34 may not include the one or more metrics that are not significant.

In some embodiments, the device 32 may be equipped with an application configured to obtain sensor data, identify personal activity metrics, and/or transmit the sensor data and/or metrics. Based upon the trained neural network's identification of metrics that are or are not relevant to accurately determining a health indicator, the application may be specifically configured to control or access one or more sensors of the device 32 to obtain targeted sensor data associated with metrics most predictive of a health indicator. Targeted sensor data may include, for example, data collected via a particular one or more particular sensors, data collected at a particular time, interval, or frequency (e.g., a predetermined “heartbeat” interval), and/or data collected when the device 32 is at a particular location. For instance, if the training unit 22 had assigned a very small weight to certain metrics during the training process (e.g., a weight below some threshold level), the application may not obtain and/or transmit any sensor data that constitutes (or is used to derive) those relatively insignificant metrics. Thus, the application may be configured in such a manner to reduce or eliminate unnecessary activation of sensors, thereby reducing or eliminating unnecessary battery usage of, and/or processing within, the device 32.

FIG. 2 depicts an environment 100 in which the technique 10 of FIG. 1 may be implemented, in accordance with some embodiments. The environment 100 generally includes a mobile electronic device 112 communicatively coupled to a server 140 via a network 150. The mobile electronic device 112 may correspond to one or more mobile electronic devices 12 (i.e., devices of one or more respective training persons) and/or one or more mobile electronic devices 32 (i.e., one or more devices of the target person) of FIG. 1. While one mobile electronic device 112 is illustrated in FIG. 2, it should be understood that a plurality of mobile electronic devices of respective training persons and/or target persons may be envisioned. In other words, in some embodiments, two or more mobile electronic devices 112 may be associated with a single training person or target person. Additional, fewer, or alternate components of the environment 100, and/or additional, fewer, or alternate actions thereof, may be envisioned, in some embodiments.

The mobile electronic device 112 and the server 140 may communicate via the network 150. The network 150 may include one or more wired and/or wireless local area networks (LANs), and/or one or more wired and/or wireless wide area networks (WANs) such as the Internet.

While the server 140 is referred to in the singular, the server 140 may include one or more distinct servers (e.g., distributed servers of a backend server farm). The server 140 includes a memory 142 (i.e., one or more memories). The memory 142 may include ROM, RAM, and/or other types of computer memory, and may include computer-executable instructions to be executed via a processor (i.e., one or more processors) 144. Generally, the computer-executable instructions, when executed via the processor 144, may cause the server 140 to receive or obtain, via a communication module 146, sensor data and/or personal activity metrics from the device 146, and/or derive personal activity metrics from received or obtained sensor data via a Personal Activity Metric (PAM) unit 148.

In some embodiments, the server 140 may receive personal activity metrics from the mobile electronic device 112 via the network 150. In other words, the mobile electronic device 112 may be configured to obtain sensor data, identify one or more personal activity metrics therefrom, and transmit the one or more identified personal activity metrics to the server 140. In some embodiments, though, the mobile electronic device 112 may obtain sensor data and transmit the sensor data to the server 140, and the server 140 may derive at least some personal activity metrics (e.g., via the PAM unit 148) based upon the received sensor data. It should be understood that a combination of the above described techniques may be employed.

In any case, the derived personal activity metrics may be stored at the memory 142. Suites of personal activity metrics corresponding to respective training persons may be stored as the training data sets 14, and a suite of personal activity metrics corresponding to the target person may be stored as the target activity data set 34.

The memory 142 may further include the neural network unit 20, which may be implemented as a software module, for example, and may be configured to (i) train the artificial neural network using the training data sets 14, and (ii) utilize the trained artificial neural network to determine a health indicator of the target person, based upon the target activity data set 34, as discussed above.

The mobile electronic device 112 may include a memory (i.e., one or more memories) 152. The memory 152 may include a ROM, RAM, and/or other computer memories. The memory 152 may include computer-executable instructions that, when executed via a processor (i.e., one or more processors) 154, cause the mobile electronic device 152 to obtain sensor data via one or more sensors at the mobile electronic device 112, derive personal activity metrics from the obtained sensor data, and/or communicate with the server 140 over the network 150, via the processor 154 and/or a communication module 156.

