed based on sensor inputs, configurations, user preferences and privacy settings, along with other information provided among others by service providers, as will be understood by a person of skill in the art.
Moreover, the disclosed technologies enable the system to determine whether there may have been assistance from the inside given to a burglar, e.g., in the context of a disgruntled employee helping a burglar with information useful to burglarize a business. This can be done by correlating profiles, where one profile is associated with the burglar and another is associated with a known/recognized user.
These are illustrative examples of the context that the disclosed technology addresses. It should be noted that these are non-limiting examples and are used for purposes of concreteness and in order to convey the general principles associated with the disclosed technology. Turning now to the manner in which illustrative embodiments of the disclosed technology address these and related situations and problems, we will break down its functionality in terms of different functional components.
One of the previously unsolved problems addressed in this disclosure is the question of how to request permission from a user of a device, when this device does not have an appropriate user interface, or where the system does not have a manner of initiating a communication with the device. A related problem that is also addressed in this disclosure relates to how to request permission from a user to detect the user based on some behavioral measurement, measurement related to biometrics, or other measure associated with the user but not a user device. In this context, there is also a problem with how to convey the request. One aspect of the difficulty for these different related problems is the fact that the correct user should be given the request, as opposed to another user, except in cases where one user acts as a proxy for another user, such proxy being given the right to respond to requests. The requests in this context relate to various aspects associated with privacy, including but not limited to sharing of associated data such as measurements related to features to which the requests pertain, the performing of services related to features to which the requests pertain, the logging of information related to the features to which the requests pertain and the associated processing and disclosure of such logs with users and service providers. Here, a feature relates to an aspect associated with the entity, where an entity may be a person, a mobile device, a pet, etc. This is further clarified and described in this disclosure. Other problems, and aspects of these, addressed in this disclosure will be recognized by a person of skill in art.
In the context of sensor networks, the tracking may be of personal devices equipped with radios, which act as proxies of identity; non-limiting examples of such devices include phones, tablets, portable computers, wearable computers such as health-related devices; as well as devices that are not typically mobile, such as Internet enabled refrigerators, TVs, and other devices that typically remain in one location. Tracking in the context of such devices might mean an assignment of functionality of the space the non-mobile device is located in, e.g., kitchen or TV room. Tracking can also be performed relative to individuals, e.g., using video cameras and facial recognition technologies, or using microphones and sound-profiling technologies that associate identified characteristics of sounds with a given user. Some of the entities tracked have traditional input/output (I/O) capabilities, e.g., a phone has a touch screen that can be used for conveying information to the user, and for the user to convey information to the system. However, some wearable technologies do not have meaningful I/O for conveying complex information, or even, any information at all. For example, the only manner in which a user can communicate with an Oura ring is using the associated app, running on a smartphone, etc.
Furthermore, some of the detection described here is of persons as opposed to devices. This is another example of a situation where a tracked entity does not have a user interface allowing notifications and user feedback. To make it worse, even some devices that can be tracked and that do have a suitable interface, may not be connected to the network. For example, some sensor networks do not interface with user devices, such as smartphones. Rather, the sensor network and the smartphones coexist side by side, without interacting with each other. Therefore, absent use of techniques disclosed herein, there exists no practical manner in which notifications from a system related to the sensor network can be relayed to such a smartphone, or receive feedback from a user associated with the smartphone. Notifications is just one of the aspects addressed in this disclosure, though, the wider area being privacy management.
Further still, even if a smartphone were connected to the sensor network or otherwise connected to the associated system, such an arrangement does not allow the system to convey notifications (such as requests) to devices that are not connected. This is a major drawback since it is beneficial to profile a user based not on only one type of input (such as the identity associated with the radio of a smartphone), but several types of input. Using several types of input will increase the accuracy of the determination associated with the tracking, and therefore enable better services to be provided to the user. For example, a personalization service provided by the sensor network or an associated party would fail to provide meaningful profiling if it uses the identification of the radio or another component associated with a smartphone, but the user associated with the smartphone has set the smartphone to charge and does not carry it with himself or herself; or the user has lent the smartphone to another user. A greater number of inputs will therefore be useful to improve the accuracy and coverage of the personalization services, while maintaining the privacy of the users.
One component of the disclosed technology is to automatically determine an association of two or more features to each other. Here, example features may correspond to an identifier associated with a user's smartphone, an identifier associated with a user's wearable computing device, and the like. It may also correspond to a user as observed by a representative location, a representative movement pattern, a representative behavior, a representative frequency when speaking, etc. The system observes and stores these features in a database, along with information related to the time and context (such as location) they were observed. The location may be represented by the sensor that reported the information, or, if reported by multiple sensors at the same or two or more closely related times, a triangulated location as determined by the sensor data. Methods to automatically determine these and other example features are described in further detail below.
A second component of the disclosed technology is to generate profiles associated with one or more features. Correlation of features may be used to establish an association between the features. For example, if two features are correlated with one another, their correlation value may be compared to a threshold. If the correlation value exceeds the threshold then the two features are associated with each other and associated with a profile. A profile, in turn, can be considered as a description of a person or a role of a person. For example, a profile may correspond to one resident of a home that has a sensor network. This profile may comprise three features that are found to be strongly correlated, such as an identifier associated with a smartphone, an identifier associated with a Fitbit™ device, and a feature related to the voice characteristics associated with the user speaking. It may be that not all three are constantly associated with the given user. In one embodiment, the profile contains information describing the typical association, the typical correlations between pairs of features, and possibly additional or alternative information.
Another way of generating profiles is to determine a relationship between two or more features. This can be done in addition to the finding of correlations, as described above, or instead of the finding of correlations. A relationship may be the belonging of two or more devices to a piconet, for example. A piconet is a group of devices connected to each other by Bluetooth or Bluetooth Low Energy (BLE), which will be collectively referred to as “Bluetooth” for simplicity. If a smartphone is the master of a piconet and wearable computing device of a type that only can have one master at a time is a slave of the piconet, then the system determines that the wearable computing device belongs to the same user(s) as the smartphone. Accordingly, these two features (information relating to the wearable computing device and information relating to the smartphone) are both added to the same profile. The system also adds to the profile information relating to a device that is a master of a piconet, for which a device already represented in a profile is a slave. If two devices belong to the same piconet and have been found to correlate to each other, as described above, then these two independent assertions of belonging together causes the system to represent features of both of these devices in the same profile. The features to be stored could be generated using methods such as those described previously herein.
A feature (such as an identifier of an iPad) may be associated with more than one profile, e.g., if the device is shared by two or more people. A profile may also correspond to one or more persons having a given task or role; for example, a person who delivers packages in a building, using a cart to carry the packages to be delivered. The features that correspond to this role may be a radio device that is associated with a motion consistent with the movement of a cart on a delivery round, whatever device is detected to exhibit this pattern; and an identifier of a device that is associated with a person who has previously been associated with the delivery of packages, and a feature that corresponds to the sound of the cart typically being used for deliveries of packages. The associated profile would comprise information describing these various features.
Another component of the disclosed technology is a mechanism to select a device with appropriate I/O capabilities to be used to convey a request to a user associated with one or more features associated with a profile. In one embodiment, one of the features associated with the profile corresponds to a device that has connected, directly or indirectly, to the network or the associated backend, and which can be established by the system to have I/O support. For example, if a user installs an app or uses a web browser on a smartphone, and accesses a service associated with the system using this, then the system is able to convey information to the user at this point. If a user repeatedly uses, for example, an app on his or her smartphone to control some aspect of the environment, e.g., lights in a room, the channel or volume of a TV, or the locks of a door, then the system can present a request to the user in the same interface, such as the app, or an associated interface such as in an email sent to an email address associated with the user in the app.
Moreover, if the system collects contact information from such a user during a session of interaction using an app or web browser, then this contact information can be used to convey a request to the user. For example, the system may collect a phone number associated with a user in a context where the user requests access to a service, and then use that phone number to transmit a request by SMS or synthetic voice in a phone call. For example, a system may require that a user wishing to gain access to the WiFi enters not a password but a phone number; then, the user will get an activation message sent by SMS to this phone number, and perform an action which may involve entering a password, installing an app, entering additional user information, or simply clicking on a link.
In response to the user performing this action, the system records the phone number and associated information about the accessing device, the latter comprising at least part of a feature and the former corresponding to an I/O information field of the associated feature. The I/O information field may also be part of a profile. The system may now grant access to this device using the WiFi. The system may also transmit a temporary password that the user can enter into another device in order to give that device access to the WiFi. As this second device first uses the WiFi, the system extracts identifying information relating to the second device and uses this to generate a new feature.
Since the second device is known to be associated with the first device (that was used to obtain the password), the system has established that these two devices are likely to be correlated. This can later be confirmed by observing co-presence of the two devices at one or more locations for one or more times. If it is confirmed that the two devices are likely to be correlated then their associated features are entered in one or more profiles. The profile information may contain information describing the assessed strength of the bond between two or more features associated with the profile. This bond is either a measure of correlation or relationship.
The system may also identify one or more devices with I/O capabilities that are co-located with a device associated with a feature. For example, the system may determine that there is one user in a room, and that this user has a device such as for example a wearable computing device, where this corresponds to a feature observed by the system. The system then uses a device with I/O capabilities in this location to convey a message to the user. For example, this may be by conveying a message “Say ‘register’ to register your wearable device, and get useful personalization opportunities. Say ‘tell me more’ to learn more. Say ‘later’ to be asked later. Say ‘never this device’ to not register your wearable device. Say ‘repeat’ to repeat these instructions.” Here, this message may, for example, be rendered on a TV that was just turned on, or conveyed using a synthetic voice over a speaker in the room. The user may respond to the request, e.g., by saying ‘register’, which will be registered by a sensor of the sensor network and conveyed to the backend, which can then perform a registration of the user.
One aspect of the disclosed technology is that the system can ask the user whether the user wishes to register multiple aspects. For example, the user may register a wearable device as described above, and then be given the message “Do you also want to register your smartphone?” followed by the options given before. The way the system would know what smartphone to associate with the user by determining what smartphone, and its associated feature, is associated with the same profile as the wearable device that was registered, assuming the bond is sufficiently strong for this association to be made.
A feature that does not correspond to a device can also be registered. For example, if the system has determined that a feature associated with the frequency band of a user speaking is associated with the same profile as a feature that has been registered, such as the wearable device in the example above, then the system can convey a question whether the user wishes to register that feature as well. For example, the system can convey the message “Do you also want to register yourself based on how your voice sounds? We will not record what you say, but only recognize that it is you. Say ‘register’ to register your voice. If you do, we will know that it is you when you give voice commands. For example, if you say ‘what are my appointments?’ I can tell you based on your calendar.” In addition to such services, the system also will be able to determine the identity of a person based on the sound of his or her voice, and apply personalization based on this, just like the system can personalize based on other features, such as wearable computing devices.
One aspect of the disclosed technology is to associate one device with multiple features, where the association depends on what user is logged in to the device. As an example, a shared device, such as a tablet, may correspond to two different features, where the first feature corresponds to a first user being logged in and a second feature corresponds to a second user being logged in. One way of determining, by the system, what feature the device corresponds to is to use information stored in or associated with a browser of the device, where two different login contexts correspond to two different stored information sets. The information can be in a form of HTML cookies, cache cookies, and other customizable objects associated with the browser. As a user uses the browser, the system may cause the browser, using redirection and traffic injection, to visit a website where the identifiers are read.
Another approach involves correlating the device information with a context or usage, where a context may be the use of a particular application, or the usage may be a particular location and/or time of the day. Such context and usage data may have been correlated with other features in the past. Once the system has a sufficiently strong correlation, i.e., exceeding a threshold value, then information related to the context and usage become part of a feature that is associated with the profile having one or more other features to which the context and usage data was correlated. If at a later point this same context or usage is detected then the device may still be assigned with said feature.
To the extent that a user of a device is unable to register the device and its corresponding feature, this can be done at setup or otherwise performed by an admin. For example, a teacher or admin may register a device as described above, or require the use of software that causes a registration. The registration of some set of features may therefore be automatic in some situations. For example, a school laptop computer may be configured by being registered, allowing, for example, making possible the tracking of the laptop computer and an automatic association with one or more features to determine who is using it. A school may also require that students register their phones, which will enable the school to track the location of the students on the premises when this is desirable. Employees of a company may have smart badges that are automatically registered, or be required to register their smartphones; the associated tracking can be used for purposes relating to access control of buildings, charging of meals onto employee meal plans or cash accounts, etc. It can also be used to correlate behavior with performance, determining for example that there may be a strong correlation between success and taking breaks every two hours. This determination can be made at a level that no individual user behavior is highlighted, but where general relationships are established. This permits the improvement of employee health, productivity and other important aspects.
Another approach to register one or more features is to offer a user an enrollment opportunity where one or more features are detected, observed or provided information about, and then the associated information is stored in a profile. For example, a user can be asked to connect a phone to a service, and the service determines one or more features associated with the phone, e.g., an HTML cookie, user agent information, and user demographics provided by the user. A local network node with sensors may further detect a MAC address and confirm one or more aspects related to the user agent information, among other things. The submitted information (which may be user input and/or other information that is part of or related to the device) is transmitted to a web site that is associated with the system, or captured by the system as the system observes network traffic using one or more network nodes. Locally captured information, such as information observed using a sensor network node, is also transmitted. The obtained information is used to generate one or more features of a profile. In one related example, an app that the user has installed on the phone has optionally obtained some demographic information, and generates one or more signals such as WiFi signals and/or Bluetooth signals, and transmits these along with an identifier such as a unique number. The same unique number can be used along with the transmission of other associated information, such as the optional demographic information, which can therefore be associated with the same profile. As a local network node detects the WiFi and/or Bluetooth signals, these are also added to the profile.
Person-related features can be obtained in an enrollment stage, too. One way is for the system to inform the user to perform an action, “Now say ‘hello, my name is’” five times. The system would then identify that spoken phrase (there may be multiple speakers, the other which should not be part of the feature) and identify the commonality in characteristics between the utterances that match the requested phrase according to a voice-to-text mapping such as a dictation software process. The commonalities are used to generate a characteristic that corresponds to the feature. Note that the repeated observation of the same phrase enables the system to perform this determination in a rather noisy environment without risk of incorporating “foreign” characteristics in the generated feature. Enrollment can be seen as a special case of registration, and it is assumed that the appropriate permissions are still given. As before, a feature is useful even if the user does not register it, e.g., for emergency response. Thus, a user may be asked to perform an enrollment action and may then be asked what features he or she wishes to register; the registration may also be implicit, and part of the enrollment.
