Information processing system, management server, update method, and program

The system addresses inefficient AI model updates in edge devices by employing dual wireless communication methods and a management server for streamlined model generation and distribution, resulting in efficient and accurate updates.

JP7947881B2Active Publication Date: 2026-11-11TOPPAN HOLDINGS INC
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
JP2022160866
Authority / Receiving Office
JP · JP
Patent Type
Patents
Current Assignee / Owner
Filing Date
2022-10-05
Publication Date
2026-11-11
Estimated Expiration
2042-10-05

AI Technical Summary

Technical Problem

Existing information processing systems face challenges in efficiently updating AI models on edge devices due to complex communication requirements with multiple apparatuses, necessitating a more streamlined and efficient update process.

Method used

An information processing system comprising an edge device with dual wireless communication capabilities and a management server that facilitates efficient model updates by using a first wireless method for normal operations and a second, faster, more power-consuming method for updates, along with a management server that generates and supplies machine learning models via wireless communication.

Benefits of technology

The system enables efficient and timely updates of machine learning models, reducing network communications and load, improving estimation accuracy, and enhancing management capabilities.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 0007947881000001
    Figure 0007947881000001
  • Figure 0007947881000002
    Figure 0007947881000002
  • Figure 0007947881000003
    Figure 0007947881000003
Patent Text Reader

Abstract

To more efficiently update a machine learning model.SOLUTION: An information processing system includes: an edge device including a device control part for executing estimation processing based on detection data detected by a sensor part and a previously learned machine learning model; and a management server capable of communicating with the edge device by using radio communication. The management server includes: an acquisition processing part for acquiring learning data including the detection data from the edge device and storing the learning data in a learning data storage part; a learning processing part for making a model generation part execute learning processing based on the learning data stored by the learning data storage part to generate a machine learning model, and storing the machine learning model in a learning model storage part; and a model supply part for transmitting the machine learning model stored in the learning model storage part to the edge device by using the radio communication.SELECTED DRAWING: Figure 1
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to an information processing system, a management server, an update method, and a program.

Background Art

[0002] In recent years, information processing systems using edge devices equipped with sensors have become widespread. Among such information processing systems, one that performs estimation processing on detection data detected by a sensor using an AI model (machine learning model) is known (see, for example, Patent Document 1).

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of Invention

Problems to be Solved by the Invention

[0004] Incidentally, in the information processing system as described above, it is desirable to update the AI model (machine learning model) used by the edge device in accordance with changes in the environment, so as to respond to the environmental changes. Note that the AI models used by edge devices are roughly divided into the following two types. (1) Types in which the edge device retains the AI model internally (2) Types in which the AI model is stored on the cloud, and the edge device downloads and uses it

[0005] However, in the information processing systems of the conventional art, for example, the edge device needs to communicate with a plurality of apparatuses, and the update processing of the AI model (machine learning model) is complicated. Therefore, in the information processing system as described above, it is required to efficiently perform the update processing of the AI model (machine learning model).

[0006] The present invention was made to solve the above problems, and its objective is to provide an information processing system, a management server, an update method, and a program that can update machine learning models more efficiently. [Means for solving the problem]

[0007] To solve the above problems, one aspect of the present invention includes an edge device equipped with a device control unit that performs estimation processing based on detection data detected by a sensor unit and a pre-trained machine learning model, and a management server that can communicate with the edge device using wireless communication, wherein the management server includes an acquisition processing unit that acquires learning data including the detection data from the edge device and stores the learning data in a learning data storage unit, a learning processing unit that causes a model generation unit to perform learning processing based on the learning data stored in the learning data storage unit to generate a machine learning model and stores the machine learning model in a learning model storage unit, and a model supply unit that transmits the machine learning model stored in the learning model storage unit to the edge device using wireless communication. The edge device comprises a first wireless communication unit that performs communication using a first wireless communication method, and a second wireless communication unit that performs communication using a second wireless communication method which consumes more power and has a faster communication speed than the first wireless communication method performed by the first wireless communication unit, and the device control unit, in normal operation mode, transmits the results of the estimation process using the first wireless communication method performed by the first wireless communication unit, and in update mode, which updates the machine learning model, receives the new machine learning model using the second wireless communication method performed by the second wireless communication unit. It is an information processing system.

[0008] Furthermore, in one aspect of the present invention, in the above-described information processing system, the acquisition processing unit may send a request to transmit the learning data to the edge device, and the edge device may, in response to the request to transmit the learning data, send the learning data to the management server.

