Device training with determination quantification

Abstract

Systems, methods, devices, and apparatus are provided for device training with deterministic quantization. For example, a device can include a battery, a communication node, a sensor configured to collect user fitness data, and a processor coupled to the battery, the communication node, and the sensor. The processor can be configured to monitor a power state of the battery, determine a usage pattern of the electronic device, and determine, based on the power state and the usage pattern, a training machine learning model for analyzing a quantification employed by the user fitness data. In addition, the processor can be configured to train the machine learning model with the determined quantization using the user fitness data and to transmit the determined quantization and updates to the trained machine learning model via the communication node without transmitting the user fitness data.

Description

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims the benefit of U.S. Provisional Application No. 63 / 517,825, filed on August 4, 2023, the contents of which are incorporated herein by reference. Technical Field

[0003] The present disclosure generally relates to apparatus, systems, non-transitory machine-readable media, and methods associated with determining a quantized training device. Background Art

[0004] For example, the computer device may be a laptop computer, a desktop computer, a tablet, a smartphone, a mobile device, a digital camera, or a wearable device (such as smart glasses, a wristband device, and / or redundant combinations thereof), among other types of computing devices.

[0005] The computing device can be used to implement the machine learning model. The computing device can also be used to train the machine learning model. Summary of the invention

[0006] On the one hand, the present disclosure provides an electronic device for training a machine learning model, comprising: a battery; a communication node; a sensor configured to collect user fitness data; and a processor coupled to the battery, the communication node, and the sensor, wherein the processor is configured to: monitor the power state of the battery; determine a usage pattern of the electronic device; determine the quantification used to train the machine learning model for analyzing the user fitness data based on the power state and the usage pattern; and transmit the determined quantification and updates to the machine learning model via the communication node without transmitting the user fitness data.

[0007] On the other hand, the present disclosure provides a system for training a machine learning model, comprising: a server configured to transmit an initial model for analyzing user fitness data; and multiple electronic devices, each of which has different hardware specifications and is each configured to: receive the initial model; collect corresponding user fitness data; determine the corresponding quantization used to train the initial model; and transmit the corresponding quantization and corresponding updates to the initial model to the server without transmitting the corresponding user fitness data; wherein the server is further configured to: aggregate the corresponding updates based on the corresponding quantization to create an updated model; and transmit the updated model to the multiple electronic devices.

[0008] On the other hand, the present disclosure provides a method for training a machine learning model, which includes: determining a time of day; determining a power state of an electronic device; determining a processing power of the electronic device; determining an amount of memory available in the electronic device; determining an activity being performed by the electronic device; collecting user fitness data by the electronic device; training the machine learning model using the user fitness data based on the following quantifications: the time of day; the amount of memory available; the power state; and the processing power; and transmitting the quantification and updates to the machine learning model. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1 An example computing system for training a machine learning model according to some embodiments of the present disclosure is described.

[0010] Figure 2 A block diagram illustrating a method for training a machine learning model according to some embodiments of the present disclosure.

[0011] Figure 3 is a flowchart corresponding to a method for training a machine learning model according to some embodiments of the present disclosure.

[0012] Figure 4 is a flowchart for training a machine learning model according to some embodiments of the present disclosure.

[0013] Figure 5 is a block diagram of an example computer system in which embodiments of the present disclosure may operate. DETAILED DESCRIPTION

[0014] Apparatus, systems, machine-readable media, and methods related to determining quantitative training electronic devices are described herein. Wearable electronic devices can enable state-of-the-art Internet of Things (IoT) applications in fitness, health, and sports (FHS). Sensors on the device worn by a user can be used to monitor the user's state and behavior. For example, the sensors can collect user fitness data, and the device can store the user's fitness data. Wearable devices come in a variety of form factors and specifications to meet the needs of low, mid, and high-end market segments.

[0015] Machine learning applications can enable better guidance to users by analyzing user activity and / or performance and can be a means to achieve FHS goals. However, this can expose users to potential privacy issues related to their personally identifiable information and make them vulnerable to adversaries who collect such data for malicious purposes, especially when the user data is transmitted to another device and / or across a network.

[0016] One or more embodiments of the present disclosure address the above and other shortcomings of some methods. For example, an electronic device may collect user fitness data, monitor the power state of a battery, and determine a usage pattern. A quantization to be employed for training a machine learning model for analyzing the user fitness data may be determined based on the power state and the usage pattern. The electronic device may use the user fitness data to determine the quantization to train the machine learning model. Indications of the determined quantization and updates to the machine learning model may be transmitted (e.g., to a central server) without transmitting the user fitness data.

[0017] In various examples, the electronic device may be used to implement federated learning. Federated learning describes the use of multiple distributed electronic devices to train an algorithm. In various instances, each of the electronic devices may receive an initial model. The electronic device receives the initial model from a centralized server. The initial model may be used by the electronic device for operation of the electronic device until user fitness data is collected and a locally trained machine learning model is generated. The locally trained machine learning model (or updates to the initial model) may be transmitted to a central server, which may aggregate updates received from multiple devices or locally trained models into a globally updated federated model, which may be delivered back to the electronic device for its operation.