Another aspect of the disclosed technology is the detection, by one or more sensors of the sensor network, of one or more features, based at least in part on receiving radio communication information, receiving sound inputs using a microphone or other sound input sensor, receiving video input signals using a camera or other video input sensor, receiving movement information using a sonar, radar or other movement input sensor, detecting actions associated with sensors, such as a light being turned on using a switch, and receiving data input such as input from a device being transmitted to a radio receiver or conveyed from a backend server. A person skilled in the art will recognize that these are only examples of inputs from which the system can determine features. One example of a processing method to generate a feature from such input, or to map an input of that type to a known feature, is the extraction of one or more device identifiers from a radio signal, e.g., a Bluetooth Device Identifier and, optionally, a master or slave identifier number in addition to the Bluetooth Device Identifier. Another example is the mapping of a sound signal to a voice-related feature that uses the mapping of the sound signal to a frequency spectrum, e.g., using an FFT, and the comparison with one or more templates associated with previously detected sounds; where at least one of these templates correspond to a feature that is identified in a record associated with the stored template.
The system detects one or more features over a period of time that is referred to as the time window, where the time window may be 30 seconds, for example. As a feature is detected, the time it was detected is recorded in a database, along with an identifier of the feature, and, optionally but preferably, information relating to the context of the detection. Example contexts are the location of the detection such as the room, the sensor or sensors that detected the feature, or a more detailed location based on triangulation of such location data. Other contexts include information about preceding events, such as an entry door was opened; an unusual sound was recorded, etc. Based on two or more detected features, and potentially auxiliary data such as data from sonar or radar that identifies where there is movement, and potentially how many people are within an area, the system determines whether two or more of the features are likely to belong to the same context, e.g., the same person or event. Here, an event may be a series of detections such as opening of a door and associated detection of pressure change; sound of a person moving around; sonar of at least one person moving; sound of closing of a door; sound of a person speaking; sonar of at least one person moving—this event, for example, may be classified as “entry.” If two or more features are associated with each other, this is preferably recorded and/or used to determine how closely these features are correlated, which is then recorded. If one or more features of a record are detected, then an optional action is taken. The action may relate to a service, for example. The rules generate predicates that are used as input to service providers in one embodiment. Examples of rules being evaluated are described elsewhere herein. In some embodiments, rules are evaluated to generate predicates. Moreover, rules are evaluated to generate control signals that are transmitted to nodes in a node network.
The nature of the action taken based on the detection of one or more features depends on whether these one or more features are recorded; and if there are multiple detected features, whether these belong to the same or different profiles. More specifically, for each profile that a feature has been detected, an optional action is taken. If such a profile is matched using at least one registered profile, then the profile match is said to be public, whereas if a profile is matched but not using any registered profile, then the match is said to be secret. The match does not need to be exact, but is preferably done with respect to a threshold and/or a confidence interval; that also applies to other matches and comparisons described herein. When a profile match is public, an action that is potentially detectable by the user corresponding to the matched profile is optionally taken, e.g., lights are turned on or off; the user is given voice notifications of recorded voice mail; the system determines that the user has a need based on a context and provides the user with a notification, a coupon, an advertisem*nt, etc. If the user gives a voice command, the system attributes this to the user associated with the matched profile. For example, if the user gives the command “Heat my car” then the system selects what connected car to send a heating command to, and then sends such a signal. In contrast, if the match is private, then the user is not provided with such services. However, the system still records the actions and location of the user in a log. If there is a disaster, such as an earthquake or a fire, and the user appears to be trapped or unconscious, then the system notifies first responders, and identifies the location of the trapped/unconscious user. As another example, if the user turns out to be a burglar and the homeowner later files a police report, then identifying data associated with the burglar's associated features are optionally transmitted to law enforcement. As yet another example, the user may be provided with the option to register a feature, as described above. Thus, based on the classification of one or more detected features into public and/or private classifications, one or more actions are taken, where the nature of the actions are based on the classifications.
Thus, the system provides services or enables the provision of services at least in part on the contents of a profile, where profiles contain at least one feature, and features may either be registered or not, where this impacts the selection of the services that are provided.
There are other examples of where access control, permissions and privacy intersect and where the system permits partial access to premises or functionality based on determination of identity or role, and where rules affecting the privacy of the associated users may be affected. For example, the system may enable the unlocking of a door for the delivery of a parcel, but causes the recording of the person entering, and if the person passes a given point (e.g., leaves the entrance room) then an alarm is activated. This is therefore an example of a special case related to burglar alarms, access control and privacy. Another example is one wherein a plumber intends to replace a water heater, and is registered in the system by being sent an SMS that allows the plumber to visit a website, in turn causing his or her smartphone to be observed and a feature created. An admin may specify the access control restrictions of the plumber by selecting from one or more rules, or by configuring one or more rules. For example, as a result of the selections and/or configurations, the system can create one or more rules that allow the plumber access to the home on a given day within a selected time interval, but only to the basem*nt and the entrance area, and triggering of a recording of him or her as he or she moves around. If the system detects that the plumber enters an area that is not allowed, then the system automatically enables all cameras, initiates a call between the plumber and a third-party security service organization, where the third party security organization can verify the circ*mstances and determine whether to initiate an alarm or not.
It may be decided that any device that connects to a network, or which is connected to a device that connects to the network, accepts to be tracked by the network. This is a reasonable place to draw the line in terms of what devices (and whose devices) are tracked when they are in the physical proximity of a network. This way, a resident of a home that corresponds to a network, who connects to the network using, say, a smartphone, would be allowed to be tracked, by this network, as it is in its proximity. Moreover, if this smartphone is connected to a wearable computing device such as a FitBit, e.g., using Bluetooth, then the wearable computing device, in spite of not connecting directly to the network, would be allowed to be tracked. This makes sense to the user who owns and/or operates the smartphone and the wearable computing device. However, if the mailman has another wearable computing device, which is connected to the mailman's smartphone, then neither these two devices would be tracked by the network. The same goes for the neighbor of the network owner, who is commonly in the proximity of the network but who is not a user of the network, and therefore would not want to be tracked. In the above, all tracking relates to tracking that can be observed by a user admin of the system, e.g., reviewing the logs, or for which the system takes configuration actions (such as turning on the lights) based solely on detecting a device being in the proximity. Consider now a burglar, and assume this person also has one or more devices, such as a smartphone and/or a wearable computing device; and assume further that neither of these are connected to the system, and therefore not trackable in the sense above. However, the system still records identifiers in a log that is not directly viewable by the user admin, but which can be handed over to law enforcement, should there be a need for this. Moreover, the user may be allowed to know of the existence of a recording of the identifiers of the burglar.
Moreover, given that many burglars may “case the joint” ahead of the burglary, the system may have additional data that it has observed, comprising feature detections in previous sessions; these may be possible to correlate to a user identity, whether by the time(s) of the appearances or by coinciding observations, such as number-plate detections of automated number plate reading systems; observations using footage from security cameras during the pre-burglary visits; and by other collected information as understood by a person of skill in the art.
In one embodiment, for any feature that is not associated with a given user, and which does not belong to a category of features that are not associated with privacy concerns (such as pets), that user is prevented from seeing information relating to such features, unless given permission by users associated with such features, which the system can verify using one of the approaches described herein, or versions thereof. Thus, this user, who may be an admin, in this embodiment, is not allowed to inspect any features of the kind for which there are privacy considerations, unless these features are associated with this user, or the user(s) associated with these features have given permission for the user wishing to see the features to do so, or is a minor that is a guardian of the admin or a party associated with the admin. This permission can be given on a case-by-case basis or the user whose features are potentially to be inspected can provide a blanket agreement, which may be limited to certain users wishing to see features; to a certain time period; to certain features or classes of features such as only device-related features; etc. Here, seeing the features means to be able to inspect information related to the features, such as descriptive information.
In another embodiment, an admin may be given partial access to descriptive information, such as generic information (“a phone that is not associated with a registration decision”); or can see full descriptors but not any associated historical information (e.g., information about when the features in question were observed or not observed); or other contextual information such as with what other features there is a correlation detected by the system. This can also be combined with the previous embodiment in which a user is granting permission, where such permission may be related to an increased insight related to the features, e.g., what features are correlated with each other.
One aspect of the disclosed technology is the clustering of features and association of these with a profile, as in certain examples given above. The clustering method can use time, location and other context to identify features that likely belong to the same entity, or which commonly coexist for a given entity. One person may correspond to multiple such clusters, as the clusters are based on observations, automated classifications, and thresholds. A person may also correspond to multiple clusters based on having multiple roles, wherein the system identifies features and correlates them to each other based on the role. For example, a person may carry and use a first smartphone (say the personal phone) in some contexts (such as during the weekend and at nights) but carry and use a second smartphone (say, a work phone) in other contexts (such as during the workday). This may lead to the creation of two profiles associated with this person. The person may decide to keep the two profiles independent, to enable the provision of services based on the role. For example, the person may want work-related notifications and alerts to be provided when he or she is in the work persona, which may not coincide with the regular hours for this persona and may instead be determined based on observations by the sensor network of the features that are detectable. The person may also decide to merge two profiles of this type, and optionally, associate different rules with different sets of features, which also enables a treatment and a service provision that is based on the features that are detected, in turn corresponding to the context of the user. We use ‘user’ and ‘person’, and other related pronouns interchangeably herein. A cluster may have only one feature associated with it. A cluster may correspond to more than one person, in the case a device corresponding to a feature is used by multiple people, or in the case of visual appearance, two people may look very much the same, as two twins may do. When a feature is associated with multiple clusters or with multiple profiles, this may correspond to one of these situations, which may prompt a user interaction using an interface, where the user can clarify whether the corresponding classification corresponds to one or more people. In some instances, the system may resolve this without user input, simply by observing two or more very clear clusterings that are distinct. This may be the case when two people take turns using one device, such as a tablet. This tablet, therefore, becomes indicative of both of these people, and the determination of what person is present is made based on the detection of other features associated with the person in question. This is still simplifying the classification as any one feature may not be indicative of a given person, but a pair or a larger number of features may be.
A cluster does not have to correspond to a person or a role. For example, one cluster may correspond to a pet, where the pet is detected based on a radio module (e.g., for operation of a pet door) which may correspond to one feature, and/or detected based on visual indications that may correspond to another feature, and/or detected based on a movement pattern that may correspond to yet another feature. These are only examples of features associated with a pet. The corresponding cluster of features are associated with a profile by the system. The system may also generate features, clusters and profiles relating to inanimate objects, such as a cleaning robot. This may be done by identifying features indicative of a radio, sound patterns associated with the robot, movement patterns associated with the robot, etc.
In one embodiment, the system automatically classifies a collection of features corresponding to a profile in terms of a type. Example types are person, pet, robot, and appliance. However, these also can be given sub-categories. For example, a person may be a resident (observed very frequently), a common visitor, an apparent service provider, a likely guest, a first-time visitor, etc. This classification is helpful as an admin may assign different types of service provision, and different types of privacy protection in a manner that is based on the classification into type and potential subcategories.
In order to convey a request to track a feature and associated profile, and/or to provide services based on the detection of one or more features or a profile, one or more rules are applied for the system to determine the desired action to take. The set of rules to be applied depends in some instances on whether a feature is registered or not; it may also depend on the class that a type is assigned to. For a feature or a profile associated with a type that is of the class “pet”, for example, there are typically no privacy considerations, and when a pet is detected in any room of a building, it is acceptable to output video of the pet, assuming of course that there are no persons in that room, which would affect the privacy determination. This is an example of how rules can be applied to a single profile but the actions taken based on the evaluation of such a rule depends on the evaluation of additional rules as well. In this example, a first rule related to the presence of a pet, and a second rule related to the presence of a person. If the first rule indicates that there is a pet in a room, then this may trigger a photo or a video to be available for viewing, but only if the second rule related to a person being in the room is not triggered. In another instance, the classification of a room (e.g., living room or bedroom) may affect the determination associated with the second rule; and in yet another instance, an additional rule related to the individual person in the room, as determined by one or more associated features, is also used to make a determination. Thus, a set of rules are evaluated based on the set of inputs, which correspond to features, profiles, classifications of types associated with the features and profiles, and also on inputs that relate to the context, which may include the classification of the location, the time of the day, and whether an emergency exists, such as a fire has been detected.
The classification of a feature can be done in a variety of ways. One way, which is relevant in the context of a Bluetooth Device Address or MAC address, for example, is to identify the type of device based on this address. It is understood that different manufacturers have been assigned different address ranges; therefore, it is possible to determine the device type based on the address. The device type can often also be determined from headers, e.g., in the case of a GET request from a browser, which contains information about the device type, the operating system (which commonly is indicative of the device type), and the like. Moreover, the friendly name of some devices are sometimes not user configurable, and remain a manufacturer-selected identifier that commonly identifies the device. For example, Oura rings have device names that include the text “Oura”, as well as a unique device identifier; these serve to identify the type of device (a wearable computing device) and to provide a unique identifier associated with the device. The system may classify smartphones, typical wearable computing devices, etc., as being associated with people. On the other hand, a MAC address associated with a Roomba™ is indicative of a device type that is a robot, and a Bluetooth device identifier associated with a pet identifier (whether worn under the skin or on a collar) is indicative of a device type that is associated with a pet. Similarly, there are sound-based features such as the sound of speaking that can correspond to a person or a loudspeaker. By determining that the source is mobile and corresponds to a detected body mass that is not consistent with a cleaning robot, for example, it can be determined that the sound corresponds to a feature with the type of person. The sound of barking is indicative of a pet, with the further subcategory of dog. Moreover, a sonar or radar detection of an entity can distinguish a person from a pet from a clearing robot; and can commonly also distinguish a large person from a small person (corresponding to approximate sub-classes related to the class person, and based on comparison to one or more thresholds); and can sometimes distinguish a large pet from a small pet. Similarly, movement patterns can be used to distinguish a person from a pet from a cleaning robot, and can sometimes be used to determine additional subcategory information. This way, each feature can be associated with a type.
By obtaining sensor data related to multiple aspects of an entity (say a person) corresponding to multiple features and their associated types, the system can obtain verification of the type classification and determine when a classification appears sufficiently certain, e.g., when there is no type discrepancies, or only a degree of type discrepancies between different associated features that fall below a threshold. One reason why there may be discrepancies is that a human may carry a pet, thereby causing two or more different type classifications at the same time. Such discrepancies are typically not permanent, though, and therefore, the system will soon be able to generate a consistent classification of type. Furthermore, by assessing records over an extended period of time, the system can determine whether a given feature is just observed (e.g., potentially corresponding to a first-time visitor); always observed with a known feature (e.g., likely to be another descriptor of the same entity); sometimes observed with one set of features and other times with another (e.g., a shared device); observed with a high frequency (e.g., a likely resident or employee); observed periodically and for limited periods of time (e.g., a guest or a service provider); observed in the context of an intrusion alarm (e.g., a potential burglar), etc. Sound-based feature classifications commonly will allow a sub-classification of gender. Movement-based feature classifications commonly will allow a sub-classification of age or medical status. A person of skill in the art will recognize that these are simply examples of classifications into types and subcategories, and that each profile can have multiple type and subcategory classifications, and that each one can have a system-assigned certainty assigned to it, where the latter may be assigned based on determining the extent to which sequential observations are consistent, or based on the extent to which different associated features are in agreement with each other, whether at a given time or over a period of time.