[0009] Furthermore, in one aspect of the present invention, in the above-described information processing system, the acquisition processing unit may determine whether or not the generation of the learning data is permitted in response to the request for permission to generate the learning data transmitted from the edge device, and if the generation of the learning data is permitted, it may send a request to transmit the learning data to the edge device.

[0010] Furthermore, in one aspect of the present invention, the above-described information processing system includes an external connection device equipped with the model generation unit, the external connection device being able to communicate with the management server, the learning processing unit transmits a generation request to the external connection device to generate the machine learning model, the model generation unit of the external connection device generates the machine learning model in response to the generation request, and stores the generated machine learning model in the learning model storage unit.

[0011] Furthermore, in one aspect of the present invention, the above-described information processing system includes a storage device comprising a learning data storage unit and a learning model storage unit, wherein the storage device is capable of communicating with the management server and the external connection device, and the model generation unit may acquire the learning data from the learning data storage unit of the storage device in response to the generation request.

[0012] Furthermore, in one aspect of the present invention, in the above-described information processing system, the management server comprises the learning data storage unit and the learning model storage unit, the learning processing unit acquires the learning data from the learning data storage unit and transmits the generation request including the learning data to the external connection device, the model generation unit generates the machine learning model in response to the generation request and transmits the generated machine learning model to the management server, and the learning processing unit may store the machine learning model received from the external connection device in the learning model storage unit.

[0013] Furthermore, in one aspect of the present invention, in the above-described information processing system, when the edge device receives a new machine learning model from the management server, it may perform the estimation process using both the existing machine learning model and the new machine learning model, compare the results of both estimation processes, and determine which of the existing machine learning model and the new machine learning model to adopt based on the comparison result.

[0014] Furthermore, in one aspect of the present invention, in the above-described information processing system, the edge device may transmit information based on the comparison results to the management server.

[0016] Furthermore, in one aspect of the present invention, in the above-mentioned information processing system, The first wireless communication method is an LPWA (Low Power Wide Area) wireless communication method. The device control unit stops the operation of the second wireless communication unit in the normal operating mode. Using the aforementioned LPWA method, the result of the estimation process is transmitted. In the update mode, the second wireless communication unit may be operated to perform communication using the second wireless communication method.

[0017] Furthermore, in one aspect of the present invention, in the above-described information processing system, the device control unit may, in the normal operating mode, change to the update mode when it receives a notification requesting an update of the machine learning model using the first wireless communication method.

[0018] Furthermore, in one aspect of the present invention, in the above-described information processing system, the device control unit may transmit information indicating that the machine learning model has been updated using the first wireless communication method.

[0019] Furthermore, one aspect of the present invention includes an edge device equipped with a device control unit that performs estimation processing based on detection data detected by a sensor unit and a pre-trained machine learning model; a management server that can communicate using wireless communication, comprising an acquisition processing unit that acquires learning data including the detection data from the edge device and stores the learning data in a learning data storage unit; a learning processing unit that causes a model generation unit to perform learning processing based on the learning data stored in the learning data storage unit to generate a machine learning model and stores the machine learning model in a learning model storage unit; and a model supply unit that transmits the machine learning model stored in the learning model storage unit to the edge device using wireless communication. The edge device comprises a first wireless communication unit that performs communication using a first wireless communication method, and a second wireless communication unit that performs communication using a second wireless communication method which consumes more power and has a faster communication speed than the first wireless communication method performed by the first wireless communication unit. The device control unit receives the results of the estimation process transmitted from the edge device using the first wireless communication method in normal operation mode, and the model supply unit transmits the new machine learning model to the edge device using the second wireless communication method performed by the second wireless communication unit in update mode for updating the machine learning model. This is the management server.

[0020] Furthermore, one aspect of the present invention is a method for updating the machine learning model by a management server communicable via wireless communication with an edge device that includes a device control unit that executes estimation processing based on detection data detected by a sensor unit and a pre-trained machine learning model, the method comprising: an acquisition processing step in which the management server acquires learning data including the detection data from the edge device and stores the learning data in a learning data storage unit; a learning processing step in which the management server causes a model generation unit to execute learning processing based on the learning data stored in the learning data storage unit to generate a machine learning model, and stores the machine learning model in a learning model storage unit; and a model supply step in which the management server transmits the machine learning model stored in the learning model storage unit to the edge device via the wireless communication Furthermore, the edge device comprises a first wireless communication unit that performs communication using a first wireless communication method, and a second wireless communication unit that performs communication using a second wireless communication method which consumes more power and has a faster communication speed than the first wireless communication method performed by the first wireless communication unit, wherein the edge device receives the results of the estimation process transmitted using the first wireless communication method in the normal operation mode by the device control unit, and in the model supply step, in the update mode in which the management server updates the machine learning model, the second wireless communication unit transmits the new machine learning model to the edge device using the second wireless communication method. This is an updating method.