[0018] The central server may share the initial model with multiple electronic devices. The processing resources and / or memory resources of the electronic device may train the initial model to create updates to the initial model and / or generate a locally trained machine learning model. The electronic device may obtain user fitness data while the electronic device is in operation. The user fitness data may be used by the electronic device to train the received initial model to create updates to the initial model and / or generate a locally trained machine learning model. The electronic devices may perform training using their local computing resources independently of each other. Quantized data, updates to the initial model, and / or a locally trained machine learning model may be uploaded from the electronic device to the central server. The difference between a locally trained machine learning model and an update to the initial model is that a locally trained machine learning model refers to the entire model, while a model update refers to any changes to the initial model, such as updated weights, biases, etc.

[0019] The electronic device determines the quantization used to train the initial model. Quantization refers to how much time and resources the electronic device invests to train the model (e.g., how “difficult” the electronic device is to train the model). Quantization may be determined by the electronic device based on the power state of the battery of the electronic device, the usage pattern of the electronic device, the memory and / or processing power of the electronic device, and / or other factors. For example, if the power state of the battery is relatively high, the initial model may be trained with a larger quantization than a relatively low power state. As another example, if the usage pattern of the device indicates relatively little use in a future time period, the initial model may be trained with a larger quantization than a usage pattern indicating relatively more use in a future time period. The electronic device may send an indication of the quantization used to train the initial model (“quantization data”) to the central server along with any updates or trained machine learning models it sends to the central server. The quantization data may be sent to the central server. The electronic device may send updates to the initial model and / or quantization data without sending user fitness data, thereby protecting any personally identifiable information that may be embedded in the user fitness data to protect the user's privacy.

[0020] The central server may collect corresponding model updates and quantization data from multiple electronic devices. The central server may aggregate the received information to update the initial model to produce an updated model. The central server may use the quantization data to determine corresponding weight parameters to be applied to model updates received from each electronic device when aggregating the updates. For example, the central server may apply a larger weight parameter to an update received from a first electronic device with an indication of greater quantization, and not to an update received from a second electronic device with an indication of less quantization, so that the update from the first electronic device has a greater impact on the global update of the initial model. Relatively larger quantization indications may imply that the update is more reliable than updates associated with relatively smaller quantization.

[0021] The central server may transmit the updated model for use by the electronic device. The electronic device may operate using the updated model. In some embodiments, the electronic device may further train the updated model and / or send additional updates to the central server.

[0022] The figures herein follow a numbering convention in which the first digit or digits correspond to the figure number and the remaining digits identify the element or component in the figure. Similar elements or components between different figures may be identified by using similar numerals. For example, reference numeral 102 may refer to Figure 1 Component "02" in , and similar components in Figure 2 202. Multiple similar elements within a figure may be referenced by a reference numeral followed by a hyphen and another number or letter. For example, 203-1 may refer to Figure 2Reference elements 203-1 and 203-2 in the illustrative example may refer to reference element 203-2 which may be similar to element 203-1. Such similar elements may be referenced as a whole without hyphens and additional numbers or letters. For example, elements 203-1 and 203-2 or other similar elements may be referenced as a whole as 203.

[0023] Figure 1 An example computing system 100 for training a machine learning model according to some embodiments of the present disclosure is illustrated. The computing system 100 may include a central server 102 and electronic devices 103-1, 103-N (e.g., devices 103-1, 103-N). The electronic device 103 may include a processor coupled to a battery, a communication node, and one or more sensors 122-1, 122-Z.

[0024] The computing system 100, the central server 102, and the electronic device 103 may include hardware, firmware, and / or software configured to train a machine learning model. As used herein, a machine learning model may include a plurality of weights, biases, and / or activation functions, as well as other variables, that may be used on the electronic device 103. The central server 102 and the electronic device 103 may further include memory subsystems 111-1, 111-2, 111-U (e.g., including non-transitory machine-readable media (MRM)) on which instructions (e.g., update instructions 115) and / or data (e.g., received training data 114, initial model 105, updated model 110, training data 107 including model updates 112 and / or quantization 113) may be stored. Although the following description refers to one processing device and one memory device, the description may also be applicable to systems having multiple processing devices and multiple memory devices. In such examples, instructions may be distributed across multiple memory devices for storage, and instructions may be distributed across multiple processing devices for execution.

[0025] The memory subsystem 111 may include a memory device. The memory device may be an electronic, magnetic, optical or other physical storage device that stores executable instructions and / or data. The memory device may be, for example, a non-volatile or volatile memory. In some examples, the memory device may be a non-transitory MRM including a random access memory (RAM), an electrically erasable programmable ROM (EEPROM), a storage drive, an optical disk, and the like. The memory subsystem 111 may be disposed within a controller, a central server 102, and / or an electronic device 103. In this example, executable instructions (e.g., update instructions 115) may be "installed" on the central server 102. For example, the memory subsystem 111 may be a portable external or remote storage medium that allows the central server 102 and / or the electronic device 103 to store instructions and / or data. In this case, the initial model 105 may be part of an "installation package" sent by the central server 102 to the electronic device 103. The memory subsystem 111-2 of the central server 102 may be encoded with executable instructions for sending the initial model 105 to the electronic device 103. In some examples, the processor 104 of the electronic device 103 may be configured to receive the initial model 105 via a communication node.