The computer-executable instructions may, in some embodiments, be included within an application 158 configured specifically to activate and/or access sensors to obtain sensor data associated with a training person or the target person (i.e., the person having the mobile electronic device 112). Sensors obtaining sensor data may include a microphone 160, a camera (and/or another imaging unit) 162, a positioning unit (e.g., GPS) 164, an accelerometer 166, a gyroscope 168, and/or other sensors, including those described herein. In some embodiments, the application 158 may further process obtained sensor data to identify activities performed by (i.e., personal activity metrics associated with) a user of the device 112. The mobile electronic device 112 may transmit the obtained sensor data and/or the identified personal activity metrics to the server 140 via the network 150. The mobile electronic device 112 may transmit acquired and/or derived personal activity metrics via the processor 154, and/or communication module 156, to the server 140 over the network 150.

In some embodiments, the mobile electronic device 112 may include one or more user interfaces 170, which may include one or more displays 172 and/or one or more input devices 174 (e.g., a touch screen, keypad, etc.) via which a user of the device 112 may interact with the device 112. In some embodiments, the user may utilize the user interface(s) 170 to expressly “opt-in” to the sensor data collection functionalities described herein (e.g., via the application 158). An “opt-in” may include a user acceptance of collection of sensor data (via one or more particular sensors) for purposes of neural network training and/or health indicator determination, as described herein.

In some embodiments, the neural network unit 20, or at least some functions or components thereof, may be disposed at the mobile electronic device 112. For example, the neural network unit 20 (or more specifically, the target health processing unit 24 depicted in FIG. 1) may be disposed at a mobile electronic device 112 of the target person. In these embodiments, the mobile electronic device 112 may cause training of the artificial neural network, and/or cause the artificial neural network to be utilized to determine the health indicator of the target person.

Additional, fewer, and/or alternate components of the environment 100 may be possible, in some embodiments.

Example Elements of an Artificial Neural Network

FIG. 3 depicts an example artificial neural network 200 that may be trained to determine health indicators of persons based upon activity data sets obtained or derived from mobile electronic devices, according to some embodiments. The neural network 200 may be trained, for example, by the neural network unit 20 (or more specifically, the training unit 22) of FIG. 1 and/or the neural network unit 120 of FIG. 2. The neural network 200, once trained, may be utilized via the same neural network unit 20 (specifically, the target health processing unit 24) of FIG. 1 and/or neural network unit 120 of FIG. 2.

The neural network 200 includes a multiplicity of neurons arranged in multiple layers, and includes an input layer 202, a number of intermediate (also referred to herein as “hidden”) layers 204-1 through 204-M, and an output layer 206. M may be any integer greater than or equal to one. Each of the layers 202, 204-1 through 204-M, and 206 may have any number of inputs/neurons/outputs (e.g., the layer 204-1 including neurons “1, 1” through “1, j,” wherein j represents the number of neurons in layer 204-1). Other configurations of the neural network 200 are possible.

The input layer 202 may correspond to personal activity metrics that may be obtained or derived from mobile electronic devices (e.g., a device 12, 32, and/or 112). Accordingly, each input x₁through x_i(wherein i represents the number of inputs) of the input layer 202 may correspond to a particular personal activity metric, such as a measure of a particular activity performed by a person having a mobile electronic device (e.g., average or total hours spent walking per week, average walking speed, average or total hours slept per night, etc.). The input layer 202 may have any number i of inputs. In some embodiments, the number of inputs used by the neural network 200 during training may be greater than the number of inputs used by the neural network after training.

Each of the intermediate layers 204-1 through 204-M may include any number of neurons, and a different number of neurons at each layer is possible. Each intermediate layer neuron may operate on one or more inputs from the input layer 202 and/or one or more outputs of other layers (e.g., the preceding intermediate layer), to generate a decision or other output.

The output layer 206 may include one or more outputs (only one output y₁is shown), each corresponding to a health indicator (e.g., a life expectancy, a life insurance premium, a health score, etc.) associated with a person. In some embodiments, outputs of the neural network 200 may be obtained not just from the output layer 206, but also from one or more of the intermediate layers 204-1 through 204-M. For example, a first intermediate layer of the intermediate layers 204 may produce an output in the form of a first health indicator (e.g., a life expectancy), and a second intermediate layer of the intermediate layers 204, or the output layer 206, may produce an output in the form of a second health indicator (e.g., a life insurance premium), such that the second health indicator is determined based upon the first health indicator (and possibly some other input(s)).