[0021] Furthermore, one aspect of the present invention is a program for causing a computer of a management server communicable via wireless communication with an edge device that includes a device control unit that executes estimation processing based on detection data detected by a sensor unit and a pre-trained machine learning model, to execute: an acquisition processing step of acquiring learning data including the detection data from the edge device and storing the learning data in a learning data storage unit; a learning processing step of causing a model generation unit to execute learning processing based on the learning data stored in the learning data storage unit to generate a machine learning model, and storing the machine learning model in a learning model storage unit; and a model supply step of transmitting the machine learning model stored in the learning model storage unit to the edge device via the wireless communication The edge device comprises a first wireless communication unit that performs communication using a first wireless communication method, and a second wireless communication unit that performs communication using a second wireless communication method which consumes more power and has a faster communication speed than the first wireless communication method performed by the first wireless communication unit, and the edge device is programmed by the device control unit to receive the results of the estimation process transmitted using the first wireless communication method in the normal operation mode, and in the update mode in which the machine learning model is updated in the model supply step, to transmit the new machine learning model to the edge device using the second wireless communication method performed by the second wireless communication unit. . [Advantageous Effects of Invention]

[0022] According to the present invention, the machine learning model can be updated more efficiently. [Brief Description of Drawings]

[0023] [Figure 1]It is a block diagram illustrating an example of the information processing system according to the first embodiment. [Figure 2] It is a diagram illustrating an example of the operation of the information processing system according to the first embodiment. [Figure 3] It is a diagram illustrating an example of the operation of the first modified example of the information processing system according to the first embodiment. [Figure 4] It is a diagram illustrating an example of the operation of the second modified example of the information processing system according to the first embodiment. [Figure 5] It is a block diagram illustrating an example of the information processing system according to the second embodiment. [Figure 6] It is a diagram illustrating an example of the operation of the information processing system according to the second embodiment. [Figure 7] It is a block diagram illustrating an example of the information processing system according to the third embodiment. [Figure 8] It is a diagram illustrating an example of the operation of the information processing system according to the third embodiment. [Figure 9] It is a block diagram illustrating an example of the information processing system according to the fourth embodiment. [Figure 10] It is the first diagram illustrating an example of the operation of the information processing system according to the fourth embodiment. [Figure 11] It is the second diagram illustrating an example of the operation of the information processing system according to the fourth embodiment. [Figure 12] It is a diagram illustrating an example of the operation of the first modified example of the information processing system according to the fourth embodiment. [Figure 13] It is a diagram illustrating an example of the operation of the second modified example of the information processing system according to the fourth embodiment. [Figure 14] It is a block diagram illustrating an example of the information processing system according to the fifth embodiment. [Figure 15] It is a diagram illustrating an example of the operation of the information processing system according to the fifth embodiment. [Description of Embodiments]

[0024] Hereinafter, an information processing system, a management server, and an update method according to embodiments of the present invention will be described with reference to the drawings.

[0025] [First Embodiment] Figure 1 is a block diagram showing an example of an information processing system 1 according to this embodiment. As shown in Figure 1, the information processing system 1 comprises a management server 10, an edge device 20, and a storage device 30.

[0026] The information processing system 1 is, for example, a system that performs estimation processing based on detection data detected by the edge device 20 and collects the estimation results. In Figure 1, for the sake of explanation, an example with one edge device 20 is shown, but it may also have multiple edge devices 20. Furthermore, in the information processing system 1, the management server 10, the edge device 20, and the storage device 30 can all be connected to the network NW1.

[0027] The edge device 20 performs estimation processing based on detection data detected by the sensor unit 22 and a pre-trained machine learning model. The edge device 20 comprises a wireless communication unit 21, a sensor unit 22, a device storage unit 23, and a device control unit 24.

[0028] The wireless communication unit 21 is a communication unit capable of wireless communication that can connect to the network NW1, and is a communication device using wireless communication methods such as LPWA (Low Power Wide Area), wireless LAN (Local Area Network), mobile communication (3G, 4G, LTE, 5G, etc.), Bluetooth (registered trademark), etc. The wireless communication unit 21 can communicate with the management server 10 via the network NW1.

[0029] The sensor unit 22 is a sensor for detecting detection data (sensing data), and includes temperature sensors, humidity sensors, strain sensors, acceleration sensors, image sensors (image capturing devices), sound sensors (sound capture devices), etc. The sensor unit 22 outputs the detected detection data to the device control unit 24.