[0026] Central server 102 may execute instructions using processor 104 - 1 (also referred to herein as processing device 104 - 1 ). Instructions may be stored in memory subsystem 111 - 2 prior to execution by processing device 104 - 1 . For example, execution of the instructions may cause initial model 105 to be provided to electronic device 103 .

[0027] In some embodiments, the central server 102 may send the initial model 105 to each electronic device 103. The electronic device 103-1 may store the received initial model 106-1. The electronic device 103-N may store the received initial model 106-Q. The central server 102 may provide the initial model 105 to the electronic device 103 using the wireless network 108 and / or the physical network 109.

[0028] The electronic device 103 may store the received initial model 106, the user fitness data, and the training data 107. The electronic device 103 including the processor 104-2, 104-P (also referred to herein as the processing device 104-2, 104-P) may utilize the processor 104-2, 104-P to execute the received initial model 106 to operate the electronic device 103. The electronic device 103 may use the user fitness data to train and update the received initial model 106. The initial model 105 may determine the quantization 113 to be trained using the user fitness data, as described in more detail below. The indication of determining the quantization is referred to herein as the quantization data. The terms quantization and quantization data may be used interchangeably.

[0029] Model update 112 is the result of updating the received initial model 106. For example, the electronic device 103 may use information based on the user's activities when the user is wearing or holding the electronic device 103 (referred to herein as user fitness data) to create model update 112. The sensor 122 in the electronic device 103 may monitor the user's behavior and state to obtain information based on the user's activities. Monitoring the user's behavior and state may involve monitoring one or more of the user's heart rate, body temperature, oxygen saturation, etc. The electronic device 103 may use the information obtained from the user's activities to train the received initial model 106, thereby creating model update 112. Model update 112 does not include user fitness data. However, user fitness data (such as monitored heart rate, body temperature, oxygen saturation, etc.) can be used to create model update 112. The electronic device 103 can transmit the model update 112 to the central server 102 without transmitting the personally identifiable information contained in the user fitness data. When the data is sent to the central server 102, the personally identifiable information of the user of the electronic device 103 remains secure and private.

[0030] The electronic device 103 may determine the quantization employed for training the received initial model 106. In some embodiments, the electronic device 103 may store data including an indication of the quantization 113 and the training data 107. The quantization 113 may include information related to the frequency of training the received initial model 106, the amount of resources of the electronic device 103 used for training, the total duration of training, and / or other information indicating how much training has been performed on the received initial model. For example, the quantization 113 may include the time of day at which the received initial model 106 is trained and how long it lasts, the amount of battery power that the electronic device 103 has when training is performed, the memory availability of the electronic device 103 when training is performed, the processing power of the electronic device 103 when training is performed, and / or the charging schedule of the electronic device 103, etc. In addition, in some embodiments, the electronic device 103 may be configured to collect corresponding user fitness data in an amount that depends on the power state of the electronic device 103. However, the embodiments are not limited thereto. The electronic device 103 may collect data based on available memory on the electronic device 103 , the processing power of the electronic device, and / or the activity of the electronic device 103 , among other things.

[0031] In some examples, the processor 104 may be configured to store the quantification data 113. The processor 104 may be configured to monitor the power state of the battery, determine the usage pattern of the wearable electronic device 103, determine the quantification 113 employed to train the received initial model 106 to analyze the user's fitness data based on the power state and the usage pattern, train the received initial model 106 using the user's fitness data to determine the quantification 113, and / or transmit the model update 112 and the quantification data 113 via the communication node without transmitting the user's personally identifiable information. In some embodiments, as described herein, the processor 104 may be configured to determine the quantification 113 based on the time of day, determine the quantification 113 based on the memory availability of the electronic device 103, determine the quantification 113 based on the time remaining before the next charge of the battery according to the usage pattern, and / or determine the quantification 113 based on environmental conditions.

[0032] The quantification 113 sent by each electronic device 103 allows the central server 102 to determine the level of importance given to the information provided in the model update 112. For example, if the quantification 113 informs the central server 102 that the electronic device 103-1 has higher processing power than another electronic device 103-N, then the model update 112-1 from the electronic device 103-1 may be given a higher importance than the model update 112-S from another electronic device 103-N when creating the updated model 110. Similarly, if the quantification 113 informs the central server 102 that the user fitness data from the electronic device 103-1 was obtained at a lower battery power and the user fitness data from the electronic device 103-N was obtained at a higher battery power, then the quantification 113-T from the electronic device 103-N may be given a higher level of importance. The central server 102 can use the training data 107 from all electronic devices 103 to create the updated model 110. The importance levels, also referred to herein as respective weight parameters, are calculated by the central server 102 and applied to the information received from each of the electronic devices 103 when creating the updated model 110 .

[0033] The central server 102 may receive output from the electronic device 103. The output may be training data 107 sent from the electronic device 103, which may be stored by the central server 102 as received training data 114. The training data 107 may be sent to the central server 102 using the wireless network 108 and / or the physical network 109. Each of the electronic devices 103 may update the received initial model 106 to provide the training data 107 to generate an updated model 110. The central server 102 may generate corrections (e.g., training feedback) for the initial model 105 using the received training data 114. The corrections may be used to modify the weights and / or biases of the initial model 105 to create the updated model 110. The corrections may be provided to the electronic device 103 (not shown). For example, when the updated model 110 is created by the central server 102, updated instructions are also created to determine when the updated model will be sent to the electronic device 103. As used herein, updated model 110 may describe an initial model that has been updated or trained using aggregated received training data 114 .