The Bluetooth Device Address (BD_ADDR) is commonly protected for privacy reasons as it is being communicated. However, the 24 lower bits of the 48-bit BD_ADDR are publicly readable, which for practical purposes can be used to establish a pseudonym with very small risk of collision. This can be read when a device interacts by Bluetooth or BLE, by scanning bands and observing a band that is being used to communicate. Similarly, the full MAC address can be read as a device that communicates with WiFi. The MAC address and the BD_ADDR are not related, and therefore cannot be matched to each other by comparing the two values. It is desirable to determine that they both correspond to the same profile, which corresponds to one likely user or role. It is further desirable to determine that they both correspond to the same device, which is information that, when accessible, is stored in the profile. To address this need, it is important to correlate the two. One way of correlating the two is by using time-based analysis. If a device is used to communicate over WiFi then, with a certain probability, it will also at the same time engage in a Bluetooth or BLE interaction, or the other way around. Both communications will relate to the same approximate location. Therefore, the two identifiers will have a positive correlation factor. If this exceeds a first threshold then the two features are determined to correspond to the same profile. If it exceeds a second threshold greater than the first threshold, then it is determined that the two features also correspond to the same device. Another way of making this determination is to automatically observe the friendly name associated with both WiFi and Bluetooth/BLE communications. This may be “John's phone”, for example. To the extent that these friendly names are locally unique, which is common and which can be determined with a high accuracy, the MAC address and the partial BD_ADDR can be determined to belong to the same device. If the system detects two or more distinct MAC addresses with the same friendly name, or two or more distinct LSB24(BD_ADDR) values with the same friendly name, then it is determined that these friendly names are not locally unique. Then time-based correlation methods, as described above, can be used to disambiguate. Similarly, location-based correlation methods can be used, as well as correlation to other features such as voice characteristics. A person of skill in the art will recognize that there are many related methods of correlating two addresses or two features to each other, or to the same device. The system can also “disprove” a device-based association between two features by near-simultaneously detecting two features in substantially different locations. The extent to which two or more different features not being associated with the same device are observed in a near-simultaneous manner in substantially different locations is an indication of the extent to which the features are not clustered in time and space at the same time. Some features, such as voice characteristics, gait characteristics and wearable computing device features are expected to be correlated in time and space.
In some implementations, where for some device the same chip houses both the Bluetooth radio and the WiFi radio, the BD_ADDR and the MAC address commonly coincide or overlap to a great extent. Therefore, a system that detects a BD_ADDR (or portions thereof) that matches a detected MAC address (such as agreeing in the 24 least significant bits) will be able to determine, heuristically, that the two features correspond to one and the same device with a very high likelihood. This, of course, is in the context of a system detecting traffic from a limited geographic neighborhood, such as one building, as opposed to an entire city; or a system that detects traffic from a large area but which detects a matching BD_ADDR and MAC address from a small geographic area, during a limited amount of time. The area may be a square mile, for example, and the time may be one hour.
One aspect of the disclosed technology is the generation or selection of representative information to describe one or more features. For example, the system may detect radio transmissions associated with the Bluetooth protocol and a given piconet associated with A Bluetooth Device Identifier, and assign that to one feature. It may further detect radio communication using the WiFi protocol, along with an associated MAC address and a user-selected descriptor such as “Alice's phone”, and may assign the MAC address and the user-selected descriptor to a second feature. The system may further detect a person speaking and perform an FFT on the sounds detected; this is associated with a third feature. During the correlation process, described above, the system in this example determines that these three features have a correlation above a threshold, which may be either a system threshold or an admin or user selected threshold. Therefore, the system associates these three example features to the same profile. The system further establishes a high likelihood that the Bluetooth communications corresponds to the same device as the WiFi communications, e.g., by determining that the Bluetooth Device Identifier and the MAC address, based on their unique identifiers and the range of addresses normally used by a given manufacturer. In this example, that brand is Apple, and the identifiers indicative that it is most likely an iPhone 7. A user would likely not recognize the Bluetooth Device Address or the MAC address as being associated with the device, but would understand the user-selected identifier “Alice's phone” to be a meaningful descriptor. In addition, a system-generated descriptor “An iPhone, most likely an iPhone 7” which is generated based on the addresses by performing lookups and comparisons, or by interaction over an API with the manufacturer or a representative thereof, where the latter performs a determination. Another meaningful descriptor is a visual representation of an iPhone 7, or a combination of some or more descriptors as those described herein. For these example representations, the determined example descriptor is associated with both the Bluetooth feature and the WiFi feature. In one embodiment, the system is asking for registration of these two features collectively; in another it may issue two requests, one corresponding to the Bluetooth tracking capability, and the other corresponding to the WiFi tracking capability. In most instances, the former is preferred, as most users are unlikely to care one way or another how exactly their phone was identified. This also applies to other forms of tracking, e.g., of a browser or an app associated with the same phone with a high likelihood. The example above involves the use of three different features in the determination of one piece of representative information of one or more features. However, a person of skill in the art will recognize that in some instances a larger or a smaller number of features are used by the system to determine the representative information, and in some instances, it is sufficient with one feature.
Features that correspond to non-radio related identifiers are also determined. For example, the system can represent a voice-based feature by one or more short recordings of the voice. The recordings should preferably correspond to a collection of single words, or short series of syllables, to make the representative information non-indicative of what was being said. For example, consider a situation where the system records a user saying “I am going to go to the match now”, and then another situation where the same user says “Do you know the time?” and a third in which the same user says “The bill is yours.” The system may generate representative information by selecting snippets corresponding to the words “go to”, “you”, “time”, and “is yours”. It is not necessary for each snippet to correspond to one or more full words. The system may have a blacklist of words never to include, and perform a mapping between the recorded series and this list. For example, the blacklist may comprise the word “kill”, and therefore, no selection that is close in hom*onym space to “kill” would be selected. Depending on the thresholds used, that may include the word “bill” in the above example. In another embodiment, the system will not use any word or syllable sequence that is spoken within a threshold time, such as two minutes, to a spoken word that is a word on the blacklist. The representative information in this example is one or more snippets of recorded speech.
As another example, a feature that comprises motion data that is indicative of a user can be represented in a variety of ways. One is for the system to relate it to another co-present feature, whether it has already been registered or not. For example, the system can request a registration for this feature by presenting the following request: “Personalization can be made based on movement patterns. The more ways in which you agree for us to detect that it is you, the better does the system work. Do you want for us to detect that it is you based on movement patterns?” Here, this request would be presented on a device associated with the co-present feature, e.g., a phone, or using a messaging channel that has been matched with the phone. This way, the system does not need to convey representative information related to the feature at hand, as it relates it to another feature for which it has an already established communication channel to the user. This can also be done for features for which there is a practical way to convey representative information that is meaningful to the end user. Another approach is for the system to make a request at a general I/O interface, e.g., on a sensor node such as a smart light switch. This is not a node that is specific to the user, but the request can be made at a time when the user is detected to be present at the location of the smart light switch, and when no other users are detected to be present. This is also possible for features for which there exists meaningful representative information to be conveyed to a user. Furthermore, the system can present unrelated information as a way to convey the request. Assume that the system has an image or a name or other identifying information related to a user. An image can be obtained from a sensor with a camera, or by a user uploading an image when registering another feature, or by the user having selected an image from a collection of selectable images that may comprise photos of the user, generic images that are available to all users to choose from, including avatars, and images that represent activities the user is associated with, e.g., a soccer ball for a user who appears to be watching many soccer games on TV. By images we also include videos, which may incorporate audio representations of relevance. A person of skill in the art will recognize that there are many other ways to generate or select visual representations that can be linked to the user, whether by a user selection, a user action such as uploading an image, or by correlation with an apparent action of the user, presentation of images relating to hardware in the apparent possession of the user, etc. Textual representations, similarly, can be made in corresponding ways, whether along with visual representations or in isolation; similarly, sound representations, as described above, can be used. Thus, it is not necessary to use a sound representation for a feature related to sound; nor a visual representation for a feature related to an image, etc. The system selects a representation and an associated explanation to clearly convey to the user what the request relates to, or selects the appropriate communications channel, or a combination of these.
Another method of obtaining registrations of features involves making blanket requests, such as “Do you want to add tracking of any other feature that can be determined to be associated with you, without having you respond to individual requests?” This is presented to a user in a way that permits the system to associate the response with at least one feature or at least one profile, in one of the ways described above or a variation thereof.
Not all users are capable of determining whether they would like to enable tracking relative to their detected features or not. A user acting as an admin can therefore make determinations on behalf of such users. For example, such an admin user can use an interface to request a list of profiles, each associated to one or more features, where each profile is preferably associated with one or more types or subcategories. The admin can also select to only view a subset of all such profiles, based on selection criteria such as a type or a type and a subcategory. An admin that selects the type pets is presented with a collection of profiles associated with the detected pets; the admin can then review representative information for some of these to verify what pets these correspond to. One pet's representative information may be “medium sized dog,” and another may be “small dog or a cat.” For the second one, the admin can specify that it is a cat, thereby correcting or improving the classification. The admin can then select both these profiles and indicate that the pets can be tracked at any time, but that the only form of personalization that should apply to the pets is that their presence in the proximity of the pet door should unlock the pet door.
The example admin then moves on to select another type of entity, namely robots. The system in this example only has one profile, and it corresponds to representative information “likely cleaning robot.” The admin selects the profile and selects a model number from a list of model numbers, and then provides an access credential for the system to be able to access and control the cleaning robot. Moreover, if the system detects that the cleaning robot has gotten stuck, it permits an automated notification to be sent to the admin or another user. The admin selects that option, and specifies who will receive the notification, from a list of users associated with the system, each one of which has been associated with a communication channel such as an app, an SMS number, an email, etc.
The example admin then selects another category, corresponding to “people who have not performed any registration decisions.” There are two such users. The admin reviews the representative information for one of these users, and realizes that this is a child. The admin selects to approve all tracking for the purposes of personalization, but provide no notifications or direct communication to this user, and to notify the admin of the location of this user, when the admin selects to want to know. The admin reviews the descriptive information associated with the profile and then edits this from “Jonathan's iPhone” to “Johnny.”
The example admin further selects the other human user profile, which has descriptive information indicating that it is the admin's spouse. The admin selects to enable voice command responses for this user, and adds a phone number, to which the system sends an SMS to verify the number. The SMS requests for the user to click on a link in the SMS while on the premises of the sensor network. The recipient reads the instructions and clicks on the link. The system receives the request for the webpage and also obtains identifiers associated with the device making the request. As the system receives these identifiers, it verifies that these correspond to one of the features associated with the profile to which the contact information was added. If this succeeds then the phone number is conformed, but if it cannot be verified then the phone number is not confirmed. Once the phone number is confirmed, the user can obtain notifications to the phone, and can also request changes to the privacy profile associated with the associated profile. In some instances, such changes are only made after two or more verifications are performed by the system to determine that the request to change privacy settings are associated with the person associated with the profile. One example method to do so is to request the user to speak words specified by the system, as a challenge, and verify that the correct words were spoken using standard methods; then verify that the FFT associated with the spoken words correspond sufficiently well to the feature of spoken words associated with the profile. A person of skill in the art will recognize that there are other methods to verify the identity of the user by matching the recorded features associated with the profile to inputs the system obtains or requests. In some embodiments, users can assign proxies that can perform privacy selections for them; in others, users are only able to make some selections and not others; in yet others, some users and not able to make any selections at all, but all such selections are made by the admin. For example, an admin may specify that all users that correspond to the class “persons” and the class “non-residents in the home” can be provided with a level-1 personalization, where this causes lights to be turned on automatically when the user enters a room, but not level-2 personalization, which permits access to the services of the system using a voice-driven interface.
The system is preconfigured to apply a first set of rules to observations of predicates. A first aspect of each rule is the qualifications associated with the rule. For example, the qualifications of one rule may specify that the rule can be applied to all registered features, whereas another rule may be qualified to be applied to non-registered features, while a third is specified to be applicable to both registered and non-registered features, but only to those features that correspond to persons. Yet another rule may specify that it is qualified only for robots, whereas a third is only qualified to robots that are not in the presence of persons. Some rules are qualified only to be applied to pets; others to persons that have not been identified as minors; whereas yet others are only applicable to persons that have been identified as adults, and who have been authorized to modify their own settings and/or selected system settings.
A second aspect of a rule is its input. The input in one instance comprises a set of inputs that may be in the form of sensor node outputs, or functions of sensor node outputs, or predicates that may be computed from sensor node outputs, auxiliary inputs from the system and/or third parties. Predicates are further described in conjunction with numerous examples provided elsewhere herein.
It is beneficial for some rules to be stateful, which means that they store a state that may be a function of prior inputs, as described above, and compute an output based on some aspects of the stored state as well as some aspects of a current input. Other rules are stateless, meaning they do not store state. The rules conditionally generate an output. Part of the output is a value (which may be a binary string, an ASCII string, or any other representation of data); and part of it is, in some instances, an indication of an entity to which the value is to be sent. Example parties include a registered user, a law enforcement entity, an alarm company entity, a first responder entity, a third party service provider, etc. Outputs can also be stored in a repository, where they can be processed by the system or other rules.
Consider a first profile comprising at least one feature that has been registered, and one of more features that have not. As the entity corresponding to the feature, which in this example may be a person, moves around then sensor nodes detect at least one of the features associated with the profile. Any feature that has been registered and which has been detected triggers the evaluation of rules that are associated with registered features, whereas any feature that has not been registered and which is detected triggers the evaluation of rules that are associated with unregistered features. There may be rules that are associated with both registered and unregistered features. In an alternative embodiment, the rules are not associated with registered/unregistered features, but instead, with profiles in which there are detected registered/unregistered features. In this embodiment, if there are one or more registered features that are detected, that will trigger the evaluation of rules associated with registered features to be evaluated relative to the profile associated with the one or more detected registered features; similarly for the unregistered features and rules associated with unregistered features. Alternatively, time is divided into quanta, and for each quantum of time, it is determined for what profiles one or more rules should be evaluated, based on what profiles there have been detections of associated features during the time quantum, and based on a match between a descriptor of the feature or profile and the rule, where example descriptors comprise registered, unregistered; person, pet, robot, non-mobile appliance and other such types; and subcategories such as a dog, or a person who has a profile associated with him or her, etc. A person of skill in the art will appreciate that there are other alternative approaches of identifying what rules to evaluate at what time to obtain the same or a similar functionality in which appropriate rules are evaluated on features, profiles or other associated information in response to the output of one or more sensors of the sensor network associated with the system.
