Wearable device, program for wearable device, and deep layer learning method of wearable device

Abstract

To provide a wearable device, a program for the wearable device, and a deep layer learning method of the wearable device capable of reducing a data amount and a communication amount by performing individual deep layer learning, in which a risk in security is small.SOLUTION: A wearable device D having an information input part 13 for inputting data on a wearer includes: a rechargeable battery 110; a learning part 14 for performing deep layer learning on the basis of the data; a temperature detection part 111 for detecting the temperature of the battery 110; and an execution control part 15 for switching between interruption and resumption of the execution of the deep layer learning by the learning part 14 according to the temperature detected by the temperature detection part 111.SELECTED DRAWING: Figure 2

Description

[Technical field]

[0001] The present invention relates to a wearable device, a program for a wearable device, and a deep learning method for a wearable device. [Background technology]

[0002] Conventionally, in areas where individuals perform physically demanding work, such as construction sites, factories, and warehouses, there has been a demand for determining the condition of individuals and preventing heatstroke, etc. One possible method for determining such an individual's condition is to obtain data including the individual's biometric information, such as pulse rate, body temperature, and number of steps, using a wearable device worn by the individual, and to issue an alert if the data exceeds a predetermined threshold.

[0003] However, what values ​​of these data should be considered abnormal varies greatly from person to person. Even for the same person, the values ​​that should be considered abnormal vary depending on the individual's environment, such as temperature and humidity. This makes it difficult to set optimal threshold values ​​and makes it difficult to make accurate judgments.

[0004] To address this issue, it is conceivable to accumulate personal data for a certain period of time, subject the characteristics of this data to deep learning, and then determine whether newly acquired data is normal or abnormal based on the obtained characteristics. This would enable accurate determination according to individual differences and differences in environment. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] JP 2023-12396 A Summary of the Invention [Problem to be solved by the invention]

[0006] Such deep learning is generally performed on a server. However, performing deep learning of biometric information acquired from a wearable device on a server has the following problems. (a) Large amounts of data must be temporarily sent to the server, which incurs communication costs. In addition, the effectiveness of the system depends on the communication environment. (b) As the number of targeted individuals increases, the capacity and processing load required for the server increases. (c) This data may be considered privacy data, which raises security concerns and means that the data cannot be sent out and processed by the server unless consent is obtained.

[0007] The present invention aims to provide a wearable device, a program for a wearable device, and a deep learning method for a wearable device that can reduce data volumes and communication volumes by performing deep learning individually and has fewer security concerns. [Means for solving the problem]

[0008] In order to achieve the above-mentioned objectives, the present invention comprises an information input unit that inputs data of the wearer, a temperature detection unit that detects the temperature of the battery, a learning unit that performs deep learning based on the data, and an execution control unit that switches between suspending and resuming the execution of the deep learning by the learning unit depending on the temperature detected by the temperature detection unit.

[0009] The present invention also relates to a program to be executed by a computer possessed by a wearable device, which executes a learning process that performs deep learning based on data of a wearer of the wearable device, and an execution control process that switches between suspending and resuming execution of the learning process depending on the temperature of the battery.

[0010] In addition, the deep learning method for a wearable device of the present invention includes a learning process in which the wearable device performs deep learning based on data of the wearer of the wearable device, and an execution control process that switches between suspending and resuming execution of the learning process depending on the temperature of the battery. Effect of the Invention

[0011] According to the present invention as described above, by performing deep learning individually, the amount of data and communication can be reduced, and there are fewer security concerns. [Brief description of the drawings]

[0012] [Figure 1] 1 is an overall configuration diagram showing a system configured by a wearable device in an embodiment. [Diagram 2] FIG. 1 is a block diagram showing a wearable device according to an embodiment. [Diagram 3] 11 is a graph showing an example of switching between execution and non-execution of deep learning depending on the battery temperature. [Figure 4] 11 is a graph showing an example of switching between execution and non-execution of deep learning depending on the battery temperature. [Diagram 5] 11 is a graph showing an example of switching between execution and non-execution of deep learning depending on the battery temperature. [Figure 6] 4 is a flowchart showing a processing procedure according to an embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0013] [overview] An embodiment for carrying out the present invention (hereinafter referred to as "the present embodiment") will be specifically described with reference to the drawings. The present embodiment is applied to a wearable device and a program for a wearable device that constitute a deep learning system. A deep learning method for a wearable device that performs the processing described below is also one aspect of the present invention. Note that the present embodiment can also be regarded as a status display device that displays the status of a wearer of a wearable device.