[0034] The central server 102 may execute the update instructions 115 using the processing device 104-1 to create the updated model 110. The update instructions 115 may be stored in the memory subsystem 111-2 before being executed by the processing device 104-1. The execution of the update instructions 115 may cause the received training data 114 (specifically the model updates 112) to be aggregated into the initial model 105, thereby creating the updated model 110. Each of the model updates 112 may be aggregated into the initial model 105 according to the quantization associated therewith. The updated model 110 may be sent to the electronic device 103. The updated model 110 is also referred to herein as a global updated model. The update instructions 115 may be software encoded in a computer-readable medium, or may be logic.

[0035] Once the updated model 110 is received by the electronic device 103, the electronic device 103 may be configured to operate based on the updated model 110. In some embodiments, the electronic device 103 may continue to train the updated model 110. For example, while the updated model 110 is being sent to the electronic device 103-1, the sensor 122-1 of the electronic device 103-1 may continue to collect data related to the use of the electronic device 103-1. The collected additional data may then be used to create additional updates to the updated model 110. Furthermore, as additional data is collected, different quantizations employed to train the updated model 110 may be determined. The electronic device 103-1 may collect additional user fitness data and train the updated model 110 using the additional user fitness data with the newly determined quantization. Additional model updates and additional quantization data may be combined to create additional training data to be sent to the central server 102 to produce additional updated models.

[0036] In various examples, the processor 104 may be internal to the memory subsystem 111, rather than external to the memory subsystem 111 as shown. For example, the processor 104 may be a processor-in-memory (PIM) processor. For example, the processor 104 may be incorporated into the sensing circuitry of the memory subsystem 111 and / or may be implemented in the periphery of the memory subsystem 111. The processor 104 may be implemented under one or more memory arrays of the memory subsystem 111.

[0037] Figure 2 Block diagram 201 illustrating a method for training a machine learning model according to some embodiments of the present disclosure. Figure 2 The central server 202 and multiple electronic devices 203-1, 203-2, 203-G are included. The central server 202 can store an initial model 205 and an updated model 210. The central server 202 can store instructions for performing weight aggregation to create an updated model 210 using training data 207 received from the electronic device 203. In various examples, the instructions can be implemented as software, hardware, and / or firmware. The instructions can be Figure 1 Executed by processor 104-1.

[0038] The electronic device 203 may be a plurality of wearable and / or portable devices. For example, the electronic device 203 may be a tablet, a cellular phone, a mobile computing device, a virtual reality (VR) headset, a smart watch, etc. Figure 2 The electronic devices 203 illustrated in FIG. 1 include a smart watch 203-1, a cellular phone 203-2, and a virtual reality (VR) headset 203-G. However, the electronic devices 203 may include different types of electronic devices other than the electronic devices shown. Each electronic device 203 includes a respective sensor 222-1, 222-2, 222-C, a respective battery 224-1, 224-2, 224-D, a respective processor 204-1, 204-2, 204-P, and a respective communication node 228-1, 228-2, 228-V, as well as other components not explicitly illustrated.

[0039] The central server 202 may transmit the initial model 205 for analyzing user data (e.g., user fitness data). The central server 202 may provide the initial model 205 by sending the initial model 205 to the electronic devices 203. For example, the central server 202 may send the initial model 205 to the first electronic device 203-1, the second electronic device 203-2, and the third electronic device 203-G. The initial model 205 may include weights, biases, and / or functions of the electronic device 203 to be used when performing daily functions (e.g., analyzing user fitness data). As described herein, each electronic device 203 may be configured to receive the initial model 205, collect corresponding user fitness data, determine corresponding quantization 213 used to train the initial model 205, and transmit the corresponding quantization data 213 and model updates 212 to the central server 202 without transmitting the user fitness data, which may include the user's corresponding personally identifiable information.

[0040] When the electronic device 203 is being used by the user, the sensor 222 of the electronic device 203 may collect data (e.g., fitness data) related to the use of the electronic device 203 and the user's status. For example, the electronic device 203 may collect information related to an inertial measurement unit (IMU), heart rate, body temperature, oxygen saturation, stress level, electrocardiogram (ECG), etc. Each electronic device 203 may use the data (e.g., fitness data) collected by the corresponding sensor 222 to train the initial model 205 received from the central server 202. When the user wears and / or holds the electronic device 203 and performs daily activities (e.g., relaxing, working, exercising, sleeping, etc.), the fitness data may be collected by the sensor 222. The data obtained by the sensor 222 may be used to create a model update 212 that can be used by the electronic device 203 when performing daily functions. For example, the sensor 222 may be a heart rate sensor for detecting the user's heart rate throughout the day when the user uses the electronic device 203. The electronic device may use the data collected by the heart rate sensor to train the initial model 205 and create a model update 212.

[0041] The model update 212 may be an update compiled into the initial model 205 based on the user-specific data. However, the model update 212 does not include the user's personally identifiable information. The electronic device 203 may use the user's fitness data to update the initial model 205 without incorporating the user's personally identifiable information into the model itself or transmitting the user's personally identifiable information with the updated model.