[0030] The device storage unit 23 stores various types of information used by the edge device 20. The device storage unit 23 includes, for example, a learning model storage unit 231 and a detection data storage unit 232.

[0031] The learning model storage unit 231 stores machine learning models (estimation models, prediction models, AI models, etc.) for performing estimation processing based on detected data. The machine learning models (AI models) stored in the learning model storage unit 231 are regenerated and updated in response to changes in the environment. The machine learning models are supplied from the management server 10 using wireless communication via the wireless communication unit 21.

[0032] The detection data storage unit 232 stores the detection data detected by the sensor unit 22. The detection data stored in the detection data storage unit 232 is used for estimation processing using a machine learning model, and is also used as training data when regenerating the machine learning model.

[0033] The device control unit 24 is, for example, a processor including a CPU (Central Processing Unit), and comprehensively controls the edge device 20. The device control unit 24 performs estimation processing based on detection data detected by the sensor unit 22 and a pre-trained machine learning model. The device control unit 24 comprises a model update processing unit 241, a detection processing unit 242, and an estimation processing unit 243.

[0034] The model update processing unit 241 executes a process to update the machine learning model. The model update processing unit 241 uses wireless communication by the wireless communication unit 21 to obtain the machine learning model from the management server 10 and stores the machine learning model in the learning model storage unit 231.

[0035] Furthermore, in response to a request from the management server 10 to generate learning data, the model update processing unit 241 acquires the detection data stored in the detection data storage unit 232 and transmits the learning data, including the detection data, to the management server 10 using wireless communication via the wireless communication unit 21.

[0036] The model update processing unit 241 may send a request to the management server 10 for permission to generate training data at predetermined time intervals or when predetermined conditions are met. Here, when predetermined conditions are met, for example, when a change in the environment is detected due to a change in the trend of the detection data of the sensor unit 22, or when the error in the result of the estimation process exceeds a threshold.

[0037] Furthermore, when the model update processing unit 241 receives a machine learning model from the management server 10, it stores the received machine learning model in the learning model storage unit 231. The model update processing unit 241 receives a machine learning model from the management server 10, for example, as a response to the above-mentioned request for the generation of training data.

[0038] The model update processing unit 241 may, at predetermined time intervals (periodically), send a model verification request including terminal information to the management server 10 using wireless communication via the wireless communication unit 21, and receive a machine learning model from the management server 10 as a response to the model verification request.

[0039] The detection processing unit 242 periodically acquires detection data from the sensor unit 22 and stores the acquired detection data in the detection data storage unit 232. Furthermore, the detection processing unit 242 instructs the estimation processing unit 243 to perform estimation processing based on the detected data. The detection processing unit 242 transmits the results of the estimation processing obtained from the estimation processing unit 243 to the management server 10 using wireless communication by the wireless communication unit 21.

[0040] The estimation processing unit 243 performs estimation processing from the detection data using the machine learning model stored in the learning model storage unit 231. The estimation processing unit 243 performs estimation processing in response to requests from the detection processing unit 242. The estimation processing here is, for example, a process of predicting (estimating) future detection data a predetermined time later from the detection data.

[0041] The storage device 30 is, for example, a storage device that can be connected to the network NW1 and can communicate with the management server 10 via the network NW1. The storage device 30 comprises a network communication unit 31 and a storage unit 32.

[0042] The NW communication unit 31 is, for example, a network adapter such as a wired LAN or wireless LAN, and can communicate with the management server 10 via the network NW1.

[0043] The memory unit 32 stores various types of data. The memory unit 32 comprises a learning data storage unit 321 and a learning model storage unit 322. The training data storage unit 321 stores training data for generating a machine learning model. The learning model storage unit 322 stores the machine learning model. Here, the machine learning model is a model that has been trained to perform the estimation process described above.

[0044] The management server 10 is a server device that manages the information processing system 1, which includes the edge device 20. The management server 10 can communicate with the edge device 20 and the storage device 30 via the network NW1. The management server 10 comprises a network communication unit 11, a server storage unit 12, and a server control unit 13.

[0045] The NW communication unit 31 is, for example, a network adapter such as a wired LAN or wireless LAN, and can communicate with the edge device 20 and the storage device 30 via the network NW1.

[0046] The server storage unit 12 stores various types of information used by the management server 10. The server storage unit 12 includes, for example, an estimated result storage unit 121. The estimation result storage unit 121 stores the result data of the estimation process described above. The estimation result storage unit 121 stores, for example, the result data of the estimation process, the identification information of the edge device 20 (for example, terminal information such as MAC (Media Access Control)), and the date and time information in association with each other.