[0014] The term "wearing" refers to a state in which the device can follow the wearer's movements to an extent that the wearer's condition can be identified. For this reason, the device may be in direct contact with the wearer, or may be in contact with clothing or other materials between the device and the wearer.

[0015] As shown in FIG. 1, a deep learning system S includes wearable devices D1 to Dn (hereinafter simply referred to as wearable device D), a server C, an access point P, and a gateway G. The wearable devices D are information communication terminals capable of transmitting and receiving information to and from each other or to and from the server C via a network N. The wearable device D includes a main unit U and a mounting part B that is attached to the human body. For example, a wristwatch-type information communication terminal is used as the wearable device D. In this case, the main unit U includes a computer equipped with a processing part, which will be described later, in a thin housing. The mounting part B is a band that is wrapped around the arm.

[0016] The server C is a device capable of transmitting and receiving information to and from the wearable devices D via the network N. The server C collects information from each wearable device D, and based on this information, the state of the wearer can be determined.

[0017] The access point P is a relay device that relays communication between the network N and the wearable device D. The access point P is typically a WiFi wireless LAN access point. The gateway G has a function of connecting the wearable devices D to each other by short-range wireless communication such as Bluetooth (registered trademark).

[0018] The wearable device D, the server C, the access point P, and the gateway G are configured by a computer controlled by a program. The computer is configured by a processor such as a CPU, a storage medium such as a memory, etc., but the implementation of the hardware and the program can be changed in various ways.

[0019] Various settings, arithmetic expressions, parameters, etc. required for the processing of each unit described later are stored in advance in a built-in or removable storage medium. A storage unit 19 described later can be configured as a part of the storage area of ​​such a storage medium.

[0020] In the following description, a functional block diagram illustrating each function is used. Note that the present invention can also be understood as a program for causing a computer to execute each process, or a computer-readable storage medium having such a program recorded thereon.

[0021] [Wearable devices] As described above, the wearable device D is a device that is worn on a part of the body of a wearer, that is, a person performing an action. As shown in Fig. 2, the wearable device D has a communication unit 10, a battery 110, a temperature detection unit 111, a connection unit 12, an information input unit 13, a learning unit 14, an execution control unit 15, a threshold setting unit 16, a determination unit 17, an information notification unit 18, and a storage unit 19.

[0022] (Communications Department) The communication unit 10 is a processing unit that transmits and receives information to and from the outside. The communication unit 10 can transmit and receive information to and from the server C and other wearable devices D via a network N or a gateway G. The network N may have any of a many-to-one, many-to-many, or one-to-one relationship. For example, the communication unit 10 has a function of connecting the server C and the wearable devices D to each other via a wireless LAN such as WiFi, WiFi Direct, or the like.

[0023] The communication unit 10 transmits the feature amount, the evaluation value of the threshold value, and the like learned by the learning unit 14 together with the identification information of each wearable device D to the server C or another wearable device D.

[0024] (battery) The battery 110 is a rechargeable power source. In other words, the battery 110 is a battery that can be used repeatedly by charging, and is also called a secondary battery, a rechargeable battery, or a storage battery. The battery 110 functions as a power source for each part that constitutes the wearable device D. If the temperature of the battery 110 rises excessively, processing will stop and the life of the battery will deteriorate. However, the heat resistance temperature differs depending on the type and performance of the battery 110. For example, lithium ion batteries are commonly used as the battery 110 of the wearable device D, but are relatively vulnerable to heat. The battery 110 of this embodiment is not limited to such batteries, but is suitable for batteries that are vulnerable to heat.

[0025] (Temperature detection section) The temperature detection unit 111 detects the temperature of the battery 110. The temperature detection unit 111 is connected to the battery 110, measures the resistance value, and obtains the temperature by calculation based on the measured resistance value. Such an embodiment is also included in "detecting the temperature." However, the temperature detection unit 111 may be a sensor that directly measures the temperature of the battery 110.

[0026] (Connection) The connection unit 12 is an interface that connects the battery 110 to an external power supply source. The connection unit 12 may be a member for wired connection, such as a conductive terminal connected to the battery 110, or a member for wireless connection, such as a power receiving coil. For example, as shown in Fig. 1, charging of the battery 110 begins when the connection unit 12 of the wearable device D is attached to a charger E whose power plug is connected to an outlet.