[0042] The training data 207 may also include quantified data 213. In some embodiments, in addition to data describing the environment around which the initial model was trained, the quantified data 213 may describe the environment around which the user's fitness data was collected. The quantified data 213 may describe environmental conditions, battery strength, time of day, strength of the processor 204 in the electronic device 203, etc. These factors may help determine the accuracy of the trained machine learning model 212 and the training data 207. For example, an electronic device 203-2 having a more powerful processing device 204-2 than another electronic device 203-G may create a more reliable model update 212-2 than an electronic device 203-G having a weaker processing device 204-P. For example, a processor 204-2 with stronger processing power will provide a more reliable model update 212-2 than a processor 204-P with weaker processing power.

[0043] In another example, a heart rate monitoring sensor may provide better tracking of heart rate at full battery power (rather than low battery power). In this example, the quantified data 213 may include the battery condition in which the information was collected. In addition, the quantified data 213 may include the weather conditions surrounding the collection of data. For example, if the electronic device 203 is collecting user fitness data related to oxygen saturation, then the quantified data 213 including weather information (e.g., sunny, rainy, etc.) may determine the accuracy and / or reliability of the collected data. Similarly, the quantified data 213 including information related to the user's activities (e.g., running, walking, sitting, etc.) when the data was collected may determine the importance level of the model update 212.

[0044] In some embodiments, the processor 204 causes the communication node 228 to communicate with the wireless network (eg, Figure 1 wireless network 108) and / or physical network (e.g. Figure 1 The training data 207 (including the quantified data 213 and the model update 212) is sent to the central server 202 without sending the user fitness data and / or the personally identifiable information by using the physical network 109 of the electronic device 203. In some embodiments, the electronic device 203 may transmit an indication of the quantified 213 based on the determined amount of memory available, the determined power state, and / or the determined processing power. The central server 202 may perform weight aggregation 220 on the training data 207 received from the electronic device 203. For example, the quantified data 213 from each electronic device 203 may be received by the central server 202. The central server 202 may analyze the quantified data 213 received from each electronic device 203 to determine the accuracy, reliability, and / or importance of the user fitness data without the personally identifiable information of the user of each electronic device 203. The central server 202 may then aggregate the corresponding received model updates 212 based on the quantified data 213 associated therewith to generate an updated model 210.

[0045] Aggregating the received data may include processing multiple sets of training data 207 received from different electronic devices 203. For example, the training data 207-1 provided by the electronic device 203-1 may be the first set of training data 207-1 (including the quantized data 213-1 and the model update 212-1), the training data 207-2 provided by the electronic device 203-2 (including the quantized data 213-2 and the model update 212-2) may be the second set of training data 207-2, and the training data 207-R provided by the electronic device 203-G (including the quantized data 213-T and the model update 212-S) may be the third set of training data 207-R. The central server 202 may process the first set of training data 207-1, the second set of training data 207-2, and the third set of training data 207-R to generate a single updated model 210. Once the updated model 210 is generated, the central server 202 may utilize a wireless network (e.g., Figure 1 wireless network 108) and / or physical network (e.g. Figure 1 The updated model 210 is sent to the electronic device 203 via the physical network 109 .

[0046] Figure 3 3 is a flowchart corresponding to a method 330 for training a machine learning model according to some embodiments of the present disclosure. In some examples, the method 330 may be performed using a computing system, such as Figure 1 and 2 The computing system described. At 331, the time of day when the electronic device is used may be determined. At 332, the power state of the electronic device may be determined. In some embodiments, the electronic device may determine the quantization used to train the initial machine learning model. The determined quantization may be represented by quantization data, which may indicate the accuracy, reliability and / or importance of the update to the initial model created by the electronic device due to training the initial model based on the user's fitness data collected from the sensor. For example, the amount of battery life that the electronic device has may indicate the amount of power that the processor has for training the model. The time of day when the training is performed may also indicate the accuracy of the training. For example, if the training is performed when the electronic device is usually busy performing other tasks, the accuracy of the training may be hindered. The central server may use the quantization data, including information related to the power state of the electronic device when the training is performed, to determine the importance of the model update, and when the updated model is generated, no personally identifiable information of the user is included. The central server may use the quantization data to determine how updates to the model are aggregated with other updates of other electronic devices into a global updated model.

[0047] At 333, the processing power of the electronic device may be determined. At 334, the amount of memory available in the electronic device may be determined. In some embodiments, the quantitative data may indicate the processing power of the processing device in the electronic device and the state of the memory in the electronic device. The capabilities of the processing device in the electronic device may determine the training accuracy of the initial model executed by the electronic device. An electronic device with a high-power processing device may provide more accurate results than an electronic device with a low-power processing device. The state of the memory may refer to the total amount of memory, the amount of free memory, the age or wear state of the memory, etc. The state of the memory of the electronic device may indicate the integrity of the collected data. For example, a sensor on an electronic device with low memory may save memory space by storing only a portion of the collected data. Using only a portion of the collected data to generate a model update may reduce its accuracy.

[0048] In some embodiments, when the quantized data is sent to the central server, the central server can use the quantized data to create an accurate and reliable updated model. The central server can prioritize each training data received when aggregating model updates received from multiple electronic devices. For example, the central server is further configured to calculate a corresponding weight parameter for each corresponding model update based on each corresponding quantized data and aggregate the corresponding model updates based on the corresponding weight parameters.