By determining that one entity (e.g., an entity with the type person) is in the area of another entity (e.g., an entity with the type non-movable appliance, such as a refrigerator), the system can determine a context of the first entity: he or she is likely to be in a kitchen or pantry, as opposed to a bedroom. Therefore, tracking provides context. Some rules may relate to this context. For example, one rule may relate to registered person entities that are in the kitchen; if this entity has not expressed a wish not to get food delivery coupons, and has not been provided such a coupon within 12 h, or another system-specified time period, then the rule triggers the generation and delivery of a suitable coupon, where suitable may refer to a match to current offers, potential user preferences, historical coupon use, and so on.
There are multiple ways of performing the processes described above. For example, registration of a feature can be done in a broad manner, as described above, where the registration applies to all relevant services. It can also be done relative to services. For example, the registration can be done selectively in the context of only some services. For example, when a user registers one or more features, he or she can be asked whether to enable all types of functionality associated with the detection and tracking of this feature (and potentially other, related features). For example, the user can be asked to select from a menu of uses, where one use may be to automatically disarm the burglar alarm, or a broader scope such as automatically access security services. Another use may relate to provision of coupons, or the generation of a list of recommendations based on observations. The user may also set such privacy preferences at a first time, and then, when installing an app or subscribing to a service, be asked to modify the preferences to enable the associated tracking needed to perform the services. An admin user may also select the registration for entities other than himself or herself, which may initiate a notification to the associated user, when applicable; or may require an associated user to approve the selection. This approval can be expressed by responding to a notification sent by the system to an address such as an email address or a phone number associated with the entity, as described above in a different context. A person of skill in the art would recognize that there are many variations on how the selections or approvals can be made, including approaches involving communication by apps, communication using messaging as described above, communication by voice, etc.; there are also many variations of how the system may optionally verify whether a selection or approval was performed with a big likelihood by the correct entity, e.g., verifying that the context of the action is consistent with the features detected by the system relative to the action. It may also involve multiple such methods. For example, if a user gets a notification by email that his or her privacy settings are set to be changed, the user may approve by performing a voice command such as speaking a phrase indicated in the notification, such as “I agree that my presence will be detected by the system in order to provide me with better physical security such as protection in the context of earthquakes or fires.” This spoken voice may be detected by the system, whether using the device that the message was delivered on or by using one or more of the sensors of the network. The system can then verify that the terms were agreed to by verifying whether the spoken words correspond sufficiently closely to the required words; and perform the verification of the correct entity agreeing by verifying that features associated with the manner in which the person speaks match the detected voice command sufficiently well, such as within a select interval of acceptance.
The services and privacy aspects disclosed herein have been described mostly in the context of a sensor node network that is installed in a residential setting. A person of skill in the art will recognize that the same principles apply to commercial settings such as offices. Moreover, the same principles also apply to publicly accessible spaces, such as malls and movie theatres. A system can be defined specifically to one such environment, or to a combination of these. In the case where a system is defined in the context of multiple areas for which the privacy preferences may differ for some users, the user should preferably be able to identify what services (and therefore, associated rules) the user wishes to receive in these various contexts. For example, a user may be happy to receive personalization in his or her home (e.g., different members of a family having their favorite settings applied as they enter a room) but find it undesirable to expose their preferences in private places. Alternatively, a user may select some services that are desirable to be performed in any location. For example, one user may specify his or her preferred shower temperature, and wish for this to be automatically set both in his or her home and in a hotel. Therefore, the user wishes a system including both her home and hotels to make a determination of her identity (which may be kept from being exposed to individuals or organizations) while still enabling a lookup of the related preference in the context of the user being in a bathroom. Thus, sensors associated with the bathroom may determine features associated with the user, transmit these to a backend associated with the system, where the backend determines what rules apply. One rule may be the determination of a shower temperature, causing the communication of this information to a unit associated with the bathroom. The system may do this whether the user has stepped in to a shower or not, or in response to the shower being turned on. As soon as the water temperature information is received, it is applied to the local system, e.g., the controller of the water temperature or the mix of hot and cold water. The system may infer some of the preferences automatically by repeated observation. For example, a user that always selects a temperature of the shower within a narrow range may be seen as implicitly stating a preference. The system may apply the enforcement of this preference automatically, or may communicate an inquiry to the user to establish whether the user wants to apply the preference in the form of a new rule. The user can also be asked about the scope of provision of the related service, e.g., only at home or anywhere, or according to some other relevant specification. The user, of course, will want to have an override operation to change the temperature of the shower in a specific instance, e.g., by interacting with a local controller. If this is repeatedly performed, the system may automatically modify the preference or inquire whether the user wishes for such a change to be made.
Illustrative embodiments disclosed herein relate to several types of privacy, including but not limited to:
1. Determination of what services to provide to what entities, such as users, and under what conditions.
2. Determination of what information to provide to third parties about entities, and under what conditions, where the information relates to the detection of one or more features during one or more time periods in one or more locations.
3. Determination of what features to group together and provide descriptions of in privacy related requests. Whereas this may not be immediately clear to be a matter that may disclose sensitive information, it is. For example, consider a first user who has a body implant that can be detected by the system, and a second user who is asked whether this detected body implant is associated with him or her; this is an invasion of the first user's privacy.
4. Determination of what users may provide configuration information related to other users and their associated features and privacy settings, and the extent to which the users whose associated features and privacy settings are modified or set are notified or asked to approve the modifications and/or settings. A failure to properly manage this is a failure of privacy as it would cause unauthorized users to change a user's privacy settings.
Further, the disclosure provides a solution to a problem relating to notification of users, where the notification may be relative to a privacy setting, but may also be a notification whose mere transmission or access may be seen as a privacy intrusion. As a concrete example of the latter, consider a user who is associated with a feature such as a Bluetooth Device Identifier, and which is also associated with a contact channel such as a phone number. The association between these two is not typically publicly known. This user may be in the proximity of a sensor that detects a feature related to the user, such as the Bluetooth Device Identifier in this example. An entity associated with the sensor may wish to provide a service to the user, but does not have, a priori, a way of contacting the user, whether to offer the service or to request permission to provide the service. However, by interacting with a server associated with the system disclosed herein, a lookup can be performed of the feature, and a determination of the associated communication channel, which in this example is a phone number. The entity wishing to provide a service can request to receive information related to the communication channel. Alternatively, the entity wishing to provide a service can provide the system with information used to convey a message to the user using the communications channel. For example, the information may be the message to be sent, which may comprise a user-friendly explanation and a hyperlink. The system determines, based on privacy settings associated with the user whether the transmission if this message to the user over the communications channel is desirable by the user, and if so, transmits the message. A user may specify conditions for which such messages are allowed to be conveyed, where one example condition is that only one class of entities such as restaurants may be allowed to transmit information, and the only information that is allowed to be transmitted is discount coupons; furthermore, the user may specify the times of the day that such messages can be sent. Another example condition is that the message can only be transmitted if the entity wishing to transmit it pays a fee, such as ten cents, where this fee may be optionally refundable if the user chooses to, but otherwise be credited to a financial account associated with the user. This latter condition strives to control the privacy of the user with respect to unwanted messaging. Another example condition may specify information associated with the user that the user is willing to be used for determining whether the system should transmit the message. For example, this condition may specify that the user is male and 25-30 years old. An entity wishing to convey messages to a male between 25-30 years old (which is targeting information) can now identify this to the system along with the feature and a message or associated information, where the system determines whether to transmit the message to the user over the associated communications channel based on the targeting information and the information specified by the user. This can be combined with other conditions, such as conditions specifying a fee to be paid or conditionally paid by the entity wishing to convey a message to the user receiving the message. In addition, the system may apply additional conditions, including determination of identity of an entity wishing to transmit a message, additional fees that this party must pay to the system for the message to be transmitted, the time and context of the request for transmission, etc. This permits an effective system for notifications and advertisem*nts that respects user privacy. The provision of requests and advertisem*nts is not limited to contexts in which the entity making the request is associated with a sensor that detects a feature associated with the user to which the transmission is to be performed. In particular, the entity wishing to transmit a message may convey information to the system regarding the desired targeting, with the system making a determination of what user or users to transmit the message to, based on evaluation of rules that specify the conditions under which this is desirable or acceptable.
One aspect of the disclosed technology is a method to balance the need to maintain low error rates (e.g., misclassification) and the need to address resource constraints such as battery limitations. Consider, for concreteness an example profile corresponding to one user, where the profile comprises five different types of detected features:
1. WiFi associated with a phone
2. Bluetooth associated with the same phone
3. Bluetooth associated with a wearable computer in the shape of a ring
4. WiFi associated with a tablet computer
5. Voice features associated with the user
Here, the phone is almost always used by one and the same user, corresponding to the user associated with the profile. Assume for concreteness that 98% of the time it is used, it is used by this user (the “owner”). However, it is sometimes not used; e.g., it is being charged during the night and parts of the evening. The phone has battery constraints, but they are not severe.
The ring is always used by the same person, the “owner” corresponding to the profile. The ring, furthermore, is almost always worn, except when it is being charged. The ring has severe battery constraints.
The tablet is used by the “owner” 60% of the time. It is being used (including being carried around) only a small portion of the time, e.g., 10% of the day. The tablet does not have any noteworthy battery constraints in the context of the resources sometimes used by devices to facilitate tracking.
The voice features have a 10% probability of being associated with the wrong person, e.g., somebody with a somewhat similar voice as the “owner.” There are, of course, no power constraints.
It is common that a person is associated with more than one feature at any one time, or during a short time period. A tracking strategy aimed at minimizing the risk of failing to detect a user would therefore attempt to track all these five features at the same time, independently of each other, and perform an assessment of the likelihood that the determination of identity is correct based on determining which ones of the five features were detected during one time period. Moreover, it is beneficial to verify whether this identity determination is consistent over time, e.g., for two or more subsequent time periods. If it is, then that further increases the likelihood that the determination is correct. If there are no signals associated with some of the features, the system may, instead of simply passively listening for signals, actively probe for some of the features. For example, the system may transmit a WiFi or Bluetooth request (such as a notification or a ping on an established channel) to the phone to determine if the phone is present. This can be done for multiple features. However, this “greedy” method is not very conscious of the resource constraints of the devices, as actively interacting causes a use of the device resources. Therefore, while the benefit of a match with the Bluetooth identifier of the ring is substantial (causing a significant reduction of the risk for misclassification), there is a real cost associated with doing it.
A more resource aware approach is to use all signals received in a passive mode of interaction (e.g., simply listening in to signals), and based on a time series of such signals or sets of signals, make an assessment of identity and an assessment of the accuracy of the identity assessment. If the assessment of accuracy is below a threshold, meaning there is a higher risk of misidentification than some measure, then the system may actively probe for one or more features. The system may determine whether a response by the tablet would result in a sufficient accuracy (based on the comparison with the threshold); and if so, based on the likelihood of the tablet being present (which in turn is possible to assess based on previous data) determine whether to send a signal to the tablet. Similarly, the system can determine what set(s) of features to probe to obtain sufficient accuracy; the likelihood of success (a measure associated with the probability of the associated entities being present, which can be determined based on historical data), and what the cost is in terms of resources. The latter can be assessed using a cost function that assigns a cost to each battery level, as applicable, and potentially also related to the assessed sizes of various batteries as measured by time to empty given typical usage. Note that voice features can be probed, but carry a high “user bother” cost. One way to do it is to request for a user in the room to state his or her name, and then determine the identity based both on the spoken name and on the voice characteristics. This should typically be left as a last resort, very much related to a step-up authentication in traditional systems, as it is user invasive. The system may generate zero or more probes, where each probe is followed by an assessment such as the one described above in which the probabilities of presence, probabilities of correct classification, and costs, are weighed against each other, and where a selection of an action is performed based on one or more such probes, zero or more passive observations, and zero or more historical determinations of identity and their underlying data used to come to these determinations. The system can also value power to different extents based on the urgency of a determination. For example, for a personalization system where lights are to be turned on as a person enters a room, and set in a way that corresponds to the preferences of the person, there is not a big cost associated with a misclassification. However, if there is a fire breaking out in a building, it is critical to determine the location of each person and important to determine the location of each pet, but not very important at all to determine the location of cleaning robots to mention one example.
One beneficial use of the system is to perform a series of determinations of identity based on passive observations, with a small number of active observations (e.g., probes) used to bring up the accuracy every once in a while, as needed based on some classification criteria; these criteria may also depend on the urgency of the situation, and similarly, on the business importance of correct classifications. A user recognizing that registering many features will lead to better precision and lower battery consumption may want to register multiple features. Tradeoffs may also be informed by the time of the day, as that will typically determine when a user is likely to recharge devices. For example, a ring with only 15% charge at 8 am in the morning may be considered seriously depleted in terms of power resources, whereas the same ring with 10% charge at 10 pm may be considered well charged, for a user who normally charges the ring at 11 pm. Thus, historical patterns may also preferably dictate the determination of how to perform the tracking. The duration of the time period also is an important parameter. When there is a higher risk of misclassification (e.g., many different users present) then the intervals need to be shorter to establish the same level of confidence as for longer time periods and fewer or less mobile users. Also, the more the users interact and/or use one another's devices, the shorter the time period is needed to establish the same accuracy. Improved accuracy can also be obtained by performing one or more probes, as described above. A person of skill in the art will recognize that this is an illustrative example only, and that many other types of trade-offs exist and can be balanced against each other using the principles explained herein.
In one embodiment, the system is constantly assessing its accuracy of identification so that it can build up a history of the percentage of time that it is assessing that it's classifications are correct. For instance, each time a device is in range, the system makes an assessment of who it is associated with at that instant. In instances where it later determines it was wrong, it logs the mistake. When it later determines it is correct—thanks to other validating information, it logs the correct assessment. When there is no validation, it logs the result as uncertain. The ongoing accuracy of its guesses can be used in the future to assess its confidence, and to provide an automated feedback loop to constantly improve its identification models.
The following are some illustrative examples of rules and services:
The rule identifies a state that corresponds to the burglar alarm being automatically disarmed when a registered user is detected, based on features detected by one or more sensors associated with one or more network nodes. When such a feature is detected, the system performs a lookup to determine whether the associated profile corresponds to a user that has been granted the ability to have the burglar alarm automatically disarmed. This is separate from the registration status value associated with the detected feature or associated profile. For example, a person who has been registered, e.g., for purposes of automatically turning on or off the lights, may not be authorized to enter the building without being accompanied by a person with those rights. Depending on the configuration, a match may cause the disarming, or it may cause a transmission of a message to the user associated with the matched profile requesting a verification, e.g., by voice or fingerprint biometrics. We note that although this relates to one rule related to automatically disarming a burglar alarm, there may be multiple rules relating to this service need, and these rules may work either independently of each other or in a way that use each other's output, or a combination of these. Also, the example is non-limiting, and only describes some aspects of this service, as will be appreciated by a person skilled in the art. Further rules to arm burglar alarms or perform services on either armed or disarmed burglar alarms associated with the disclosed system can be used, and may work in unison, e.g., exchange data and triggering actions across the rule sets. One example of the triggering of rules to evaluate is based on the output of one or more rules of another set; similarly, one or more rules can be used to untrigger the evaluation of a set of rules, i.e., cause previously triggering of evaluation of rules to no longer remain in effect, until triggered again.