[0027] (Information input section) The information input unit 13 is a processing unit that inputs data of the wearer of the wearable device D. The data of the wearer in this embodiment includes the wearer's motion information and biometric information. Hereinafter, information including at least one of the motion information and biometric information will be referred to as biometric information. The data in this embodiment includes biometric information. The information input unit 13 has an acceleration sensor 131, a geomagnetic sensor 132, and a gyro sensor 133 as input means for motion information, and a pulse sensor 134 and a temperature sensor 135 as input means for biometric information.

[0028] The acceleration sensor 131 is a sensor that senses tilt and vibration and converts it into an electric signal according to the degree of tilt and vibration. The acceleration sensor 131 is a sensor that outputs signals in three mutually orthogonal axial directions: X-axis, Y-axis, and Z-axis. The geomagnetic sensor 132 is a sensor that senses the magnitude and direction of a magnetic field and converts it into an electric signal according to the direction. The gyro sensor 133 is a sensor that detects angular velocity in the three axial directions of X, Y, and Z.

[0029] The pulse sensor 134 is a sensor that comes into contact with the wearer and detects the pulse of the wearer. For example, a sensor that detects the pulse by irradiation and reflection of ultraviolet light can be used as the pulse sensor 134. In this embodiment, the pulse detected by the pulse sensor 134 is used as the heart rate of the worker. The temperature sensor 135 is a sensor that detects the body temperature of the wearer. The temperature sensor 135 is provided separately from the temperature detection unit 111 that detects the temperature of the battery 110.

[0030] (Learning Department) The learning unit 14 performs deep learning based on the input data (learning process). In this embodiment, the learning unit 14 performs deep learning while the battery 110 is charging. The learning unit 14 starts deep learning when it detects that the battery 110 has started charging. For example, deep learning uses a VAE (variational autoencoder), which is a neural network. The VAE encodes (compresses) the input data into a specific latent space, and decodes (reconstructs) the original data from the latent space.

[0031] This VAE has two main parts: an encoder that performs encoding and a decoder that performs decoding. The encoder receives input data, converts it into two parameters, the mean and standard deviation, and reduces it to a latent variable. At this time, constraints are added so that the latent variable follows a normal distribution. The decoder reconstructs the original input data based on the latent variable. Features are parameterized by having the VAE learn to make the input and output the same for actual measured values.

[0032] (Executing Control Unit) The execution control unit 15 switches between interrupting and resuming the execution of learning by the learning unit 14 according to the temperature detected by the temperature detection unit 111 (execution control process). That is, the execution control unit 15 executes deep learning while monitoring the temperature of the battery 110. More specifically, the execution control unit 15 switches between execution and non-execution based on a first threshold value that is the upper limit temperature and a second threshold value that is the lower limit temperature. After the start of deep learning by the learning unit 14, if the temperature reaches the first threshold value that is the upper limit, the execution control unit 15 interrupts deep learning by the learning unit 14, and if the temperature reaches the second threshold value that is the lower limit, the execution control unit 15 resumes deep learning by the learning unit 14. Note that the first threshold value that is the upper limit is a threshold value on the high temperature side, and the actual temperature of the battery 110 may be higher than the first threshold value. The second threshold value that is the lower limit is a threshold value on the low temperature side, and the actual temperature of the battery 110 may be lower than the second threshold value. The first threshold value is higher than the second threshold value, and the second threshold value is lower than the first threshold value.

[0033] When the learning unit 14 starts calculations based on deep learning, the temperature rises relatively quickly. When the calculations are stopped, the temperature drops relatively slowly. However, the rate at which the temperature rises and falls varies depending on the room temperature and the season. When the amount of deep learning is small, the temperature can be kept low. The first threshold is set to a value lower than the temperature at which the wearable device D stops. The second threshold is set to a temperature lower than the first threshold, so that the temperature does not continue to stagnate near the first threshold.

[0034] Here, Figs. 3 to 5 show an example of temperature transition when deep learning is performed on the heart rate per minute of the wearer of the wearable device D during charging. In Figs. 3 to 5, the vertical axis is the battery temperature [°C], and the horizontal axis is the time [hours:minutes:seconds]. Note that the horizontal axis does not completely display the deep learning from start to finish. Fig. 3 shows an example in which the first threshold is 48°C, the second threshold is 43°C, and the time required from start to finish of deep learning is 52 minutes. Fig. 4 shows an example in which the first threshold is 53°C, the second threshold is 48°C, and the time required from start to finish of deep learning is 34 minutes. Fig. 5 shows an example in which the first threshold is 58°C, the second threshold is 53°C, and the time required from start to finish of deep learning is 26 minutes.