In some embodiments, the neural network 200 may be a recurrent neural network, wherein decisions or outputs from at least one layer of the neural network 200 are fed back to at least one previous layer during training to provide an indication of significance (e.g., a “weight”) of a particular input or intermediate layer output in determining a particular decision or calculation. For example, outputs of an intermediate layer 204 and/or output layer 206 may be utilized to weight personal activity metrics at the input layer 202. As a result of training, in some embodiments, insignificant inputs of inputs 202, and/or insignificant neurons of layers 204-1 through 204-M, may be bypassed in order to reduce processing demands in determining a health indicator of a person.

FIG. 4 depicts an example neuron 220 that may correspond to a neuron of the neural network 200 of FIG. 3 (e.g., a neuron “1, 1” of the layer 204-1). One or more inputs x₁through x_imay be provided to 220. A particular input may be, for example, an input to the neural network 200 itself (e.g., a personal activity metric from an activity data set), or an output of a neuron of another layer of the neural network 200. Each input may be assigned a respective weight (w₁through w_i), wherein the weight of each input may be determined during the training process. In some cases, an input may be determined to be insignificant to a decision or calculation of a neuron, and may accordingly be assigned a zero or near-zero weight.

The weighted inputs x₁through x_imay be provided to a summing node 222 of the neuron 220. The sum of the weighted inputs, z₁, may be provided as an input to a function 224, labeled in FIG. 4 as F_1,1(z₁). The function 224 may represent any suitable linear or non-linear operation on z₁. As shown in FIG. 4, the output of function 224 may be provided to a number of neurons of the next layer, and/or may be provided as an output of neural network 200. For example, the output may indicate a location of a segment or segment portion, or may be a parameter that is calculated or determined as an interim step when determining such a location.

In other embodiments, and/or in other training scenarios, neuron 220 may be arranged differently than is shown in FIG. 4. For example, summing node 222 may be omitted, and function 224 may operate directly on one or more of the inputs x₁through x_i. As another example, neuron 220 may not apply weights to any of the inputs x₁through x_i.

Health Indicators and Their Utilization

The health indicators generated by the trained artificial neural network described above may generally be utilized to represent, in some direct or indirect manner, the health of a target person. A health indicator may be, for example, a life expectancy, a life insurance premium cost, or a health insurance premium cost for the target person. In some embodiments, a health indicator may be an overall “health score” representing an aggregate health (e.g., fitness) level of the target person. It should be understood that, in some cases, correlation may exist between different health indicators. A health score may, for example, be utilized at least in part to determine a life expectancy and/or life insurance premium for the target person. As another example, a target person's life expectancy may be a significant factor in determining a life insurance premium and/or other parameters of a life or health insurance policy.

In some embodiments, the health indicator may be utilized to “cross-sell” a life and/or health insurance policy to a target person already having some other insurance policy. The target person may have an existing automotive insurance policy, for example, in connection with which at least some sensor data is obtained from a mobile electronic device associated with the target person (e.g., as a part of a “good driving” rewards program). Based upon at least this sensor data, personal activity metrics may be derived, and an entity (e.g., an insurance provider) may utilize the artificial neural network to determine a health indicator and offer a life and/or health insurance policy to the automotive insurance customer, based at least in part upon the determined health indicator. Additionally or alternatively, the entity may utilize the artificial neural network to determine an updated health indicator of an existing life and/or health insurance customer, and update a life and/or health insurance premium, for example, based upon the determined health indicator.

Example Method For Determining a Health Indicator of a Person

FIG. 5 illustrates an example computer-implemented method 300 for determining a health indicator of a target person using a proxy model (e.g., an artificial neural network) as a proxy for a health assessment process, in accordance with some embodiments. The method 300 may be performed, for example, by the server 140 of FIG. 2, or by both the mobile electronic device 112 and the server 140 of FIG. 2.

The method 300 may include training, using labeled training data, the proxy model to enable the proxy model to determine health indicators of persons based upon personal activity (block 302). The labeled training data may include, for example, a plurality of activity data training sets, each corresponding to a different one of a plurality of training persons. Each activity data training set may comprise a suite of personal activity metrics obtained or derived from (i.e., obtained from, and/or derived from sensor data obtained from) one or more mobile electronic devices of the respective training person. The training data may be “labeled” in that each activity data training set is labeled by one or more health indicators assigned to the respective training person by the health assessment process.