[0047] The server control unit 13 is a processor, such as a CPU, and comprehensively controls the management server 10. The server control unit 13 performs tasks such as controlling the generation and updating of machine learning models, and controlling the collection of estimation processing results from the edge device 20. The server control unit 13 comprises an acquisition processing unit 131, a learning processing unit 132, a model generation unit 133, and a model supply unit 134.

[0048] The acquisition processing unit 131 acquires training data, including detection data, from the edge device 20 and stores the training data in the training data storage unit 321. The acquisition processing unit 131 sends a training data transmission request to the edge device 20 via the network communication unit 11 and acquires the training data from the edge device 20 via the network communication unit 11.

[0049] The acquisition processing unit 131 may periodically (for example, every month) send a request to the edge device 20 to send training data, or it may send a request to the edge device 20 to send training data in response to a request from the edge device 20 to allow the sending of training data. In other words, the acquisition processing unit 131 determines whether or not the generation of training data is permitted in response to a request for permission to generate training data sent from the edge device 20, and if the generation of training data is permitted, it sends a request to send training data to the edge device 20.

[0050] The acquisition processing unit 131 may, for example, determine whether or not the generation of training data is permitted based on the permission information for the generation of training data stored in the storage device 30. The acquisition processing unit 131 transmits the acquired learning data to the storage device 30 via the network communication unit 11, and stores it in the learning data storage unit 321 of the storage device 30.

[0051] The learning processing unit 132 causes the model generation unit 133 (described later) to perform learning processing based on the learning data stored in the learning data storage unit 321 to generate a machine learning model, and also stores the machine learning model in the learning model storage unit 322.

[0052] The learning processing unit 132 acquires learning data from the learning data storage unit 321 of the storage device 30 via the network communication unit 11, and causes the model generation unit 133 to perform learning processing based on the learning data to generate a machine learning model. The learning processing unit 132 also stores the machine learning model in the learning model storage unit 322 of the storage device 30 via the network communication unit 11.

[0053] The model generation unit 133 executes a learning process and generates a machine learning model based on the learning data stored in the learning data storage unit 321. For example, in response to the supply of learning data from the learning processing unit 132, the model generation unit 133 executes a learning process using the learning data and generates a machine learning model. The model generation unit 133 supplies the generated machine learning model to the learning processing unit 132.

[0054] The model supply unit 134 transmits the machine learning model stored in the learning model storage unit 322 to the edge device 20 using wireless communication. For example, in response to a model confirmation request from the edge device 20, the model supply unit 134 retrieves the machine learning model from the learning model storage unit 322 of the storage device 30 via the NW communication unit 11 and transmits the machine learning model to the edge device 20.

[0055] Next, the operation of the information processing system 1 according to this embodiment will be described with reference to the drawings. Figure 2 shows an example of the operation of the information processing system 1 according to this embodiment.

[0056] As shown in Figure 2, the management server 10 first sends a request to the edge device 20 to generate training data (step S101). The acquisition processing unit 131 of the management server 10 sends the request to the edge device 20 to generate training data via the network communication unit 11.

[0057] Next, the edge device 20 performs the training data generation process (step S102). The model update processing unit 241 of the edge device 20 receives the training data generation process via wireless communication from the wireless communication unit 21, and in accordance with the training data generation process, acquires the detection data stored in the detection data storage unit 232, and generates training data including the said detection data as the training data generation process.

[0058] Next, the edge device 20 transmits the training data to the management server 10 (step S103). The model update processing unit 241 transmits the generated training data to the management server 10 using wireless communication by the wireless communication unit 21.

[0059] Next, the management server 10 transmits the learning data to the storage device 30 (step S104). The acquisition processing unit 131 transmits the learning data acquired from the edge device 20 to the storage device 30 via the NW communication unit 11.

[0060] Next, the storage device 30 stores the learning data (step S105). The storage device 30 stores the learning data received via the NW communication unit 31 in the learning data storage unit 321.

[0061] Furthermore, the learning processing unit 132 of the management server 10 determines whether the number of learning data is sufficient (step S106). The learning processing unit 132 determines whether the number of learning data acquired from the edge device 20 is sufficient to perform the learning process. If the number of learning data is sufficient (step S106: YES), the learning processing unit 132 proceeds to step S107. If the number of learning data is not sufficient (step S106: NO), the learning processing unit 132 returns to step S103 and waits until sufficient learning data is secured.

[0062] In step S107, the learning processing unit 132 acquires learning data from the storage device 30. The learning processing unit 132 sends a request to the storage device 30 to transmit learning data via the network communication unit 11. In response to the request to transmit learning data, the storage device 30 transmits the learning data stored in the learning data storage unit 321 to the management server 10 via the network communication unit 31, and the learning processing unit 132 acquires the learning data.