[0035] As shown in Figures 3 to 5, after the start of deep learning, when the first threshold is reached, the temperature drops due to the interruption of deep learning, and when the second threshold is reached, the temperature rises due to the start of deep learning. This phenomenon is repeated. It can be seen that the time required for deep learning changes by changing the settings of the first and second thresholds.

[0036] When the amount of data and calculations to be performed in deep learning is large, the temperature is likely to become high. Since it is preferable that the temperature is low when the wearable device D is in use, for example, it is preferable that the execution control unit 15 stops learning when the time is a predetermined time before the scheduled wearing time set in advance. Note that there are cases where the above-mentioned control does not function. For example, if the amount of data is such that the battery temperature does not reach the upper limit, deep learning can be performed from start to finish without interruption and resumption.

[0037] Furthermore, when deep learning is interrupted, the execution control unit 15 stores the parameters generated by deep learning together with the loss value in the memory unit 19 (storage process). The loss value is a value that represents the difference between a target output and an actual output, and is calculated each time deep learning is performed by the learning unit 14. The smaller the loss value, the more accurately the target output can be generated. When deep learning is resumed, the execution control unit 15 continues deep learning based on the parameters with the smaller loss value. For example, when use of the wearable device D is started during deep learning and new data is input, the execution control unit 15 compares the loss value of the parameters based on the new data with the loss value of the parameters based on the data up to that point, and uses the parameters with the smaller loss value.

[0038] Note that when the execution control unit 15 detects that the battery 110 has started charging, the execution control unit 15 may cause the learning unit 14 to start deep learning. In this case, when the temperature detected by the temperature detection unit 111 is equal to the first threshold value, the execution control unit 15 may wait until the temperature reaches the second threshold value before starting deep learning. In other words, the execution control unit 15 may switch between starting, suspending, and resuming the execution of deep learning by the learning unit 14 according to the temperature detected by the temperature detection unit 111.

[0039] (Threshold setting section) The threshold setting unit 16 sets a first threshold and a second threshold (threshold setting process). It is preferable to set the first threshold so that the number of interruptions is not excessive, and the second threshold so that a situation does not occur where the operation is not resumed even when the temperature drops. It is also preferable to minimize damage to the battery 110. However, it is not easy to determine what values ​​are optimal for the first threshold and the second threshold. Therefore, the threshold setting unit 16 determines the optimal values ​​using an evaluation value based on the following two values ​​(A) and (B) obtained as a result of actual learning.

[0040] (A) Time required for deep learning (seconds) This is the time from the start to the end of learning, and is based on the fact that it is preferable that the time required for learning is short as a result. (B) α×Σ(battery temperature – reference temperature) (seconds) This is based on the fact that it is preferable that the time during which the battery temperature exceeds the reference temperature is short. Σ(battery temperature-reference temperature) (sec) is the integral value (amount exceeded x time (sec)) of the time (sec) during which the battery temperature exceeds the reference temperature. The reference temperature is a reference temperature at which the battery 110 can be considered to be high. This reference temperature is set to prevent the battery 110 from reaching a temperature higher than or equal to a temperature at which the battery 110 is damaged, and is therefore set to a temperature lower than the temperature at which the battery 110 is damaged. The reason for setting the reference temperature is to avoid a relative difference becoming small in the range from 0 degrees, which reduces the resolution and causes the temperature to become dull. As a guideline for the reference temperature, since the value depends on the room temperature in a stopped state and is too low, it is sufficient to set it to about the temperature during normal operation. α is a coefficient set according to the amount of damage that the battery 110 can tolerate. In other words, it is set to a large value when damage is to be prevented, and to a small value when damage is to be tolerated to a certain extent.

[0041] Examples of such evaluation values ​​are described below. For example, in the cases of FIG. 3 (first threshold 48°C, second threshold 43°C), FIG. 4 (first threshold 53°C, second threshold 48°C), and FIG. 5 (first threshold 58°C, second threshold 53°C), the evaluation values ​​of (A)+(B) were as follows: Figure 3: 3090+4443=7533 Figure 4: 2038+5311=7349 Figure 5: 1551+5924=7475 Since the evaluation value is the lowest in the case of FIG. 4, threshold value setting unit 16 sets the first threshold value to 53°C and the second threshold value to 48°C.