[0049] At 335, the activity being performed by the electronic device may be determined. In some embodiments, the activity being performed, environmental factors, and the state of the electronic device may be used to create quantified data. The level of importance of the collected data may vary depending on the activity being performed by the electronic device. For example, heart rate information obtained when a user is running (e.g., the electronic device is in exercise mode) may have a higher importance than heart rate information obtained when the user is resting. Therefore, the quantified data may provide information related to the level of importance of the training data when it is provided to the central server using information such as the activity of the electronic device. The quantified data may provide information related to the importance of the activity used to generate the user's fitness data without providing the user's fitness data itself and / or personally identifiable information.

[0050] At 336, user fitness data may be collected by the electronic device. In some embodiments, the user fitness data may be used to train an initial machine learning model. The electronic device may collect user fitness data using sensors based on the user using the electronic device.

[0051] At 337, the initial model may be trained using the user's fitness data in a quantitative manner. The training may result in model updates. In some embodiments, the electronic device may use machine learning to generate improvements to the initial model provided by the central server. For example, when the user uses the electronic device and / or provides input to the electronic device, the electronic device may learn user preferences and other more efficient ways to perform functions.

[0052] In some embodiments, the electronic device can learn a more efficient way to perform a function by analyzing the information collected by the sensor and the activities of the user and the electronic device. For example, the results of the sensor collecting the user's oxygen saturation data in rainy days and sunny days or when the user is running and the user is walking. Analyzing the differences in each situation can lead to a more efficient way for the electronic device to perform a function. In addition, the user can make choices related to the data collected by the sensor. The choices made by the user can be used by the electronic device to provide training to the electronic device and develop a trained machine learning model.

[0053] At 338, quantization and model updates to the initial machine learning model may be transmitted. The training data may be transmitted to the central server via a communication node located on the electronic device without transmitting user fitness data. The central server may aggregate model updates received from multiple electronic devices to develop updates to the initial model, thereby creating an updated model.

[0054] The quantified data may include information about the environment surrounding how the training of the initial machine learning model occurred. Examples include environmental conditions when the training occurred, battery conditions of the electronic device when the training occurred, charging cycles of the electronic device when the training occurred, processing power and / or memory levels of the electronic device when the training occurred, time of day when the training occurred, and / or activity levels of the user when the data was collected. The quantified data may help the central server determine how to combine training data received from multiple electronic devices. Based on the quantified data, the central server will determine the importance, accuracy, reliability, and / or relevance of model updates provided by multiple electronic devices in order to aggregate individual updates into a global updated model. The global updated model may then be sent to the electronic device. The electronic device may operate based on the global updated model received from the central server. In some instances, the electronic device may provide recommendations to the user based on the updated model.

[0055] Figure 4 is a flowchart 440 for training a machine learning model according to some embodiments of the present disclosure. Figure 4 Describes the use of user fitness data (while omitting personally identifiable information) and information about electronic devices (such as Figure 1 electronic device 103) quantitative data (e.g. Figure 1 quantification 113) to update the central server (e.g. Figure 1The initial model (e.g. Figure 1 At 441, flow chart 440 describes an example of an electronic device receiving an initial model (e.g., Figure 1 At 442, flow chart 440 describes the electronic device setting the initial model as the primary model. The central server can send the initial model to one electronic device or multiple electronic devices. In some embodiments, the type of initial model sent to the electronic device can depend on the type of electronic device. The processor of the electronic device (e.g. Figure 1 The processor 104 of the electronic device can use the initial model for the initial operation of the electronic device. The electronic device can use the initial model until the electronic device has a chance to collect user fitness data.

[0056] At 443, flowchart 440 describes determining the quantization level of the electronic device based on the resources used, the time, and the state of the electronic state. The electronic device may use resources (e.g., processing power, etc.), data collection time, the state of the electronic device, and other quantitative data to determine the quantization used to train the initial model. At 444, flowchart 440 describes monitoring the behavior of the user when using the electronic device. The user's behavior can be monitored to generate user fitness data. The quantization level of the electronic device can be determined before collecting the user fitness data, during the user fitness data collection, and / or immediately after collecting the user fitness data. The quantization level can provide a perspective on how to apply the user fitness data without providing personally identifiable information.

[0057] At 445, flow chart 440 describes using the monitored user behavior to determine quantitative training of a machine learning model of an electronic device. The machine learning model may be trained using user fitness data. For example, information from user activity may be used to train the machine learning model. However, the trained machine learning model will not include personally identifiable information that may be contained in the user fitness data. At 446, the electronic device may then compare the quantitative data collected by the electronic device with the model update (e.g., Figure 2 At 447, flowchart 440 describes calculating, by the central server, a respective weight parameter for each respective electronic device based on each respective quantization. At 448, flowchart 440 describes aggregating, by the central server, the respective updates based on the respective weight parameters of the quantization to create a global updated model (e.g., Figure 1 The central server uses the quantitative data received from each electronic device to determine the importance of the information collected when creating a global updated model of the electronic device.