[0063] Next, the learning processing unit 132 executes the machine learning data generation process (step S108). The learning processing unit 132 instructs the model generation unit 133 to execute the learning process using the acquired learning data to generate a new machine learning model. The model generation unit 133 executes the learning process and generates a new machine learning model.

[0064] Next, the learning processing unit 132 transmits the machine learning model to the storage device 30 (step S109). The learning processing unit 132 transmits the machine learning model generated by the model generation unit 133 to the storage device 30 via the network communication unit 11.

[0065] Next, the storage device 30 saves the machine learning model (step S110). The storage device 30 stores the machine learning model received via the NW communication unit 31 in the learning model storage unit 322.

[0066] Next, the edge device 20 sends a model verification request to the management server 10 (step S111). The model update processing unit 241 periodically sends a model verification request to the management server 10 using wireless communication by the wireless communication unit 21.

[0067] Next, the model supply unit 134 of the management server 10 obtains a machine learning model from the storage device 30 (step S112). The model supply unit 134, via the network communication unit 11, sends a request to the storage device 30 to transmit the machine learning model, in response to a model confirmation request received from the edge device 20. In response to the request to transmit the machine learning model, the storage device 30 transmits the machine learning model stored in the learning model storage unit 322 to the management server 10 via the network communication unit 31, and the model supply unit 134 obtains the machine learning model.

[0068] Next, the model supply unit 134 transmits the machine learning model acquired from the storage device 30 to the edge device 20 (step S113). The model supply unit 134 transmits the machine learning model to the edge device 20 via the network communication unit 11.

[0069] Next, the model update processing unit 241 of the edge device 20 updates the machine learning model (step S114). The model update processing unit 241 stores the machine learning model received via wireless communication by the wireless communication unit 21 in the learning model storage unit 231.

[0070] Next, the edge device 20 performs detection and estimation processing (step S115). The detection processing unit 242 of the edge device 20 periodically acquires detection data from the sensor unit 22 and instructs the estimation processing unit 243 to perform estimation processing using the detection data and the updated machine learning model.

[0071] In the process shown in Figure 2, steps S101 to S110 correspond to the machine learning model generation process, and steps S111 to S114 correspond to the machine learning model update process. Furthermore, step S115 corresponds to the normal operation of the edge device 20.

[0072] <First modified example of the first embodiment> Next, with reference to Figure 3, a modified example (first modified example) of the information processing system 1 according to this embodiment will be described. Figure 3 shows an example of the operation of the first modified version of the information processing system 1 according to this embodiment. Here, as the first modified version of this embodiment, a modification in which the generation of a machine learning model is requested from the edge device 20 will be described.

[0073] As shown in Figure 3, when requesting the generation of a machine learning model from the edge device 20, the edge device 20 first sends a request for permission to generate training data to the management server 10 (step S121). The model update processing unit 241 of the edge device 20 sends a request for permission to generate training data to the management server 10 at predetermined time intervals or when predetermined conditions are met.

[0074] Next, the management server 10 confirms permission to generate training data (step S122). The acquisition processing unit 131 of the management server 10 determines whether or not permission to generate training data is granted in response to the training data generation permission request sent from the edge device 20. If permission to generate training data is granted, the acquisition processing unit 131 sends a training data transmission request to the edge device 20 (step S123).

[0075] The subsequent processes from step S123 to step S132 are the same as the processes from step S101 to step S110 shown in Figure 2 above, so their explanation will be omitted here.

[0076] As described above, in this first modified example of the embodiment, the edge device 20 sends a request to the management server 10 for permission to generate training data. The acquisition processing unit 131 of the management server 10 determines whether or not the generation of training data is permitted in response to the request for permission to generate training data sent from the edge device 20, and if the generation of training data is permitted, it sends a request to the edge device 20 for transmission of training data.

[0077] <Second variation of the first embodiment> Next, with reference to Figure 4, a modified example (second modified example) of the information processing system 1 according to this embodiment will be described.

[0078] Figure 4 shows an example of the operation of a second modified version of the information processing system 1 according to this embodiment. Here, as a second modified version of this embodiment, we will describe a modification in which the edge device 20 compares an existing machine learning model with a new machine learning model and adopts the machine learning model that has better estimation accuracy.

[0079] In Figure 4, the processes from step S141 to step S143 are the same as the processes from step S111 to step S113 shown in Figure 2 above, so their explanation is omitted here.

[0080] In step S144, the model update processing unit 241 of the edge device 20 adds a machine learning model. The model update processing unit 241 stores the machine learning model received via wireless communication by the wireless communication unit 21 in the learning model storage unit 231.