[0042] The threshold setting unit 16 sets the first threshold and the second threshold when the evaluation value calculated when the first threshold and the second threshold are changed within a predetermined range is the lowest. For example, based on data already stored in the storage unit 19, the evaluation value when learning is performed using the first threshold and the second threshold set by default, and the first threshold and the second threshold are adjusted (changed) within a predetermined range. Then, the evaluation value when learning is performed each time the adjustment is performed is obtained, and the first threshold and the second threshold that are the lowest evaluation value are set as the optimal threshold. At this time, it is preferable to adjust the first threshold and the second threshold under set conditions such as the upper limit of the first threshold (for example, 60°C) and the difference between the first threshold and the second threshold (for example, 5°C). Alternatively, the upper limit and the lower limit of the difference between the first threshold and the second threshold may be set, and the adjustment may be performed within the range of the difference.

[0043] Such threshold setting may be performed periodically by the wearable device D alone. However, the threshold setting unit 16 may compare the evaluation value calculated by itself with the evaluation value acquired from the other wearable devices D via the communication unit 10, and select and set the first threshold and the second threshold with the lowest evaluation value. In this way, the optimal first threshold and the second threshold can be set and shared without imposing a processing load on each wearable device D, so that it is also possible to perform the setting every time deep learning is executed.

[0044] (Judgment Department) The determination unit 17 determines whether the wearer's condition is normal or abnormal (determination process). For example, when new data that has not been used for learning is input to the VAE, if data with similar characteristics can be generated, it can be determined to be normal, and if a different output is produced, it can be determined to be abnormal. This allows the state of the wearer of the wearable device D to be determined.

[0045] (Information Notification Department) The information notification unit 18 notifies the user of information such as a warning (alert) in a manner that can be identified by the user according to the result of the determination by the determination unit 17 (notification process). For example, when the determination unit 17 determines that there is an abnormality, it can notify the user of a warning. The information notification unit 18 has an audio output unit 181, a vibration unit 182, and a display unit 183. The audio output unit 181 is a means for transmitting information to the wearer by voice, such as a speaker or a buzzer. The vibration unit 182 is a means for transmitting information to the wearer through contact with the wearer. The vibration unit 182 can be, for example, a vibrator that generates vibrations using a motor as a driving source. The vibration unit 182 can output intermittent vibrations according to a pulse signal, not limited to continuous vibrations, according to the information. The display unit 183 is, for example, a display that displays an image that the user visually recognizes, such as a still image, a video, or text.

[0046] (Storage part) The storage unit 19 stores information necessary for the processing of each unit. The information stored in the storage unit 19 includes data input from the information input unit 13, a first threshold value, a second threshold value, a battery temperature, a coefficient α, a reference temperature, an evaluation value, an adjustment width, a difference value, a range of the difference value, and the like. The wearable device D has an input unit for inputting various information necessary for the processing of each unit. The input unit includes a touch panel, an operation button, and the like.

[0047] [Processing Procedure] The processing procedure of this embodiment as described above will be described with reference to the flowchart of FIG. 6 in addition to the above drawings.

[0048] First, while the wearable device D is being used, data of the wearer wearing the wearable device D is continuously input by the information input unit 13 at predetermined time intervals and stored in the storage unit 19. For example, the storage unit 19 stores the heart rate per minute.

[0049] Then, after finishing use, the wearer attaches the connection unit 12 of the wearable device D to the charger E to start charging the battery 110 (step S101). The learning unit 14, which detects the start of charging, starts deep learning (step S102). For example, deep learning is performed based on the heart rate per minute stored in the memory unit 19. After starting deep learning, the execution control unit 15 performs an execution control process that switches between suspending and resuming execution of the learning process according to the temperature of the battery 110. That is, when the battery temperature detected by the temperature detection unit 111 rises and reaches a first threshold value (YES in step S103), the execution control unit 15 causes the learning unit 14 to suspend deep learning (step S104).

[0050] When the battery temperature detected by the temperature detection unit 111 drops and reaches the second threshold value (YES in step S105) due to interruption of deep learning, the execution control unit 15 causes the learning unit 14 to resume deep learning (step S106). When data to be used for deep learning remains (YES in step S107), deep learning is continued (step S103 and thereafter). When there is no data to be used for deep learning (NO in step S107), the learning unit 14 ends deep learning (step S108).