Some traditional systems automatically turn on lights as a person enters a room, and then automatically turn the lights off after a time has passed from when the person was last detected as being in the room. This is not always accurate, and can cause lights to be turned on by large pets, for example. Also, it can cause the lights to be turned off even when a person is still in the room, e.g., if the person is sitting very still. A person watching TV or speaking on the phone may therefore experience lights being automatically turned off every once in a while.
Instead a system that is based on detection of one or more features, as described herein, can be used to determine whether an entity is present, and if so, whether that corresponds to a person. This may be performed, for example, using the type of one or more detected features. It is beneficial to use multiple features for this determination, as described in other places in this disclosure. For example, a person sitting on a couch in a TV room may accidentally drop his or her phone in the couch. Any feature associated with the phone would therefore remain associated with the TV room even after the person leaves. Therefore, using additional features strengthens the system against such misclassification. Consider, for example, a system based on detecting the location of a person based on two features, such as a feature related to a phone and a feature related to the voice characteristics of the person. Further features are beneficial to use, but for simplicity, we will only consider these two herein. In a context where a person drops his or her phone in the couch and then leaves the room, and where the system detects the presence of the person elsewhere, e.g., based on the person speaking to somebody in the kitchen, the system can determine that the phone, while still associated with the person, is no longer in the presence of the person. Therefore, the presence detector should no longer use features associated with the phone to determine the location of the person, until the system can determine a likelihood of the phone being with the person again. This can be done at different standards of accuracy. For example, consider a case where the phone is detected to be moving from the previous location (the TV room) to another location (such as the hallway outside the TV room). The system can determine that it is somewhat likely that the phone has been reunited with the person it is associated with. However, it is possible that another person moved the phone. Therefore, a higher standard (corresponding to a higher degree of confidence) can be achieved if the system determines that the voice-related feature is detected in the hallway (for example) as the phone-related feature is also detected to be located in the hallway. Here, the location can be approximated by determination based on what sensor nodes detects signals related to the feature; as well as triangulation methods in which multiple sensor nodes of the node network receive such inputs and determine based on signal strength, attenuation, potential delays, and other methods understood by a person skilled in the art, a likely and more precise location.
Going back to the pet, the system will recognize the type associated with the features related to the pet as being dog features (for example), and therefore not trigger the evaluation of rules such as rules used to control the lights. Other rules can be triggered. If a pet owner wishes to locate a pet remotely, for example, and see footage of the pet, the system can determine the location of the pet and provide footage. This rule can be triggered by a request from an app used by the user to watch photos or videos of the pet.
The system can also be used to locate entities. Consider the example above related to the phone that was accidentally dropped in the couch. A user not knowing where his or her phone is can request for the system to locate it. The system can determine the last location at which one or more features associated with the phone were detected by the system, and identify the location to the user. This interaction can be voice driven, for example, allowing a user to ask “Where is my phone?” after having activated a voice request feature which can be done by stating a wake-up word as understood by a person skilled in the art, or by performing another action such as pressing and holding a button associated with a light switch. The system, which preferably has classified various locations based on their use, or which has received a classification from a user or admin associated with the system, can provide the location by referring to the functionality of the room, e.g., by transmitting a response “Your phone is in the TV room” to the user. Alternatively, the system may cause signaling in the location, e.g., the TV room, where the signaling may be a sound emitted in the approximate location of the lost phone. Alternatively, the system can determine the approximate location of the user making the request, and based on this, generate a response. A user who is not at home would likely not be able to hear a sound signal emitted in the TV room. Therefore a simple response such as the address of the location or a label such as “at your home” would be a useful response. The user can ask for additional detail, and could learn “your phone was last spotted in your basem*nt, potentially in your TV room.” The system may provide additional information, such as “The last time the phone was detected was approximately three hours ago. The phone may be out of battery.” Alternatively, in some cases wherein the system is able to send signals to the device being searched for, it can do that. However, this is not applicable to all forms of devices, and is not applicable to devices that cannot be communicated with by the system.
In the context of the location of the phone, we note that this has important privacy implications, and it is not always desirable for any user to be able to locate any entity. It is, most commonly, appropriate for a user to locate an entity associated with one or more features that belong to the same profile as that which is used to describe the user in question (in other words, users should be able to locate their own devices, their own pets, etc.). Therefore, the location service corresponds to a rule that cannot be evaluated without determination of who is initiating a location request, and a comparison of the features of this person to the profile including features of the entity to be located. However, a user may permit a second user to perform such actions. For example, a user may enable his or her spouse to locate the user's phone, but only in contexts where the phone does not appear to be co-located with the user himself or herself, where the co-location is determined by identifying the location associated with two or more features, where one feature is associated with the entity to be located (such as the phone) and the other is a feature associated with the user associated with the phone (e.g., the owner of the phone). If these appear to not be located in the same approximate location, then the rules will permit the location of the entity (e.g., the phone) to an authorized party (e.g., the spouse); another rule may use another mechanism for determining when to permit tracking, and multiple rules, whether related to the same type of service or not, can be evaluated in the response to the same event, where the event may be a request of a user to locate an entity, as in this example. Authorization as described in this context may be performed based on detection of features (e.g., determining that the request is likely to come from the spouse) or based on a rule (e.g., a person in a user's home), or a combination of such conditions (e.g., limited to family members when in the user's home). These examples, as will be understood by a person skilled in the art, are non-limiting and included to describe the uses of the disclosed system.
The illustrative embodiments shown in FIGS. 23 through 30 will now be described in more detail. The particular arrangements of devices and their operation as illustrated in these figures should not be construed as limiting in any way.
FIG. 23 shows an example association process. In step 2301, a user device 2401 is detected by a control system 2410 and a feature 2501 is created and stored by control system 2410 in step 2302. In step 2303, a user installs an app 2602 on a user device 2402, where the app comprises a storage area storing data 2622, which may include unique information such as an app identifier, an HTML cookie set by control system 2410, etc. In step 2304, the user provides data 2622, configuration information 2623 and identifying information 2624, not necessarily in that order, to the control system 2410 using the app 2602 on user device 2402. In response to receiving data 2622, configuration information 2623 and identifying information 2624, again not necessarily in that order, the control system 2410 creates and stores a feature 2502, as indicated in step 2305. For example, feature 2502 may comprise or relate to identifying information 2624 or portions thereof; this may be a MAC address associated with user device 2402. In optional step 2306, the system also creates and stores a feature 2503, which may for example be a friendly name associated with user device 2402. An example friendly name is “Alice's phone.”
In step 2307, the control system 2410 further generates and stores a profile 2510 and associates features 2502 and 2503 with profile 2510. The control system 2410 may further store data 2622, configuration information 2623 and identifying information 2624, or portions thereof, in field 2511 associated with profile 2510.
In step 2308, the control system 2410 determines that feature 2501 is associated with feature 2502 and/or feature 2503, thereby establishing that feature 2501 is likely to be associated with profile 2510. Accordingly, in step 2309, the control system 2410 adds feature 2501 (or a reference to feature 2501, such as a pointer) to profile 2510.
FIG. 24 shows an exemplary simplified system comprising user devices 2401 and 2402, node network 2420 preferably connected to or able to observe user devices 2401 and 2402. It also shows connections to Internet 2430 and control system 2410. User device 2401 may be the same as user device 2402, although this fact may not be known to control system 2410. Also, node network 2420 may be configured to permit access to the Internet 2430 to user devices connected to node network 2420; node network 2420 may also communicate with control system 2410 using Internet 2430.
FIG. 25 shows profile 2510 that is associated with features 2502 and 2503, but before the control system 2410 determines in step 2308 that feature 2501 is associated with profile 2510. Profile 2510 is optionally referencing field 2511 comprising additional information.
FIG. 26 shows user device 2402 communicating over a network with control system 2410, with the app 2602 conveying data 2622 and configuration information 2623 to control system 2410 via user device 2402, and user device 2402 additionally providing identifying information 2624 to control system 2410. An example of data 2622 comprises but is not limited to a unique identifier in the context of user device 2402, account information associated with user device 2402, or information known by both user device 2402 and control system 2410, but preferably not known by other entities not associated with the operator of user device 2402. An example of configuration information 2623 comprises demographic data input by a user; and preference data or other settings related to the provision of service enabled by control system 210. Examples of identifying information 2624 comprise an HTML cookie, a MAC address, a BD_ADDR or a portion thereof, and user agent information associated with a browser executing on user device 2402. These are non-limiting examples and included for the sake of clarity.
FIG. 27 shows data 2622, configuration information 2623 and identifying information 2624 being associated with feature 2701. Features 2701 and 2702 have been determined by control system 2410 to correspond to the same device, and are therefore both associated with device record 2710. Device record 2710 (and therefore also features 2701 and 2702) are associated to the same profile 2510 as feature 2703, indicating that these are all determined to relate to the same user or a role for a significant part of the time. Profile 2510 may optionally contain data useful to evaluate logic associated with control system 2410 to determine the likelihood at any time that these associated features relate to the same end user. Feature 2704 has not been determined to be associated with profile 2510.
FIG. 28 shows a feature 2801 with associated registration data 2802, accuracy data 2803 and probing cost data 2804. Registration data 2802 describes whether the feature has been registered, and if not, whether a user has been requested to make a registration decision. Additional information such as the time of registration, time of request, etc., may also be included. Accuracy data 2803 describes the error rate associated with the feature 2801. This can be a score, a probability, a classification, or other indication of reliability of a determination. Different methods of computing the accuracy score use potentially different qualifying inputs to do so, as illustrated in FIG. 29.
FIG. 29 shows a timeline 2901 with two plots of accuracy scores 2902 and 2903, where accuracy score 2902 is computed using a first error rate method and accuracy score 2903 is computed using a second error rate method. Both are shown as numbers ranging from 0 to 1, but they can be other assessments as will be understood by a person skilled in the art. Events 2910, 2911 and 2912 are observed by the system. Events 2910 and 2912 are used to compute accuracy score 2902, whereas all three events are used to compute accuracy score 2903. At time 2920, a service provider wishes to know the pseudonym, identity or profile identifier associated with a context, and selects a metric corresponding to a method of computing an accuracy score. This selection can also be made a priori, during a setup session associated with the service provider. The score value 2921 of accuracy score 2903 at time 2920 is about 0.75, whereas the score value 2922 of accuracy score 2902 at time 2920 is about 0.38.
FIG. 30 shows an example user interface for permission requests displayed on screen 3000 that may be associated with a device such as the device referenced in first explanation 3010, or an I/O device that is not tracked but part of a sensor network. First explanation 3010 relates to one or more features of a first device. Second explanation 3020 relates to one or more features of a second device. Third explanation 3030 relates to one or more features associated with a person. First explanation 3010 is associated with exemplary user selection buttons 3011, 3012 and 3013, corresponding to possible user responses of Yes, No and More Information, respectively. Similar buttons exist for second explanation 3020 and third explanation 3030. Third explanation is also associated with an option 3034 allowing a user to play samples associated with third explanation 3030. User selection 3041 can be clicked to see additional features that can be tracked. Alternative user selection methods are also possible, e.g., scrolling to see more and swiping left on an explanation to approve a request, or swiping right to deny a request, etc. User selection 3042 can be clicked for a user to see already approved tracking, and to edit the permissions for associated features.
Like other embodiments disclosed herein, the embodiments described above in conjunction with FIGS. 23 through 30 are for purposes of illustration only. Numerous other arrangements and associated operating scenarios are possible in other embodiments.
In some embodiments, methods are provided for complying with a need to obtain permission by users for a sensor node network by observing features useful for tracking of users, objects and pets, etc.; determining a representative description of features; associating features with a user communication channel; presenting a request using the user communication channel, where the representative description is optionally used to convey the request; receiving a response to the request; and based on the response, associating a registration status value with an associated feature and storing the registration value in a profile; where an action is taken based on an observation of a feature, a registration value and at least one rule associated with the sensor node network, and wherein the action may optionally depend on information associated with at least one profile.
In some embodiments, systems are provided for performing privacy related processing of information. A given such system illustratively comprises at least one unit configured to obtain a first feature; at least one unit configured to obtain a second feature; and at least one processor. The at least one processor is configured for receiving information about the first feature and the second feature; determining whether these are associated according to one or more criteria relating to one or more threshold values; and conditionally, based at least in part on the determination, presenting a request to an interface unit, where the request relates at least in part to the first feature, and where the interface unit is physically separate from the entity corresponding with the first feature. For example, the request in some embodiments of this type is illustratively delivered to a device other than the device that is detected, based on a clustering of at least features. This is in contrast to cookie-related privacy requests that are always presented on devices that are the same as the ones to which the cookie applies.
Additional illustrative embodiments will now be described in further detail with reference to FIGS. 31 through 35. These embodiments are illustratively configured to implement various aspects of infrastructure support to enhance resource-constrained device capabilities. Such embodiments in some implementations utilize one or more aspects of device location services or privacy enhancement using derived data disclosure as described elsewhere herein, and can additionally or alternatively implement other types of complementary functionality, such as functionality relating to determining preferences and events and generating associated outreach therefrom.
A person with a wearable computing device, such as a Fitbit™ or an Apple Watch™, may use the device to monitor her health or to get alerts when her activity is anomalous. The wearable computing devices determine the activity of the user by performing computations on sensor data of sensors associated with the wearable computing devices. For example, using an accelerometer, the devices can determine the likely type of movement, and the amount of the corresponding activity. For example, this can be used to determine the number of steps a person has taken during a given day. If a user suddenly falls below a threshold of activity, or an undesirable event such as a fall is observed, an alarm may go off, whether on the wearable device or on a backend device that in turn can be associated with one or more computers to which alert notifications are sent. However, if the interpretation of the events is flawed, this may result in either false positives (i.e., false alarms) or false negatives (i.e., the absence of an alarm that should have been initiated).