The health assessment process may be, in some embodiments, a partially or entirely manual process, and may include, for example, a physical examination and/or a process that relies upon medical history information associated with the respective training person (e.g., an underwriting process).

Training the proxy model may, in some embodiments, include comparing (i) a health indicator determined by the proxy model processing an activity data training set corresponding to a particular training person from among the training persons to (ii) a health indicator assigned to the particular training person by the health assessment process. In other words, a health indicator generated for a training person by the proxy model may be compared to the assigned or “known” health indicator for the same training person. This process may be performed iteratively, e.g., with various weights of the neural network being adjusted after each iteration, until the proxy model outputs, for some or all of the activity data training sets, a health indicator that matches (exactly, or within some acceptable margin of error) the known health indicator of the respective training person.

In some embodiments, the proxy model may be a recurrent neural network, and may include a plurality of hidden (or “intermediate”) layers. As noted above, the training process may involve the adjustment of various weights of the neural network. For example, training the proxy model may include determining respective weights of one or more personal activity metrics provided as inputs to the proxy model. Additionally or alternatively, training the proxy model may include determining respective weights of one or more outputs generated by neurons of one or more of the plurality of hidden layers.

In some embodiments, one or more of the plurality of activity data training sets may be obtained via an application installed at respective mobile electronic devices of one or more respective training persons of the plurality of training persons. The application, when installed on a particular mobile electronic device, may be configured to obtain sensor data at a predetermined frequency of time, via activation of one or more sensors of the particular mobile electronic device.

The method 300 may further include receiving an activity data set corresponding to the target person (block 304). The activity data set corresponding to the target person may include a suite of personal activity metrics obtained or derived (i.e., obtained from, and/or derived from sensor data obtained from) from one or more mobile electronic devices of the target person.

In some embodiments, the activity data set corresponding to the target person may be obtained via an application installed at at least one device of the one or more mobile electronic devices of the target person. The application may be configured to obtain sensor data at a predetermined frequency of time, via activation of one or more sensors of the at least one device.

In some embodiments, training the proxy model includes identifying, from the one or more personal activity metrics, one or more significant metrics having a significant predictive effect upon the health indicator, and/or identifying one or more insignificant metrics not having a significant predictive effect upon the health indicator. In these embodiments, the application may be specifically configured to activate the one or more sensors of the at least one device to obtain sensor data associated with the one or more significant metrics, while not being configured to obtain and/or retain sensor data associated with the one or more insignificant metrics. Effectively, the application may be configured to activate one or more sensors of the at least one device in a manner to avoid unnecessary sensor activation.

The method 300 may further include determining a health indicator of the target person, at least by processing the activity data set using the trained proxy model (block 306). In some embodiments, the health indicator may be determined based upon other data (e.g., other health-related and/or financial data associated with the target person) in addition to the activity data set associated with the target person.

The method 300 may include additional, fewer, or alternate operations, in some embodiments.

Example Method For Health Assessment

Various embodiments can include an example computer-implemented method for health assessment. The method can comprise providing an artificial neural network trained with training data. The training data can include training metrics and health indicators corresponding to training persons. The training data can be based on data from mobile electronic devices associated with the training persons. The artificial neural network can be trained by at least identifying one or more first ones of the training metrics as having one or more predicted effects on one or more of the health indicators.

In many embodiments, the method further can include receiving activity data comprising personal activity metrics corresponding to a target person. The activity data can be based on data from one or more sensors associated with the target person and associated with one or more of the training metrics. The method also can include determining, by the artificial neural network, a health indicator of the target person based upon the activity data corresponding to the target person. Additionally, the method can include outputting the health indicator of the target person to an electronic device for communication of the health indicator to a user of the electronic device.

In some embodiments, the artificial neural network can be further trained by at least identifying one or more second ones of the training metrics as not having a predicted effect on one or more of the health indicators. The health indicators can be based on medical history information associated with the training persons. In a few embodiments, the artificial neural network can be trained with the training data by comparing a health indicator determined by the artificial neural network processing a portion of the training data corresponding to a particular training person of the training persons to a known health indicator assigned to the particular training person by a health assessment process. In a number of embodiments, the artificial neural network can include a recurrent neural network comprising hidden layers. The artificial neural network can be trained by determining respective weights of at least one of: (i) one or more activity metrics provided as one or more inputs to the artificial neural network, or (ii) one or more outputs generated by one or more of the hidden layers.