[0051] [Effects] (1) This embodiment is a wearable device D having an information input unit 13 for inputting data of a wearer, and includes a rechargeable battery 110, a learning unit 14 that performs deep learning based on the data, a temperature detection unit 111 that detects the temperature of the battery 110, and an execution control unit 15 that switches between whether or not deep learning is performed by the learning unit 14 depending on the temperature detected by the temperature detection unit 111. In this embodiment, the learning unit 14 performs deep learning while the battery 110 is charging.

[0052] In addition, this embodiment is a program executed by a computer included in the wearable device D, which executes a learning process for performing deep learning based on data of the wearer, and an execution control process for switching between suspending and resuming execution of the learning process according to the temperature of the battery 110. In this embodiment, the learning process is performed while the battery 110 is charging.

[0053] In this way, when the temperature of the battery 110 rises due to deep learning, the deep learning can be interrupted, and when the temperature drops, the deep learning can be resumed. This prevents the wearable device D from stopping and the deep learning from being unable to continue due to an excessive rise in temperature of the battery 110, thereby preventing the battery 110 from deteriorating.

[0054] Typically, deep learning is performed on server C because it requires processing large amounts of data. However, in recent years, the processing performance of the hardware in wearable devices D has improved, making deep learning possible at the level of the data volume of the individual wearing the device. This trend is expected to continue to advance in the future. In addition, the algorithms and models of neural networks in deep learning have evolved, making it possible to clarify unnecessary layers, neurons, nodes, etc. and omit them. This has resulted in streamlining of calculations and lighter processing loads.

[0055] However, when deep learning is performed, power consumption increases, and the battery capacity becomes insufficient. Therefore, the inventors thought of preventing the battery from running out by performing deep learning while the wearable device D is charging. However, a phenomenon occurred in which the wearable device D stopped even when deep learning was performed while charging. The inventors found that this was caused by heat generation from the processor, and after extensive research, they arrived at the present embodiment as a technology to solve this problem.

[0056] In other words, deep learning consumes a large amount of power from the battery 110 due to the large amount of calculations. Therefore, performing deep learning while the battery 110 is in operation tends to cause the battery to run out. Therefore, it was decided to perform deep learning in a state where power is shared while the battery 110 is being charged. However, when deep learning is performed, heat is generated from devices such as the computer circuit and the battery 110. The heat damages the device and causes it to break down, or accelerates the deterioration of the battery 110. If this heat leads to an abnormality in the battery 110, there is a risk of a fire accident. In addition, if the system exceeds the guaranteed operating temperature, it will be forced to stop and will become uncontrollable. The heat increases depending on the time required for calculation, and is affected by heat sources such as room temperature, air flow, and the surrounding wearable device D. Therefore, a deep learning method that takes heat into account is required for the wearable device D.

[0057] In this embodiment, as described above, the battery temperature is referred to as the internal temperature, and an excessive temperature rise is prevented, making it possible to efficiently terminate deep learning and suppress damage to the device and the battery. Such an effect can be obtained whether deep learning is performed during charging or not (when the battery 110 is in operation), but in particular, by performing deep learning during charging, the time when the wearable device D is not in use can be effectively utilized. In addition, since there is no need to transmit a large amount of data to the server C, communication costs are not incurred. Furthermore, it can be executed even in a poor communication environment. Furthermore, there is no need to transmit private data to the outside, and security can be ensured.

[0058] Biological information such as the heart rate and movements of the wearer of the wearable device D has different characteristics for each person. For this reason, an approach that allows learning according to individual characteristics is more appropriate than statistical processing. In addition, biological information changes depending on the season, health condition, muscle strength, physical strength, work habituation, etc., so daily continuous learning is important. In addition, since it is possible to follow the state by learning the obtained data, it is compatible with learning in the wearable device D.

[0059] (2) After causing the learning unit 14 to start performing deep learning, the execution control unit 15 causes the learning unit 14 to suspend deep learning when a first threshold value, which is an upper limit, is reached, and causes the learning unit 14 to resume deep learning when a second threshold value, which is a lower limit, is reached.

[0060] The execution control process interrupts deep learning when a first threshold value, which is an upper limit, is reached after the deep learning starts, and resumes deep learning when a second threshold value, which is a lower limit, is reached.