[0058] At 449, the central server then sends the global updated model to the electronic device connected to the central server. In some embodiments, the global updated model sent to the electronic device may be based on the type of electronic device receiving the global updated model. For example, as described herein, the electronic device may receive the same global updated model or a global updated model customized for the type of electronic device receiving the global updated model. In some embodiments, the electronic device may set the global updated model as a base operation and train a machine learning model based on the global updated model.

[0059] Figure 5 is a block diagram of an example computer system 590 in which embodiments of the present disclosure may operate. For example, Figure 5 An example machine illustrating a computer system 590 within which a set of instructions for causing the machine to perform any one or more of the methodologies discussed herein may be executed. In some embodiments, the computer system 590 may correspond to a host system that includes, is coupled to, or utilizes a memory subsystem (e.g., Figure 1 The computer system 590 may be used to perform the operations described herein (eg, to perform operations corresponding to Figure 1 In some embodiments, the machine may be connected (e.g., using a network) to other machines. The machine may operate in the capacity of a server or a client machine in a client-server network environment, as a peer machine in a peer-to-peer (or distributed) network environment, or as a server or a client machine in a cloud computing infrastructure or environment.

[0060] The machine may be a personal computer (PC), a tablet PC, a set-top box (STB), a personal digital assistant (PDA), a cellular phone, a network appliance, a server, a network router, a switch or a bridge, or any machine capable of executing (sequentially or otherwise) a set of instructions that specify actions to be taken by the machine. Furthermore, while a single machine is described, the term "machine" shall also be taken to include any collection of machines that individually or jointly execute a set (or multiple sets) of instructions to perform any one or more of the methodologies discussed herein.

[0061] The example computer system 590 includes a processing device (e.g., a processor) 591, a main memory 593 (e.g., read-only memory (ROM), flash memory, dynamic random access memory (DRAM), such as synchronous DRAM (SDRAM) or Rambus DRAM (RDRAM)), etc.), a static memory 597 (e.g., flash memory, static random access memory (SRAM)), etc.), and a data storage system 598, which communicate with each other via a bus 596.

[0062] The processing device 591 represents one or more general-purpose processing devices, such as a microprocessor, a central processing unit, or the like. More specifically, the processing device 591 may be a complex instruction set computing (CISC) microprocessor, a reduced instruction set computing (RISC) microprocessor, a very long instruction word (VLIW) microprocessor, or a processor implementing other instruction sets or multiple processors implementing a combination of instruction sets. The processing device 591 may also be one or more special-purpose processing devices, such as an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), a digital signal processor (DSP), a network processor, or the like. The processing device 591 is configured to execute instructions 592 for performing the operations and steps discussed herein. The computer system 590 may further include a network interface device 594 to communicate over a network 595.

[0063] The data storage system 598 may include a machine-readable storage medium 599 (also referred to as a computer-readable medium) on which is stored one or more sets of instructions 592 or software embodying any one or more of the methodologies or functions described herein. The instructions 592 may also reside, completely or at least partially, within the main memory 593 and / or the processing device 591 during execution thereof by the computer system 590, the main memory 593 and the processing device 591 also constituting machine-readable storage media. The machine-readable storage medium 599, the data storage system 598, and / or the main memory 593 may correspond to Figure 1 Memory subsystems 111-1, 111-2, 111-U.

[0064] In one embodiment, instructions 592 include instructions for implementing instructions corresponding to determining a quantized training machine learning model (e.g., using Figure 1 1, 104-2, 104-P) of the processors 104-1, 104-2, 104-P). Although the machine-readable storage medium 599 is shown as a single medium in the example embodiment, the term "machine-readable storage medium" should be taken to include a single medium or multiple media storing one or more sets of instructions. The term "machine-readable storage medium" should also be taken to include any medium capable of storing or encoding a set of instructions for machine execution and causing the machine to perform any one or more of the methodologies of the present disclosure. Therefore, the term "machine-readable storage medium" should be taken to include, but not limited to, solid-state memory, optical media, and magnetic media.

[0065] Some portions of the foregoing detailed description have been presented in terms of algorithms and symbolic representations of operations on data bits within a computer memory. These algorithmic descriptions and representations are the means used by those skilled in the data processing arts to most effectively convey the substance of their work to others skilled in the art. An algorithm is generally conceived herein as a self-consistent sequence of operations leading to a desired result. Operations are those requiring physical manipulation of physical quantities. Typically, but not necessarily, these quantities take the form of electrical or magnetic signals capable of being stored, combined, compared, and otherwise manipulated. It has proven convenient at times, primarily for reasons of common usage, to refer to these signals as bits, values, elements, symbols, characters, terms, numbers, or the like.

[0066] It should be remembered, however, that all of these and similar terms should be associated with the appropriate physical quantities and are merely convenient labels applied to these quantities. The present disclosure may involve the actions and processes of computer systems or similar electronic computing devices that manipulate and transform data represented as physical (electronic) quantities within the computer system's registers and memories into other data similarly represented as physical quantities within the computer system's memories or registers or other such information storage systems.

[0067] The present disclosure also relates to an apparatus for performing the operations described herein. This apparatus may be specially constructed for the intended purpose, or it may comprise a general purpose computer selectively activated or reconfigured by a computer program stored in the computer. This computer program may be stored in a computer-readable storage medium, such as, but not limited to, any type of disk, including floppy disks, optical disks, CD-ROMs and magneto-optical disks, read-only memories (ROMs), random access memories (RAMs), EPROMs, EEPROMs, magnetic or optical cards, or any type of medium suitable for storing electronic instructions, each coupled to a computer system bus.