In several embodiments, the training data can be obtained from applications installed on the mobile electronic devices associated with the training persons. The applications can be configured to obtain sensor data from sensors of the mobile electronic devices at one or more predetermined frequencies of time.

In a number of embodiments, the activity data associated with the target person can be obtained from an application installed on at least one device associated with the target person, wherein the at least one device comprises the one or more sensors. The application can be configured to obtain sensor data from the one or more sensors at a predetermined frequency of time.

Example System For Health Assessment

Various embodiments can include an example computer system for health assessment. The computer system can comprise one or more processors and one or more memories storing computer-executable instructions. The computer-executable instructions, when executed by the one or more processors, can cause the computing system to perform one or more acts. In many embodiments, the one or more acts can include providing an artificial neural network trained with training data. The training data can comprise training metrics and health indicators corresponding to training persons. The training data can be based on data from mobile electronic devices associated with the training persons. Further, the artificial neural network can be trained by at least identifying one or more first ones of the training metrics as having one or more predicted effects on one or more of the health indicators.

In a number of embodiments, the one or more acts additionally can include receiving activity data comprising personal activity metrics corresponding to a target person. The activity data can be based on data from one or more sensors associated with the target person and associated with one or more of the training metrics. In some embodiments, the one or more acts also can include determining, by the artificial neural network, a health indicator of the target person based upon the activity data corresponding to the target person. In several embodiments, the one or more acts further can include outputting the health indicator of the target person to an electronic device for communication of the health indicator to a user of the electronic device.

In many embodiments, the artificial neural network can be further trained by at least identifying one or more second ones of the training metrics as not having a predicted effect on one or more of the health indicators. The health indicators can be based on medical history information associated with the training persons. In a number of embodiments, the artificial neural network can be trained with the training data by comparing a health indicator determined by the artificial neural network processing a portion of the training data corresponding to a particular training person of the training persons to a known health indicator assigned to the particular training person by a health assessment process.

In some embodiments, the artificial neural network can include a recurrent neural network comprising hidden layers. The artificial neural network can be trained by determining respective weights of at least one of: (i) one or more activity metrics provided as one or more inputs to the artificial neural network, or (ii) one or more outputs generated by one or more of the hidden layers. In a few embodiments, the training data can be obtained from applications installed on the mobile electronic devices associated with the training persons. The applications can be configured to obtain sensor data from sensors of the mobile electronic devices at one or more predetermined frequencies of time. In several embodiments, the activity data associated with the target person can be obtained from an application installed on at least one device associated with the target person. The at least one device can comprise the one or more sensors. The application can be configured to obtain sensor data from the one or more sensors at a predetermined frequency of time.

Example Medium For Health Assessment

Various embodiments can include an example non-transitory, computer-readable medium for health assessment. The non-transitory, computer-readable medium storing instructions, when executed by one or more processors, can cause the one or more processors to perform one or more acts. In many embodiments, the one or more acts can include providing an artificial neural network trained with training data. The training data can comprise training metrics and health indicators corresponding to training persons. The training data can be based on data from mobile electronic devices associated with the training persons. The artificial neural network can be trained by at least identifying one or more first ones of the training metrics as having one or more predicted effects on one or more of the health indicators.

In a number of embodiments, the one or more acts further can include receiving activity data comprising personal activity metrics corresponding to a target person. The activity data can be based on data from one or more sensors associated with the target person and associated with one or more of the training metrics. In several embodiments, the one or more acts further can include determining a health indicator of the target person based upon the activity data corresponding to the target person. In certain embodiments, the one or more acts further can include outputting the health indicator of the target person to an electronic device for communication of the health indicator to a user of the electronic device.

In a few embodiments, the artificial neural network can be further trained by at least identifying one or more second ones of the training metrics as not having a predicted effect on one or more of the health indicators. The health indicators can be based on medical history information associated with the training persons. In a number of embodiments, the artificial neural network is trained with the training data by comparing a health indicator determined by the artificial neural network processing a portion of the training data corresponding to a particular training person of the training persons to a known health indicator assigned to the particular training person by a health assessment process.

In some embodiments, the artificial neural network can comprise a recurrent neural network comprising hidden layers. The artificial neural network can be trained by determining respective weights of at least one of: (i) one or more activity metrics provided as one or more inputs to the artificial neural network, or (ii) one or more outputs generated by one or more of the hidden layers.