[0061] In this way, by setting not only the first threshold value, which is the upper limit, but also the second threshold value, which is the lower limit, deep learning can be intermittently executed until completion. In particular, by performing deep learning during charging, deep learning can be completed during the time when the wearable device D is not being used. By appropriately setting the first threshold value, frequent interruptions can be prevented. By appropriately setting the second threshold value, it is possible to prevent the learning from not being resumed even if the temperature drops.

[0062] (3) The wearable device D has a threshold setting unit 16 that sets a first threshold and a second threshold based on an evaluation value calculated from the time required for deep learning and the time at which the temperature of the battery 110 becomes higher than a reference temperature.

[0063] The program for a wearable device causes the computer to execute a threshold setting process that sets a first threshold and a second threshold based on an evaluation value calculated from the time required for deep learning and the time it takes for the temperature of the battery 110 to become higher than a reference temperature.

[0064] Therefore, the first and second threshold values ​​can be set so that the test is completed in a short time and causes little damage to the battery 110.

[0065] (4) The wearable device D has a memory unit 19 that stores the parameters generated by deep learning together with the loss value when the learning unit 14 interrupts deep learning, and when the learning unit 14 resumes deep learning, it performs deep learning based on the parameters with the smaller loss value.

[0066] The program for the wearable device causes the computer to execute a storage process to store the parameters generated by deep learning together with the loss value when deep learning is interrupted, and when deep learning is resumed, continues deep learning based on the parameters with the smaller loss value.

[0067] This makes it possible to achieve highly accurate learning even if deep learning is interrupted and restarted.

[0068] (5) Wearable device D has a judgment unit 17 that judges whether the wearer's condition is normal or abnormal based on the results of deep learning by learning unit 14.

[0069] The program for the wearable device causes the computer to execute a determination process to determine whether the wearer's condition is normal or abnormal based on the results of deep learning.

[0070] As a result, deep learning can be performed only on each individual wearable device D, and the wearer's condition can be determined based on that, making it possible to make highly accurate determinations without disclosing private data to the outside.

[0071] [Variations] (1) Deep learning tools and data subject to deep learning are not limited to the examples given above.

[0072] (2) The determination of normality or abnormality by the determination unit 17 and the notification by the information notification unit 18 are not limited to the above-mentioned aspects. For example, the determination unit 17 may predict a change in state by time series prediction, determine that a situation that is likely to be avoided is abnormal, and the information notification unit 18 may notify with a warning or the like. The determination unit 17 may also obtain a heart rate using a movement (frequency component of acceleration) or a step pitch as an input value, compare it with the heart rate based on the pulse, and determine that a situation in which the correlation between data is different from normal is abnormal, and the information notification unit 18 may notify with a warning or the like.

[0073] (3) Between the wearable devices D, and between each wearable device D and the server C, the evaluation values ​​based on the respective deep learning and the parameters obtained by learning are shared via the network N and the gateway G, so that the judgment unit 17 can judge whether or not the abnormality is due to some external factor or the work content. By judging whether or not the abnormality is only of the person in question, the importance and danger can be judged. The evaluation values ​​and parameters do not directly become private data, and the original data cannot be obtained based on them. For this reason, data such as the evaluation values ​​and parameters can be shared via the network N and the gateway G to be easily handled. This can be useful not only for individual judgments of the wearable devices D, but also for overall analysis.

[0074] (4) For the reasons described above, it is preferable that the deep learning by the learning unit 14 is performed during charging. However, it is not limited to being performed during charging. The learning unit 14 may perform deep learning not during charging, but when the wearer is wearing and using the wearable device D. In this way, even when deep learning is performed during use, the execution control unit 15 switches between suspending and resuming the execution of learning by the learning unit 14 according to the temperature detected by the temperature detection unit 111, thereby preventing the wearer from becoming uncomfortable due to heat or from functioning out. This is effective, for example, in the early stages when the amount of data is relatively small, or when it is necessary to reset the data after completing deep learning for some reason (when an unspecified number of people use the device, when data cannot be saved due to contractual reasons).