Misclassification is not unusual, and is commonly the result of insufficient information. For example, a person wearing a Fitbit™ or an Apple Watch™ may be sitting still for an entire afternoon, knitting a sweater, while the device determines, based on accelerometer data, that the user is taking a very long walk. This misclassification could easily have been avoided by using Global Positioning System (GPS) data to determine that the movement of the device does not result in the user moving from one location to another. However, GPS location determination is well understood to consume a lot of power, thereby potentially draining the battery of the wearable device. Some wearable devices address this problem by using the GPS on a smartphone that is associated with the wearable device. This requires an application (“app”) to be installed on the smartphone, where the application can communicate with the wearable device, e.g., using Bluetooth or Bluetooth Low Energy (BLE), and is also capable of requesting GPS location data from the host smartphone. However, this is a very partial remedy, as it is limited to devices that are paired; requires an application to be installed; and still, causes a significant drainage of power. The latter is because the smartphone, too, is a resource-constrained device, although not as severely constrained as the wearable device.
Another limitation relates to devices having limited sets of sensors. A wearable device with an inertial sensor, commonly including a gyroscope and an accelerometer, can sometimes not distinguish between the user falling and the user slamming his hand on the table or sitting down in a rapid manner. However, sensors in the surroundings of the user are capable of making this distinction. For example, inertial sensors mounted close to the floor will be able to distinguish a slam on the table or a rapid sitting down from a fall. These sensors would also be able to detect other types of events. For example, a person falling down is a locally observable event, whereas an earthquake would be an event that can be observed in every room of a house. Thus, the limitations in terms of the type and/or number of sensors of mobile devices, including both wearable devices and smartphones, cause false positives and false negatives alike, and result in misclassifications of important events.
Yet another problem with resource-constrained devices is that many such devices generate activity logs, and if these are not periodically synchronized, then the limitations in terms of storage cause old activity log entries to be overwritten. This either results in user inconvenience in the form of a need to synchronize, or occasional loss of data. Moreover, it requires the users to travel with the devices, such as smartphones, that the wearable devices are associated with, or the logs cannot be synchronized.
A further problem is that time-sensitive alerts can only be transmitted to a backend in charge of obtaining help if the wearable device is equipped with a cellular modem and within range of cell phone towers, or is in the presence of a paired smartphone that can convey alerts or other data streams using cellular communications or other methods of accessing the Internet. This means that if the smartphone is not charged, is not turned on, is not within radio distance of a typical constrained wearable computing device, or is not capable of connecting to the backend, then alerts will not be reported in a timely manner.
Thus, the functionality of resource-constrained devices of various types is stunted by battery limitations; form factor or cost considerations causing a limited set of sensors to be incorporated; limitations in terms of capability to interpret limited sensor data; limitations in terms of connectivity; and numerous others. The problems are not addressed by pairing resource-constrained devices such as wearable computing devices to smartphones or tablets, as these devices also suffer similar limitations and constraints, and moreover, can be separated from the wearable computing device or otherwise made inoperable. In fact, mobile devices such as smartphones suffer very much the same types of limitations as wearable computing devices do, only less acutely. The somewhat lesser degree of the problem is due to the larger size of smartphones and their batteries, and their greater number of sensors. However, this increase in size and components incur costs and inconveniences, and it would be beneficial for such devices to enjoy longer battery life, lower production costs and smaller size. Such improvements may be made possible if the smartphones could offload some capability to other devices with yet lesser constraints. This is not possible with today's technology.
One aspect of the disclosure is an advertising of resources. For example, one entity may advertise that it has the capability of location determination, and another may advertise that it has the capability of Internet connectivity. An advertisem*nt may comprise a priority rating. For example, an electric outlet may have been configured to know the GPS location it is in and convey this to nearby devices. Since the electric outlet is in a static position, and it is not powered by a battery, there is no need for costly operations for it to be able to determine and share its GPS location. Thus, it may indicate that its priority is 10, assuming by way of example only that the priority values are on a scale from 1 to 10, with 1 indicating the lowest priority and 10 indicating the highest priority. Other types of values and scaling arrangements can be used in other embodiments, such as an inversion of the low to high priority value scale.
A smartphone may have the capability of determining GPS location as well. However, the smartphone, being mobile, needs to periodically determine its GPS location. Moreover, the smartphone, being battery powered, has a cost associated with determining the GPS location. Therefore, the smartphone may set the priority value to 5. As the battery resources of the smartphone falls below 25%, the priority value is changed to 4, and as it falls below 10%, it is changed to 1.
A smart watch may also be able to determine GPS location. However, the battery of a smart watch is more constrained than that of a smartphone; therefore, the priority may be 2 for the smart watch when its battery is above 75% full, and then changing to 1 as it falls below 75%, and to “unavailable” (which may be expressed as 0) when it falls below 25%. Similarly, when the resource corresponds to Internet access, the priority may correspond to a number such as 10 for an electric outlet that is both powered and has a wired Internet connection, whereas it may be set to 6 for a smartphone with a WiFi connection and 3 for a smartphone using a 2-bar cellular connection. The status of an associated subscriber's monthly data allotment and the day of the billing period may be considerations affecting the priority value as well. The priority indicates, among other things, the “willingness” of an entity to help another device.
In addition to priority, an entity may advertise the quality level of a resource. In terms of location determination, the quality may correspond to the time since the most recent determination. In some embodiments, the quality values, like the priority values, are also on a scale from 1 to 10, with 1 indicating the lowest quality and 10 indicating the highest quality, although again other types of values and scaling arrangements can be used.
A phone that performed a GPS determination 5 minutes ago, and which has detected rather limited motion using its accelerometer may advertise a high quality level, such as 8, whereas a phone that performed a GPS determination 15 minutes ago and which as detected moderate motion using its accelerometer since the last GPS determination may advertise a quality level of 6. In contrast, the smart outlet that is not mobile advertises a quality level of 10 for its GPS location, since it most likely has not moved at all since its GPS location was determined. (This is not applicable to some outlets, such as one mounted in a motorhome, but to most outlets; their nature can be determined in a configuration phase or by interaction with other devices, as described below.)
Quality level also applies to other types of services. For example, when the service is Internet connectivity, the quality level may correspond to the bandwidth, the estimated roundtrip time or another aspect of a typical communication link. If the service is not general Internet connectivity, but just access to an Intranet, e.g., using WiFi, the quality level may instead be an indication of the network load, e.g., whether there are lots of collisions or not. Thus, for each type of service, the quality level may have a slightly different meaning. An advertisem*nt may correspond to multiple services, and associated priorities and quality levels. The quality level can also be seen as an indication of reliability (e.g., for network communication) or accuracy (e.g., for GPS location determination); such indicators can alternatively be added in addition to the quality level, or the system can use a quality level that comprises multiple numbers, each number representing one aspect of quality. For example, a network connection can have speed, bandwidth, availability, and error rates as aspects of a quality measurement.
Nodes may advertise their services periodically; in response to observing general activity such as network activity, motion, the turning on or off of lights, sounds, etc.; or in response to a service request. A device may generate a service request in response to a computational process having a need (such as the need to transmit data, synchronize data with another entity, determining location, obtaining environmental data, etc.) A service request may take the form of an identity associated with the requesting device, an identity of a user associated with the device, a description of the service or services needed, and an optional minimum requirement in terms of quality level. A response to a service request may be conditional on the identity of the requested device or associated user, what services are needed, and additional conditions such as minimal quality needs. A service advertisem*nt may contain the identity of the entity advertising the service(s), one or more priority levels, one or more quality levels, and information about access constraints. For example, some service providers may offer services to any device; others only to devices that have been pre-registered with the service provider or an associated entity; or only to devices that have a particular certification or membership. Different services may be associated with different constraints.
Service requests and/or service advertisem*nts may be transmitted in cleartext, encrypted, authenticated, or a combination of these. Services can advertise a capability that they do not currently have, e.g., the capability to transmit data packets to another entity. This may be indicated using an availability value. A device may indicate a general ability to do so, but the absence of this ability at the time of the advertisem*nt by setting the availability to 0. A service requester may transmit the packets to one or more entities with the capability but not the ability to transmit the packets to their destination at the time, as a way to optimistically obtain a service. The final recipient of the packets may signal back in a similar way that the packets have been received in order for the service requester to know that they were received.
Some devices may both be providers of services and consumer of services, whether for the same or different service. For example, a smartphone may be a consumer of GPS location information, provided by a smart outlet or smart light switch at the same time as it may be the provider of GPS location information to a wearable computer such as a smart ring or a smart watch. When the smartphone provides the service of GPS location to such a device, it may use GPS data that it produced itself or it may use GPS data that it received from another entity. In the latter case, it may modify the quality description of the GPS data to take into consideration that the measurement has aged for five minutes since it was received by the smartphone, but that the smartphone has been largely sedentary since then; this particular situation may correspond to a decrease of the quality level by one unit.
A stationary device that does not have a GPS sensor may request GPS location information from one or more devices in the neighborhood, and use the responses to determine a location that it later serves other devices. For example, an outlet may determine that it does not have a location, e.g., after first installation, and then request a location or listen for advertisem*nts for location. After receiving the location it may store this data in a non-volatile storage, such as an electrically erasable programmable read-only memory (EEPROM) or a flash unit. The outlet may occasionally request a location from nearby service providers offering location information, and use such information to determine whether it is likely that it has been relocated; if so, the stored location may be erased, replaced, invalidated, or its quality level may be degraded, e.g., from 10 to 4, where the extent of the degradation could be a function of the quality of the data used to invalidate it. If the quality level comprises multiple metrics, such as a speed aspect and a reliability aspect, the latter is the aspect that would be degraded. Alternatively, instead of using a quality measure, the system may use several measures, such as multiple ones of the measures described herein that can comprise a quality measure.
As one example use case, in a busy train compartment, there may be multiple devices running applications that consume GPS data. One of the devices may obtain an updated GPS location reading, whether using a GPS sensor or from another node, and then advertise this to other devices, which would update their information provided the quality of the new information exceeds the quality of the stored information. This way, a new GPS location may cascade through the train on a periodic basis. In addition to priority and quality, freshness may be advertised, or this may be an aspect of the quality. The freshness is an indication of the time since the underlying measurement; the freshness may alternatively be expressed as the time of the underlying measurement. A device may choose to replace an older measurement with a newer measurement even if the quality level of the newer one is the same or slightly lower. Freshness can be seen as an indication of reliability, which can be an aspect of a quality value, or be used instead of or in addition to a quality measure.
In some contexts, the provision of service may be for free, whereas in other contexts it may require a payment, a subscription or a membership. Example payments include micropayments, such as those described in, for example, S. Micali et al., “Micropayments revisited” and M. Jakobsson et al., “A Micro-Payment Scheme Encouraging Collaboration in Multi-Hop Cellular Networks,” both incorporated by reference herein. Another example payment method is a crypto payment scheme, such as BitCoin.
Subscriptions and memberships may use digital certificates or other authentication methods as a way to indicate whether a device is permitted to gain access to a service. Such methods may also be used to determine that a requester or service provider is compliant with a set of requirements. Example requirements comprise compliance with security standards, certification of an original equipment manufacturer (OEM), the use of technical solutions such as digital rights management (DRM) solutions, and user agreement to complying with usage agreements, e.g., limiting the extent to which a given type of service may be used, limiting the ratio between receiving and providing services, and more.
A device may receive the same or similar services from multiple service providers during a limited time period. For example, a device may request the location from several service providers; the associated quality levels may be the same or not. The requester can combine the different location assessments and generate a more precise or higher quality location assessment from this. It can then act as a service provider and offer this improved location assessment to other devices, including the devices from which it received location information. A device can similarly use multiple service providers to transmit or receive data, whether in the manner described above wherein no service providers have an ability at the time of the transmission, or by dividing a transmission into sets of packets and requesting to have different sets transmitted by different service providers. This can be used to increase bandwidth. For example, a device that has a Bluetooth radio that is distinct from its WiFi radio can transmit or receive using both of these at the same time in order to increase the bandwidth.
One authentication method that requesters and service providers may use is a traditional digital signature. In the case of nodes that have established pairwise connections, a symmetric authentication method such as a message authentication code (MAC) may also be used. Hash chain based authentication methods, such as that described in A. Perrig et al., “The TESLA Broadcast Authentication Protocol,” CryptoBytes, 5:2, Summer/Fall 2002, pp. 2-13, which is incorporated by reference herein, can also be used. Other methods include the use of fractal hash chain traversal, as described in D. Coppersmith et al., “Almost optimal hash sequence traversal,” Proceedings of the Fourth Conference on Financial Cryptography (FC '02), which is also incorporated by reference herein. Hybrid methods relying both on digital signatures and hash chains may also be used for authentication. One example combination is the use of an RSA digital signature and a hash chain, where a first time verification requires verifying both the RSA digital signature and at least a portion of the hash chain, but where consecutive verifications only require the verification of the hash chain component.
In some embodiments, no explicit authentication is used. For example, if a message related to a request or a service provision is communicated over a channel that has some form of built-in authentication or limitation in terms of ability to use the channel, the recipient may determine that a correctly formed message that was received must be authentic. In some embodiments, keyed error correction codes or other methods providing limited authentication guarantees may be used. In some embodiments, no authentication is used for at least some of the messages relating to requests, responses, advertisem*nts, and service provision messages in which data is communicated.
Some messages may, further, be encrypted. In one approach, the encryption is provided by the communication medium, e.g., for Bluetooth communication. In another approach, encryption is provided as an end-to-end manner, not involving the service provider in the encrypting or decrypting of messages. In yet another approach, the service that is provided comprises encryption. For example, a service provider device with the capability of performing costly public-key encryption may receive unencrypted or symmetrically encrypted messages from a message originator that corresponds to the service requester, decrypt these where applicable, and then encrypt the data using the public keys of one or more recipients. This is an example of a service that comprises computation, and also, an example where it is clear that the requester will not want to receive service from untrusted service providers. Thus, in such a situation, it is beneficial for the channel between the requester and service provider to be encrypted and/or authenticated, using low-cost cryptographic methods such as symmetric methods. In situations where the exposure of sensitive data is not likely between a requester and a service provider, such as where the channel is implemented using NFC, no local security may be required. The security needs may be negotiated between the requester and service provider during a setup step that may comprise a service request message, a response message, a message for advertising of services, or a message following one of these types of messages.
One beneficial service that can be provided by the disclosed system is a ubiquitous synchronization of devices. Consider two devices, such as a health tracking device and a smartphone, where the health tracking device makes measurements and periodically transmits values representing at least some of the measurements to the smartphone. The smartphone can then analyze the health data, provide recommendations and warnings, as well as provide updates to the health tracking device, such as firmware updates or weights that are used to generate the values representing the measurements in a manner that optimizes the conveyance of measurement data of greatest importance to the health of the user. However, the communication between the health tracking device and the smartphone only takes place during some conditions, which may not always be satisfied. For example, the two devices must be in each other's presence, given that the typical method of communication is BLE, which has a very limited range. The greater the distance, the more energy is consumed to perform the synchronization. If the device receiving the signal from the health tracker had a very strong antenna, which the smartphone may not have, and lower constraints on the powering of this antenna, then the bandwidth can be increased while the relevant power consumption is limited.