[0081] Next, the edge device 20 performs detection processing and estimation processing using both the existing machine learning model and the new machine learning model (step S145). The detection processing unit 242 of the edge device 20 acquires detection data from the sensor unit 22 and instructs the estimation processing unit 243 to perform estimation processing using both the existing machine learning model and the new machine learning model.

[0082] Next, the detection processing unit 242 performs an accuracy comparison (step S146). The detection processing unit 242 compares the results of the estimation process using the existing machine learning model with the results of the estimation process using the new machine learning model. If the existing model is more accurate, the process proceeds to step S147, and the existing machine learning model is adopted as is. After the process in step S147, the detection processing unit 242 proceeds to step S149.

[0083] Furthermore, if the detection processing unit 242 finds that the new model has better accuracy as a result of the accuracy comparison, it proceeds to step S148 and adopts the new machine learning model.

[0084] Next, in step S149, the detection processing unit 242 transmits feedback information to the management server 10. The detection processing unit 242 uses wireless communication by the wireless communication unit 21 to transmit feedback information to the management server 10 indicating which of the existing machine learning model or the new machine learning model was adopted as a result of the accuracy comparison.

[0085] Thus, in this second modified embodiment, when the edge device 20 receives a new machine learning model (new machine learning model) from the management server 10, it performs estimation processing using both the existing machine learning model and the new machine learning model, and compares the results of both estimation processes. Based on the comparison results, the edge device 20 determines which of the existing machine learning model and the new machine learning model to adopt.

[0086] As described above, the information processing system 1 according to this embodiment comprises an edge device 20 and a management server 10. The edge device 20 includes a device control unit 24 that performs estimation processing based on detection data detected by the sensor unit 22 and a pre-trained machine learning model. The management server 10 is capable of communicating with the edge device 20 using wireless communication and comprises an acquisition processing unit 131, a learning processing unit 132, and a model supply unit 134. The acquisition processing unit 131 acquires learning data, including detection data, from the edge device 20 and stores the learning data in the learning data storage unit 321. The learning processing unit 132 causes the model generation unit 133 to perform learning processing based on the learning data stored in the learning data storage unit 321 to generate a machine learning model and stores the machine learning model in the learning model storage unit 322. The model supply unit 134 transmits the machine learning model stored in the learning model storage unit 322 to the edge device 20 using wireless communication.

[0087] As a result, the information processing system 1 according to this embodiment allows, for example, an edge device 20 to acquire a new machine learning model simply by communicating (accessing) with the management server 10. Therefore, the information processing system 1 according to this embodiment simplifies the processing of the edge device 20 when updating the machine learning model, reduces the number of network communications, speeds up processing due to the reduced number of communications, and reduces the network load. Thus, the information processing system 1 according to this embodiment can update machine learning models more efficiently.

[0088] In this embodiment, the acquisition processing unit 131 sends a request to the edge device 20 to send training data. The edge device 20 sends the training data to the management server 10 in response to the training data transmission request.

[0089] As a result, the information processing system 1 according to this embodiment can acquire training data from the edge device 20, starting from the management server 10, and update the machine learning model. Therefore, the information processing system 1 according to this embodiment can update the machine learning model even more efficiently.

[0090] In this embodiment, the acquisition processing unit 131 determines whether or not the generation of training data is permitted in response to the training data generation permission request sent from the edge device 20, and if the generation of training data is permitted, it sends a training data transmission request to the edge device 20.

[0091] As a result, the information processing system 1 according to this embodiment can update the machine learning model starting from the edge device 20. Because the information processing system 1 according to this embodiment can update the machine learning model at an appropriate timing determined by the edge device 20, the machine learning model can be updated even more efficiently.

[0092] Furthermore, in this embodiment, when the edge device 20 receives a new machine learning model from the management server 10, it performs estimation processing using both the existing machine learning model and the new machine learning model, and compares the results of both estimation processes. Based on the comparison results, the edge device 20 determines which of the existing machine learning model and the new machine learning model to adopt.

[0093] As a result, the information processing system 1 according to this embodiment can improve the estimation accuracy by updating the machine learning model, for example by adopting the result of both estimation processes, whichever has better accuracy. Therefore, the information processing system 1 according to this embodiment can update the machine learning model even more efficiently.

[0094] In this embodiment, the edge device 20 transmits information based on the comparison results to the management server 10. As a result, in the information processing system 1 according to this embodiment, the management server 10 can grasp information such as the adopted machine learning model. Therefore, in the information processing system 1 according to this embodiment, the management server 10 can appropriately manage updates to the machine learning model.