[0068] The algorithms and displays presented herein are not inherently related to any particular computer or other device. Various general purpose systems may be used in conjunction with the programs according to the teachings herein, or it may prove convenient to construct more specialized equipment to perform the methods. The structures of various of these systems will appear as set forth in the description below. In addition, the present disclosure is not described with reference to any particular programming language. It should be appreciated that various programming languages ​​may be used to implement the teachings of the present disclosure described herein.

[0069] The present disclosure may be provided as a computer program product or software that may include a machine-readable medium having instructions stored thereon, which instructions may be used to program a computer system (or other electronic device) to perform a process according to the present disclosure. A machine-readable medium includes any mechanism for storing information in a form that can be read by a machine (e.g., a computer). In some embodiments, a machine-readable (e.g., computer-readable) medium includes a machine (e.g., computer) readable storage medium, such as a read-only memory ("ROM"), a random access memory ("RAM"), a magnetic disk storage medium, an optical storage medium, a flash memory device, etc.

[0070] In the foregoing description, embodiments of the present disclosure have been described with reference to specific example embodiments of the present disclosure. It should be understood that various modifications may be made to the present disclosure without departing from the broader spirit and scope of the embodiments of the present disclosure as set forth in the appended claims. Therefore, the description and drawings should be viewed in an illustrative rather than a restrictive sense.

Claims

1. An electronic device for training a machine learning model, comprising: Battery (224); Communication node (228); a sensor (222) configured to collect user fitness data; and a processor (204) coupled to the battery, the communication node, and the sensor, wherein the processor is configured to: monitoring a power state of the battery; determining a usage pattern of the electronic device; determining, based on the power state and the usage pattern, a quantification to employ for training the machine learning model (106) for analyzing the user fitness data; and The determined quantification and updates to the machine learning model are transmitted via the communication node (338) without transmitting the user fitness data. 2 . The electronic device of claim 1 , wherein the processor is further configured to determine the quantification based on a time of day. 3 . The electronic device of claim 2 , wherein the processor is further configured to determine the quantification based on memory availability of the electronic device. 4 . The electronic device of claim 3 , wherein the processor is further configured to determine the quantification based on a time remaining before a next charge of the battery according to the usage pattern. The electronic device of claim 4 , wherein the processor is further configured to determine the quantification based on environmental conditions. 6 . The electronic device of claim 1 , wherein the user fitness data includes one or more of heart rate, body temperature, and oxygen saturation.

7. The electronic device of any one of claims 1 to 5, wherein the processor is configured to: receiving the machine learning model as an initial model via the communication node; and An updated model is received via the communication node after transmitting the determined quantification and the update to the machine learning model.

8. The electronic device of claim 7, wherein the processor is further configured to provide recommendations to a user based on the updated model and based on the user fitness data.

9. A system for training a machine learning model, comprising: A server (102) configured to transmit an initial model (105) for analyzing user fitness data; and A plurality of electronic devices (103), each having different hardware specifications and each configured to: Receiving the initial model (106); Collect corresponding user fitness data; determining a corresponding quantization to be used for training the initial model (113); and transmitting the corresponding quantification and the corresponding update (112) to the initial model to the server without transmitting the corresponding user fitness data; The server is further configured to: aggregating the respective updates based on the respective quantifications to create an updated model (110); and The updated model is transmitted to the plurality of electronic devices.

10. The system of claim 9, wherein the server is further configured to calculate a respective weight parameter of each respective trained machine learning model based on each respective quantization; and The respective updates are aggregated based on the respective weight parameters.

11. The system of claim 9, wherein the plurality of electronic devices are each configured to operate based on the updated model.

12. The system of claim 9, wherein each of the plurality of electronic devices is configured to: Collect additional corresponding user fitness data; and The updated model is trained with the additional corresponding user fitness data at the determined quantification.

13. The system of claim 9, wherein each of the plurality of electronic devices is configured to omit personally identifiable information of the user when transmitting data to the server.

14. The system of claim 9, wherein each of the plurality of electronic devices is configured to collect the respective user fitness data in an amount that depends on a power state of the electronic device.

15. A method for training a machine learning model, comprising: Determine the time of day (331); determining a power state of the electronic device (332); determining a processing power of the electronic device (333); determining an amount of memory available in the electronic device (334); determining an activity being performed by the electronic device (335); Collecting user fitness data by the electronic device (336); Training the machine learning model (337) using the user fitness data based on the quantification of: the time of day; The available amount of the memory; the power state; and the processing power; and The quantification and updates to the machine learning model are transmitted (338).

16. The method of claim 15, wherein transmitting the updates to the machine learning model comprises transmitting updated weights. 17 . The method of claim 15 , further comprising receiving a global updated model based on an aggregation of updates for a plurality of electronic devices.

18. The method of claim 17, further comprising operating the electronic device based on the global updated model.

19. The method of claim 17, further comprising providing recommendations to a user based on the global updated model.

20. The method of claim 17, comprising transmitting the quantification and the update without transmitting the user fitness data.

Images