Moreover, given multiple entities related to infrastructure devices or other devices in the presence of the health tracking device, there would be a greater coverage of synchronization capabilities, which enables more timely receipt of information of relevance to alerts. This has a significant health benefit. Not having to be in the presence of the smartphone is also a convenience for the user, who may wish to have the health tracking device synchronize with a node on his or her bed stand every ten minutes throughout the night, using the measurements to adjust the temperature of the room to improve the user's sleep. This type of benefit would not be available to a user who did not sleep next to his or her phone, which many users may not like to do. The disclosed technology allows the synchronization of devices, such as health tracking devices, using nearby service providing devices as proxies, whether the synchronization is performed with the smartphone, a backend server providing analysis and alert services, or a combination of these. By using a paired phone, a user is enabled to indicate what service providers or types of service providers the paired health tracking device can use, and initiate the establishment of service provision to such devices.
The system benefits from the use of certificates to be used to provide assurance of compliance with desired behavior. These certificates may have an expiration date, after which they have to be renewed. The renewal can be performed by the security system or by a free-standing entity in charge of evaluating the behavior of entities. A certificate may be renewed unless there is significant negative feedback related to the associated device, where negative feedback is preferably automatically generated by a device receiving a service from one or more service providers, and where the device receiving services identifies discrepancies or anomalies and report these to the system. The report may also comprise a location, whether implicitly and in the form of an IP address or explicitly in the form of a GPS address. Alternatively, the system may know the location of some of the devices of the report, e.g., due to a configuration phase in which the location of a non-mobile device is transmitted to the backend from the configured device or a proxy of this. A report may comprise of a set of identities and associated service types, a time of service provision, and the identity of the entity generating the report; the report may optionally be digitally signed by the reporting device, or otherwise authenticated, e.g., using a MAC for which the system has access to the symmetric key used to generate the MAC.
The system (or a free-standing entity of the type described above) receives a multiplicity of reports. It tags reports so that devices that are mentioned in larger-than-normal quantities of reports are suspected of malfunctioning, and such devices are suspected of filing incorrect reports; thus, the reports of suspected devices are tagged as being potentially incorrect. Based on the reports that are not tagged in this manner, the system determines whether some devices are overrepresented among the reported devices. If so, these devices are identified as potentially malfunctioning. Other related techniques for identifying potentially malfunctioning devices based on the reports can also be performed, whether using rule-based techniques, expert systems, machine learning techniques or a combination of such techniques.
The potentially malfunctioning devices, based on the determinations, will not have their certificates automatically renewed, and their identities may be placed on a blacklist that is made available to entities. The blacklist may contain individual device identities, a series of identifiers such as an interval of serial numbers, a type of device, a manufacturer of devices, or a combination of such identifiers. Blacklist updates may be automatically transmitted to devices observed to be in a geographical area that is close to a geographical area associated with entities on the blacklist update. This way, if individual devices are malfunctioning or maliciously modified to generate incorrect advertisem*nts not representing their actual behavior, or to request services that the devices should not request, or otherwise behave incorrectly or in an undesirable manner, then such devices will be excluded by means of their certificates not being renewed, their identities being placed on a blacklist, or a combination of this.
The blacklist is preferably digitally signed by the party generating it; similarly partial blacklist updates are also preferably digitally signed. A device receiving a blacklist may advertise having this blacklist in its possession, and be willing to share it with nearby devices. Blacklists and partial blacklists are preferably time-stamped as well as digitally signed, which enables devices to refer to these lists by their timestamp when advertising them. Alternatively, the devices advertising having such a list can indicate its age using a quality measure, as described for other types of services. A person of skill in the art will recognize that there are many variations of these techniques, including reputation-base techniques, anomaly-based evaluation of reports, and more.
A certificate revocation list (CRL) is a type of blacklist that can be used to identify malfunctioning, malicious or otherwise misbehaving entities, whether these are acting as consumers of services, providers of services, or both. A CRL can be made available in its entirety; in terms of the most recent updated; in terms of geographically relevant entries; and in terms of types of services, and a combination of these. The CRL can be operated by a security service provider that makes it available to any party, to certified parties, to paying subscribers, and according to other policies. The CRL can also be made available as a service, by nodes performing the role of service providers, as described herein. It is desirable that the CRL is digitally signed by a trusted authority and has a date stamp or other freshness indicator. A node only interested in obtaining services related to location may only want access to the CRL related to location. A node that typically does not leave a neighborhood corresponding to two zip codes may only want access to updates related to these two zip codes. If there are no such updates, or a very small number, they may want updates related to nearby zip codes or other geographic areas as well. A device (also called a node) requesting a CRL may specify what types of entries are of interest, e.g., by location and service. A device may also request to receive updates using a push service when it enters a new location for which it does not have the CRL entries yet; when a blacklisted node enters its geographic location, and more. This is a form of subscription that requires one or more service providers to be aware of the preferences associated with the device, and to determine when a new update is needed. Alternatively, the device itself can identify a need for an update, e.g., as it enters an area for which it does not have CRL entries, as it discovers a new device in its network neighborhood, and more.
The disclosed technology can be used to provide a large number of services, the breadth of which has been illustrated by examples. A person of skill in the art will recognize that the techniques disclosed herein can be used for provision of additional services, variants of the illustrated services, and combinations of the illustrated services. For example, the system can be used to enable storage of information, for devices with information to be stored (requesters) by devices with storage facilities or the ability to communicate with storage facilities (service providing devices). The system can be used to distribute anti-virus updates, anti-scam updates, and other software updates, including updates of operating systems, firmware, etc. For example, service providing devices can facilitate access to such updates upon request, or can receive updates from backend service providers and store and later distribute such updates after being requested. The updates may be encrypted to limit access to authorized devices, and authenticated to guarantee, to the requester, that the updates are authentic and therefore safe. The service providing devices can also maintain lists of available updates, obtained from other service providers, but not maintain the associated updates. The lists of available updates can include identifiers of the updates, impact or importance ratings, authenticators, version information, and more.
The disclosed system facilitates trusted connections in a system of vouching. Consider a collection of entities (such as wearable computers, mobile devices, routers, radio-equipped appliances, smart outlets and more). Some of the entities have already established trust relationships with other entities. For example, entity A has determined that entity B is trustworthy; this may have been performed using a combination of determining that B's certificates are signed by a trusted entity, still valid, and not on a blacklist; that B is owned, operated and configured by the same person as A is; that A and B run the same DRM technology; and similar. A may also have a trust relationship with entity C, for similar or other reasons. Entity C establishes a trust relationship with entity D. After doing that, it communicates this fact to entity A, as it interacts with A. This communication is referred to as C vouching for D to A. Since A trusts C, and C trusts D, A determines that D can be trusted. This is a transitive closure of length 2, as D is two steps away from A. A may communicate the fact that it now trusts D when it interacts with B. It also conveys that D is two steps away. B may accept D as trustworthy, as A did, or it may determine that it does not accept transitive closures with distance 3 without additional verification. In the latter case, it may perform an additional verification, such as contacting a CRL service or requesting from a peer device a recent CRL list of relevance.
What actions are sufficient is based on a policy related to the receiving entities, which in this case is entity B. The policy may depend on the accessibility of resources, such as the availability of a network connection to request an update from a CRL service. It may also depend on the resources of the entity, e.g., B may accept the vouching by A of D without any further verification if B has near-depleted battery resources and is in need of the capability associated with D. This may be a time-limited acceptance of D's trustworthiness. The entity B may also accept the entity D as trustworthy after testing D, e.g., by requesting a service both from D an already-trusted entity, and comparing the outcomes; if these match, then B may determine that D is trustworthy. When one entity vouches for another entity to a third entity, the voicing entity preferably authenticates the message, e.g., using a digital signature, a message authentication code, or using other methods allowing the recipient of the information to verify that it came from an entity it trusts.
In the absence of pre-established trust, a receiving device may use a certification of the vouching device as a method to establish likely trustworthiness; alternatively, if the two devices are operated by the same user, the receiving device may simply accept the list of trusted entities communicated by the vouching entity, without any verification. Two friends or colleagues may also establish a bond between their devices (or other entities) that have the same trust impact as if the devices of the two persons were owned by the same person. This is useful, for example, for an employee to automatically benefit from the trust relationships established by an employer, or for a visitor to a person's home to benefit from the trust relationships of devices and other entities of this home. A person of skill in the art will recognize that these are illustrative examples demonstrating how trust can be established, and that there are many variants of this that can be used.
In one embodiment, nodes that provide communication capabilities are introduced and networked. For example, a residence or an office may have integrated a number of low-power nodes that receive and transmit BLE signals, allowing them to communicate with mobile and wearable computers, and also, a capability to transmit signals to a controller that is connected to the Internet, and which can connect to other devices, whether using a similar BLE-capable node; by WiFi, or by other means, including wired methods. The BLE-capable nodes may be integrated in seating such as chairs and sofas; in desks and tables; in turnstiles; in scanners used to gain access to a space; and in numerous other objects or devices. The BLE-capable nodes may thus be integrated into existing electronics (such as the scanner to access a space) or be free-standing (e.g., integrated into a desk or chair). The integrated nodes are preferably powered in the same manner as the electronics into which they are integrated, e.g., from the same power source; the free-standing nodes may have their own power source, such as a battery (e.g., as a BLE-capable node in a chair) or receive power from an outlet (as a BLE-capable node in an office desk might). A BLE-capable node connects to a user device, such as a mobile or wearable computing device, where this connection can be made using BLE or other low-power wireless radio method such as NFC. The BLE-capable node may impersonate a BLE-capable device with which the user device has bonded, such as a phone. This is done with the approval of the user, and can be done in a variety of ways, as described elsewhere herein, and/or in one or more of the above-cited U.S. patent applications.
Another functionality enabled by the disclosed technology is theft protection. Traditionally, computers, phones and tablets manage access control by using passwords and/or biometrics for user authentication. Such credentials are also sometimes used to “wrap” keys that are used to decrypt contents stored on the devices. This means that without access to the credential, it is not possible to decrypt the contents that are encrypted using the wrapped keys. Traditionally, either a biometric match or a password match causes the successful unwrapping of keys, and access to the device and its associated information. In the disclosed technology, this is strengthened. A first key is wrapped using a first set of one or more credentials, such as biometrically derived credentials, passwords and PINs, and a second key is wrapped using a second set of one or more credentials, including a key that is stored by the infrastructure. A third key can be wrapped using another set of one or more credentials, and so on. If a user associates a location with a device, e.g., a computer with her home, then a first access policy dictates how she can authenticate to the computer (i.e., unlock the computer and log in) in the home environment. This first policy may allow any one of a password, a PIN, or a biometric to be used, meaning correct match with stored data associated with one of these would cause a successful login when the computer is in the area of the home, where it will receive the associated key stored on or forwarded by one or more of the nodes of the network at home.
We will call the forwarded key associated with the home the “infrastructure key.” This action corresponds to unwrapping the corresponding key. In fact, there may be multiple keys that are wrapped and associated with the first policy, such as one that is wrapped using the infrastructure key and the PIN of the user; a second wrapped key associated with the first policy that is wrapped using the infrastructure key and the password of the user; and a third wrapped key associated with the first policy that is wrapped using the infrastructure key and data related to the successful authentication using biometrics. Any one of these three keys can be used to access the computer and its stored data.
One way to do that is to have a master key used to access the device encrypted using the first wrapped key, resulting in a first encrypted master key; the same master key encrypted using the second wrapped key, resulting in a second encrypted master key; and the same master key encrypted using the third wrapped key, resulting in a third encrypted master key. A second policy may relate to a second trusted location, such as the office of the user. This may correspond to another access policy, e.g., requiring both a PIN and biometrics or a password login, and in addition to one of these, also the access to the infrastructure key associated with the office. This results in two additional encrypted master keys, using the same general approach described above. When the computer is not in the vicinity of the home network or the office network, a third policy may control the access, which may correspond to requiring either a password and a biometric login, or a PIN and a low-error biometric login, where a low-error login is one that requires a more stringent match to a template than a biometric login that is not low-error. These two options correspond to yet two more encrypted master keys.
The encryption of the master key using a wrapped key can use, e.g., standard symmetric key encryption algorithms. The encrypted master keys are stored on the computer, as are unwrapped versions of the wrapped keys. A node in a network such as the home network or the office network would determine that a recognized computational device is in the proximity, e.g., using the media access control address of the computational device and comparing that to a stored media access control address. After the node verifies the identity of the computational device, it optionally determines a distance or a network distance to the computational device by performing a round-trip time measurement, such as described in S. Brands et al., “Distance-bounding protocols,” EUROCRYPT '93 Workshop on Theory and Application of Cryptographic Techniques. Upon successful verification by the node, the node transmits the infrastructure key to the computational device. This can be done over a secure channel associated with the computational device. The computational device uses the infrastructure key, as described above, and then erases it. The transmission of the infrastructure key may be initiated by a network message sent by the computational device as it is being started or being unlocked.
The illustrative embodiments shown in FIGS. 31 through 35 will now be described in more detail. The particular arrangements of devices and their operation as illustrated in these figures should not be construed as limiting in any way.
FIG. 31 shows a simplified setting in which a requesting device 3101 sends a service request 3103 to a service providing device 3102. The service providing device transmits a service advertisem*nt 3104 to the requesting device. This may be done either in response to receiving service request 3103 or in response to a detected event, such as the detected presence of a device or person, the flow of time, or another event. Service request 3103 may be sent to request service advertisem*nt 3104, or in response to service advertisem*nt 3104. After the service providing device 3102 has determined that the requesting device 3101 is permitted to receive service, the service providing device 3102 performs service provision 3105, utilizing resource 3106 which may be part of service providing device 3102 or external to it.
The service advertisem*nt 3104 is an example of what is also referred to herein as a resource-based service advertisem*nt, and other types of service advertisem*nts, or more generally other arrangements for making resource-based services known to resource-constrained devices, can be used in other embodiments. These and other resource-based service advertisem*nts may be viewed as examples of what are still more generally referred to herein as “service availability notifications.” The term “service availability notification” as used herein is intended to be broadly construed, and should not be viewed as requiring any particular communication protocol or messaging format.
In the FIG. 31 embodiment, the requesting device 3101 is an example of what is more generally referred to herein as a “resource-constrained device” relative to a particular service provided by the service providing device 3102. The particular service illustratively comprises a service that is capable of being performed in the requesting device 3101 but is offloaded from the requesting device 3101 to the service providing device 3102 in order to conserve resources of the requesting device 3101. The service providing device 3102 performs the particular service for the requesting device 3101 in accordance with the request, for example, in the manner described above. It is to be appreciated that the term “resource-constrained device” as used herein is intended to be broadly construed, so as to encompass a wide variety of different types of user devices or other processing devices having at least temporary constraints on their abilities to perform particular services. For example, a device may have an at least temporary constraint on one or more of its resources relative to another device for a particular service, but may not have such constraints relative to the other device for one or more other services. Thus, the resource constraint of requesting device 3101 may be, for example, a temporary constraint on its resources for performing the particular service, while it has no such constraints on its resources for performing one or more other services. Numerous other arrangements of resource-constrained devices are possible.