[0075] (5) The wearer of the wearable device D may be any living organism capable of measuring acceleration signals, steps, and heart rate. For this reason, the wearer may be an animal other than a human. The form of the wearable device D is not limited to a wristwatch type. In other words, the wearer may wear the device on the body to the extent that the wearer can obtain acceleration signals, steps, and heart rate that can identify the movement. The device may be worn on any part of the body, such as the arm, fingers, head, neck, chest, waist, or foot. For this reason, the wearer may be a band, belt, string, glasses, earplugs, adhesive tape, double-sided adhesive sheet, hook-and-loop fastener, or the like. Clothing, hats, supports, accessories, footwear, or the like may be used as the wearer. The main unit may not directly touch the skin. In other words, the contact for transmitting information may be direct to the skin or indirect through a part of the wearer.

[0076] [Other embodiments] Although the embodiment of the present invention and the modified examples of each part have been described above, these embodiments and the modified examples of each part are presented as examples and are not intended to limit the scope of the invention. These novel embodiments described above can be implemented in various other forms, and various omissions, substitutions, modifications, and combinations can be made without departing from the gist of the invention. These embodiments and their modifications are included in the scope and gist of the invention, and are included in the invention described in the claims. [Explanation of symbols]

[0077] 10. Communications Department 12 Connection 13 Information input section 14 Learning Department 15 Execution Control Unit 16 Threshold setting section 17 Judgment section 18 Information Notification Department 19 Memory section 110 Battery 111 Temperature detection unit 131 Acceleration Sensor 132 Geomagnetic Sensor 133 Gyro sensor 134 Pulse Sensor 135 Temperature Sensor 181 Audio output section 182 Vibration part 183 Display section B Attachment part C Server D. Wearable Devices E-charger G Gateway N Network P Access point S Deep Learning System U Main unit

Claims

1. An information input unit for inputting data of a wearer; a temperature detection unit for detecting the temperature of the battery; A learning unit that performs deep learning based on the data; An execution control unit that switches between interruption and resumption of the execution of the deep learning by the learning unit according to the temperature detected by the temperature detection unit; A wearable device comprising:

2. The wearable device according to claim 1 , wherein the learning unit performs the deep learning while the battery is being charged.

3. The wearable device of claim 1, characterized in that the execution control unit suspends the deep learning by the learning unit when a first threshold value, which is an upper limit, is reached after the learning unit starts executing the deep learning, and resumes the deep learning by the learning unit when a second threshold value, which is a lower limit, is reached.

4. The wearable device according to claim 3, characterized in that it has a threshold setting unit that sets the first threshold and the second threshold based on an evaluation value obtained from the time required for the deep learning and the time when the temperature of the battery becomes higher than a reference temperature.

5. A storage unit that stores the parameters generated by the deep learning together with a loss value when the learning unit interrupts the deep learning, The wearable device according to claim 1 , wherein the learning unit performs the deep learning based on the parameters having the smaller loss value when the deep learning is resumed.

6. The wearable device according to any one of claims 1 to 5, further comprising a judgment unit that judges whether the wearer's condition is normal or abnormal based on the results of the deep learning by the learning unit.

7. A program to be executed by a computer included in a wearable device, A learning process for performing deep learning based on data of a wearer of the wearable device; an execution control process for switching between interruption and resumption of the execution of the learning process depending on the temperature of the battery; A program for a wearable device, comprising:

8. 8. The program for a wearable device according to claim 7, wherein the learning process is performed while the battery is being charged.

9. The program for a wearable device as described in claim 7, characterized in that the execution control process interrupts the deep learning when a first threshold value, which is an upper limit, is reached after starting the learning process, and resumes the deep learning when a second threshold value, which is a lower limit, is reached.

10. The program for a wearable device according to claim 9, characterized in that the computer is caused to execute a threshold setting process that sets the first threshold and the second threshold based on an evaluation value obtained from the time required for the deep learning and the time when the temperature of the battery becomes higher than a reference temperature.

11. The computer includes: When the deep learning is interrupted, a storage process is executed to store the parameters generated by the deep learning together with a loss value; The program for a wearable device according to claim 7, characterized in that, when the deep learning is resumed, the deep learning based on the parameters with the smaller loss value is continued.

12. A program for a wearable device as described in any one of claims 7 to 11, characterized in that the computer is caused to execute a judgment process to judge whether the wearer's condition is normal or abnormal based on the results of the deep learning.

13. Wearable devices are A learning process for performing deep learning based on data of a wearer of the wearable device; an execution control process for switching between interruption and resumption of the execution of the learning process depending on the temperature of the battery; A deep learning method for a wearable device, comprising:

Images