One or more resources 3106 accessible to the service providing device 3102 are used to perform one or more particular services on behalf of the requesting device 3101, so as to permit the requesting device 3101 to conserve its constrained resources.
Although the one or more resources 3106 of the service providing device 3102 are shown as being separate from that device in the figure, this is by way of illustrative example only, and at least a portion of such resources are illustratively internal to the service providing device 3102 in other embodiments.
FIG. 32 illustrates an example resource-based service advertisem*nt 3200, illustratively corresponding to one possible implementation of the service advertisem*nt 3104 of FIG. 31. The advertisem*nt 3200 comprises an identifier 3201 of service providing device 3102, an optional certificate 3202 indicating that service providing device 3102 is trusted with providing a service, and an optional cost indicator 3203 indicating whether the service provision is free or has a cost. Certificate 3202 may comprise an expiration date, after which the certificate is no longer valid, and a digital signature of a certificate authority. The advertisem*nt 3200 further comprises a descriptor of a service type 3204, where an example service type value may be “location;” a priority value 3205; a quality value 3206; and an optional availability value 3207 that may be set, for example, to 1 if the service is currently available or 0 if it is currently unavailable. An advertisem*nt in certain embodiments may comprise values related to multiple service types, or to just one service type in other embodiments.
FIG. 33 illustrates a system entity 3300 comprising GPS unit 3301, WiFi unit 3302, Bluetooth unit 3303, and processor 3304. In some instances, WiFi unit 3302 and Bluetooth unit 3303 are implemented by the same chip. Processor 3304 may comprise a DRM engine, and data associated with the DRM engine may be stored in secure storage 3311 of storage 3310. Storage 3310 also comprises a portion storing a certificate 3312, a key 3313 and a trust list 3314. In one embodiment, data related to certificate 3312, key 3313 and trust list 3314 are stored in secure storage 3311. For example, secure storage 3311 may be used to store a decryption key, checksum values and/or an authentication key for data related to certificate 3312, key 3313 and trust list 3314. Certificate 3312 is obtained from an external entity that is trusted, and is transmitted using WiFi unit 3302 or Bluetooth unit 3303 to a party wishing to interact with entity 3300, whether to provide entity 3300 with a service or receive a service from entity 3300. Trust list 3314 comprises at least one identifier of an entity other than entity 3300 with which entity 3300 has a trust relationship. An example trust relationship is to have established a connection to receive or give a service. Key 3313 is used to authenticate information, and is associated with certificate 3312. Storage 3310 may comprise additional data not described here, as will be appreciated by a person of skill in the art.
FIG. 34 illustrates a service enabled by the disclosed technology. Wearable computing device 3401 is equipped with a limited-range radio such as a BLE radio, and was manufactured to synchronize with master 3405, which is also equipped with limited-range radio compatible with that of wearable computing device 3401. Traditionally, wearable computing device 3401 and master 3405, which may be a phone, a laptop or a tablet, for example, would be connected using their limited-range radios when they are in each other's proximity, e.g., within five inches from each other. Node 3402 advertises 3410 the capability to synchronize with distant devices, and optionally provides a specification of the devices that are available to be synchronized with. The communication between wearable computing device 3401 and node 3402 is performed using limited range radio communication, matching the type of communication wearable communication device 3401 is capable of performing directly with master 3405. This may depend on what devices are accessible by node 3402 over network 3403 at the time. Wearable computing device 3401 responds 3411 with an indication that it wishes to synchronize with master 3405. Node 3402 or an associated backend device determines how to route the request over network 3403 and initiates a synchronization 3412, which is propagated as synchronization 3413 to node 3404 in the proximity of master 3405, using limited range radio corresponding to the type master 3405 is configured to use to synchronize with wearable computing device 3401.
Alternatively, master 3405 is connected to node 3404 in an alternative manner, including wired or using an alternative wireless standard that is not compatible with the communication wearable computing device 3401 is capable of. Connection 3414 is synchronization, corresponding to synchronization 3412, 3413, and also to synchronization 3415 performed between node 3405 and wearable computing device 3401. Thus, node 3402, network 3403 and node 3404 act as a proxy between wearable computing device 3401 and master 3405.
In one embodiment, one or more messages of synchronizations 3412 and 3413 use additional security beyond what is used in one or more messages of synchronization 3415 and potentially also one or more messages of synchronization 3414. This is because limited-range communication typically may have lower security needs than communication over network 3403, which may be a LAN, a WAN, the Internet, or any combination of these. An example of this is that signals being part of synchronization 3415 may use no encryption, or a relatively light-weight form of encryption, to save power resources of wearable computing device 3401, whereas network traffic going over network 3403 should use military-strength encryption, such as public-key cryptography and AES-512. Moreover, messages that are part of synchronization 3415 may use a checksum method to identify packet loss, while the messages conveyed over network 3403 in addition may use message authentication codes and/or digital signatures, which provide additional protection against malicious modification of packets by malicious intermediaries. Thus, traffic associated with synchronizations 3412 and 3413 may be different from traffic associated with synchronizations 3414 and/or 3415, but correspond to the same payload.
FIG. 35 illustrates a location service. Instead of determining its location using GPS, computing device 3501 obtains a location externally. In signal 3510, node 3502 advertises that it can provide location information. Signal 3510 may comprise information used by device 3501 to determine whether the location signal being advertised is to be trusted, e.g., based on an identity assertion associated with node 3502, or a certification associated with node 3502. Signal 3511 is a request for location, and signal 3512 comprises location information. Computing device 3501 may have a GPS unit, but does not need to have such a unit. Node 3502 may be a peer device such as a device similar to computing device 3501, and may or may not have a GPS unit. It may be stationary, e.g., part of an electric outlet or a router, in which case the location information it has may be obtained during a configuration stage, or from a device such as computing device 3501 during a previous session in which node 3502 is the recipient of location information. Optionally, node 3502 communicates with a backend 3504 connected to node 3502 using a network 3503, using connections 3513 and 3514.
The communication between node 3502 and backend 3504 may comprise information regarding the identity of computing device 3501, allowing a centralized keeping of information of the location of computing device 3501. This is beneficial if, for example, backend 3504 operates a service by which approved users may request information regarding the location of their friends or children, by way of requesting information about where the computational devices of these parties are currently located. Backend 3504 may also be responsible for conveying location and/or identity information to other entities (not shown) in order for these to provide services to computing device 3501. In one embodiment, node 3502 notifies computing device 3501 that it will convey its location to backend 3504; this may allow computing device 3501 to not communicate its location to a party that needs it. In one embodiment, request 3511 contains identity or address location of an entity which node 3502 is requested by computing device 3501 to notify about its location. In one embodiment, a series of computational devices take turns determining their common location by querying a GPS unit. This reduces the power expended by the set of collaborating devices.
Like other embodiments disclosed herein, the embodiments described above in conjunction with FIGS. 31 through 35 are for purposes of illustration only. Numerous other arrangements and associated operating scenarios are possible in other embodiments.
Again, the particular features and other advantages described above are examples presented in the context of illustrative embodiments, and therefore such features and advantages need not be present in other embodiments. Such features and advantages of illustrative embodiments should not be viewed as limiting in any way.
As indicated previously, references herein to “a system” or “the system” in conjunction with various distinct types of features or functionality should not be construed as a requirement that all such features or functionality be present within the same single system. Instead, different systems in different embodiments can include different combinations or other arrangements of the various disclosed features and functionality.
In addition, as noted above, references herein to particular features or other aspects as being “optional” refer to utilization in one or more particular embodiments, and should not be construed as an indication that any other features or aspects, such as features or aspects not explicitly referred to as optional, are required in any particular embodiments.
The above-described systems and other processing entities described herein may be part of an information processing system. A given such entity in an information processing system as described herein is illustratively configured utilizing a corresponding processing device comprising a processor coupled to a memory. The processor executes software program code stored in the memory in order to control the performance of processing operations and other functionality. The processing device also comprises a network interface that supports communication over one or more networks.
The processor may comprise, for example, a microprocessor, a microcontroller, an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), a central processing unit (CPU), an arithmetic logic unit (ALU), a digital signal processor (DSP), a graphics processing unit (GPU) or other similar processing device component, as well as other types and arrangements of processing circuitry, in any combination.
The memory stores software program code for execution by the processor in implementing portions of the functionality of the processing device. A given such memory that stores such program code for execution by a corresponding processor is an example of what is more generally referred to herein as a processor-readable storage medium having program code embodied therein, and may comprise, for example, electronic memory such as SRAM, DRAM or other types of random access memory, read-only memory (ROM), flash memory, magnetic memory, optical memory, or other types of storage devices in any combination.
Articles of manufacture comprising such processor-readable storage media are considered embodiments of the invention. The term “article of manufacture” as used herein should be understood to exclude transitory, propagating signals.
Other types of computer program products comprising processor-readable storage media can be implemented in other embodiments.
In addition, embodiments of the invention may be implemented in the form of integrated circuits comprising processing circuitry configured to implement processing operations associated with the embodiments described herein.
Processing devices in a given embodiment can include, for example, laptop, tablet or desktop personal computers, mobile telephones, or other types of computers or communication devices, in any combination.
Communications between the various elements of an information processing system comprising processing devices associated with respective parties or other system entities may take place over one or more networks. Such networks can illustratively include, for example, a global computer network such as the Internet, a wide area network (WAN), a local area network (LAN), a satellite network, a telephone or cable network, a cellular network such as a 4G or 5G network, a wireless network implemented using a wireless protocol such as Bluetooth, BLE, NFC, ZigBee, WiFi or WiMAX, or various portions or combinations of these and other types of communication networks.
An information processing system as disclosed herein may be implemented using one or more processing platforms, or portions thereof.
For example, one illustrative embodiment of a processing platform that may be used to implement at least a portion of an information processing system comprises cloud infrastructure including virtual machines implemented using a hypervisor that runs on physical infrastructure. Such virtual machines may comprise respective processing devices that communicate with one another over one or more networks.
The cloud infrastructure in such an embodiment may further comprise one or more sets of applications running on respective ones of the virtual machines under the control of the hypervisor. It is also possible to use multiple hypervisors each providing a set of virtual machines using at least one underlying physical machine. Different sets of virtual machines provided by one or more hypervisors may be utilized in configuring multiple instances of various components of the information processing system.
Another illustrative embodiment of a processing platform that may be used to implement at least a portion of an information processing system as disclosed herein comprises a plurality of processing devices which communicate with one another over at least one network. As indicated previously, the network may comprise any type of network, including by way of example a global computer network such as the Internet, a WAN, a LAN, a satellite network, a telephone or cable network, a cellular network such as a 4G or 5G network, a wireless network implemented using a wireless protocol such as Bluetooth, BLE, NFC, ZigBee, WiFi or WiMAX, or various portions or combinations of these and other types of communication networks.
Each processing device of the processing platform comprises a processor coupled to a memory. As indicated above, the processor may comprise a microprocessor, a microcontroller, an ASIC, an FPGA, a CPU, an ALU, a DSP, a GPU or other type of processing circuitry, as well as portions or combinations of such circuitry elements. The memory may comprise RAM, ROM, flash memory or other types of memory, in any combination.
Again, the memory and other memories disclosed herein should be viewed as illustrative examples of what are more generally referred to as “processor-readable storage media” storing program code of one or more software programs.
As mentioned previously, articles of manufacture comprising such processor-readable storage media are considered embodiments of the present invention. A given such article of manufacture may comprise, for example, a storage array, a storage disk, an integrated circuit containing RAM, ROM, flash memory or other electronic memory, or any of a wide variety of other types of computer program products.
Also included in the processing device is network interface circuitry, which is used to interface the processing device with the network and other system components, and may comprise conventional transceivers.
Again, these particular processing platforms are presented by way of example only, and an information processing system may include additional or alternative processing platforms, as well as numerous distinct processing platforms in any combination, with each such platform comprising one or more computers, servers, storage devices or other processing devices.
For example, other processing platforms used to implement embodiments of the invention can comprise different types of virtualization infrastructure in place of or in addition to virtualization infrastructure comprising virtual machines, such as operating system level virtualization infrastructure comprising Docker containers or other types of containers implemented using respective Linux kernel control groups. Thus, it is possible in some embodiments that system components can run at least in part in cloud infrastructure or other types of virtualization infrastructure.
It should therefore be understood that in other embodiments different arrangements of additional or alternative elements may be used. At least a subset of these elements may be collectively implemented on a common processing platform, or each such element may be implemented on a separate processing platform.
Also, numerous other arrangements of computers, servers, storage devices or other components are possible in an information processing system. Such components can communicate with other elements of the information processing system over any type of network or other communication media.
As indicated previously, components or functionality of the system as disclosed herein can be implemented at least in part in the form of one or more software programs stored in memory and executed by a processor of a processing device.
Accordingly, a given component of an information processing system implementing functionality as described herein is illustratively configured utilizing a corresponding processing device comprising a processor coupled to a memory. The processor executes program code stored in the memory in order to control the performance of processing operations and other functionality. The processing device also comprises a network interface that supports communication over one or more networks.
The particular configurations of information processing systems described herein are exemplary only, and a given such system in other embodiments may include other elements in addition to or in place of those specifically shown, including one or more elements of a type commonly found in a conventional implementation of such a system.
For example, in some embodiments, an information processing system may be configured to utilize the disclosed techniques to provide additional or alternative functionality in other contexts. The disclosed techniques can be similarly adapted for use in a wide variety of other types of information processing systems.
It is also to be appreciated that the particular process steps used in the embodiments described above are exemplary only, and other embodiments can utilize different types and arrangements of processing operations. For example, certain process steps described as being performed serially in the illustrative embodiments can in other embodiments be performed at least in part in parallel with one another.
It should again be emphasized that the embodiments of the invention as described herein are intended to be illustrative only. Other embodiments of the invention can be implemented utilizing a wide variety of different types and arrangements of information processing systems, networks and processing devices than those utilized in the particular illustrative embodiments described herein, and in numerous alternative processing contexts, including, by way of example, contexts relating to at least one of infrastructure support to enhance resource-constrained device capabilities, device location services, privacy enhancement using derived data disclosure and/or privacy control and enhancements for distributed networks, as disclosed herein. Also, the particular types and configurations of system entities, processing devices and process operations can be varied in other embodiments. In addition, the particular assumptions made herein in the context of describing aspects of certain illustrative embodiments need not apply in other embodiments. These and numerous other alternative embodiments will be readily apparent to those skilled in the art.