[0095] Furthermore, the management server 10 according to this embodiment includes an edge device 20 equipped with a device control unit 24 that performs estimation processing based on detection data detected by the sensor unit 22 and a pre-trained machine learning model, and a management server that can communicate using wireless communication, comprising an acquisition processing unit 131, a learning processing unit 132, and a model supply unit 134. The acquisition processing unit 131 acquires learning data including detection data from the edge device 20 and stores the learning data in the learning data storage unit 321. The learning processing unit 132 causes the model generation unit 133 to perform learning processing based on the learning data stored in the learning data storage unit 321 to generate a machine learning model, and stores the machine learning model in the learning model storage unit 322. The model supply unit 134 transmits the machine learning model stored in the learning model storage unit 322 to the edge device 20 using wireless communication. As a result, the management server 10 according to this embodiment achieves the same effects as the information processing system 1 described above, and can update machine learning models more efficiently.

[0096] Furthermore, the method for updating a machine learning model according to this embodiment is a method for updating a machine learning model by a management server 10, and includes an acquisition processing step, a learning processing step, and a model supply step. Here, the management server 10 can communicate with an edge device 20 equipped with a device control unit 24 that performs estimation processing based on detection data detected by a sensor unit 22 and a pre-trained machine learning model, using wireless communication. In the acquisition processing step, the management server 10 acquires learning data including detection data from the edge device 20 and stores the learning data in the learning data storage unit 321. In the learning processing step, the management server 10 causes the model generation unit 133 to perform learning processing based on the learning data stored in the learning data storage unit 321 to generate a machine learning model, and also stores the machine learning model in the learning model storage unit 322. In the model supply step, the management server 10 transmits the machine learning model stored in the learning model storage unit 322 to the edge device 20 using wireless communication. As a result, the update method according to this embodiment has the same effects as the information processing system 1 described above, and the machine learning model can be updated more efficiently.

[0097] [Second Embodiment] Next, with reference to the drawings, an information processing system 1a according to a second embodiment will be described. In this embodiment, a modified example will be described in which the model generation unit 42 of the external connection device 40 generates the machine learning model instead of the model generation unit 133 of the management server 10.

[0098] Figure 5 is a block diagram showing an example of an information processing system 1a according to the second embodiment. As shown in Figure 5, the information processing system 1a comprises a management server 10a, an edge device 20, a storage device 30, and an external connection device 40. In Figure 5, components identical to those in the first embodiment shown in Figure 1 are given the same reference numerals, and their descriptions are omitted.

[0099] The external connection device 40 is, for example, a user's personal computer or an external server device, and can be connected to the management server 10a and the storage device 30 via the network NW1. Furthermore, the external connection device 40 includes an NW communication unit 41 and a model generation unit 42.

[0100] The NW communication unit 41 is, for example, a network adapter such as a wired LAN or wireless LAN, and can communicate with the management server 10a and the storage device 30 via the network NW1.

[0101] The model generation unit 42 has the same functions as the model generation unit 133 in the first embodiment, and executes a learning process and generates a machine learning model based on the learning data stored in the learning data storage unit 321. The model generation unit 42 generates a machine learning model in response to a request for machine learning model generation from the management server 10a, and stores the generated machine learning model in the learning model storage unit 322.

[0102] Specifically, the model generation unit 42, in response to a request from the management server 10a to generate a machine learning model, obtains training data from the training data storage unit 321 of the storage device 30 via the network communication unit 41. The model generation unit 42 then executes a training process based on the training data to generate a machine learning model. The model generation unit 42 also stores the machine learning model in the training model storage unit 322 of the storage device 30 via the network communication unit 41.

[0103] The management server 10a is a server device that manages the information processing system 1a, which includes the edge device 20. The management server 10a can communicate with the edge device 20, the storage device 30, and the external connection device 40 via the network NW1. The management server 10a comprises a network communication unit 11, a server storage unit 12, and a server control unit 13a.

[0104] The server control unit 13a is a processor, such as a CPU, and comprehensively controls the management server 10a. The server control unit 13a comprises an acquisition processing unit 131, a learning processing unit 132a, and a model supply unit 134.

[0105] The learning processing unit 132a causes the model generation unit 42 of the external connection device 40 to perform learning processing based on the learning data stored in the learning data storage unit 321 to generate a machine learning model. The learning processing unit 132a sends a machine learning model generation request to the external connection device 40 via the NW communication unit 11 as a generation request to generate a machine learning model, causing the model generation unit 42 of the external connection device 40 to generate a machine learning model.

[0106] Next, the operation of the information processing system 1a according to this embodiment will be described with reference to the drawings. Figure 6 shows an example of the operation of the information processing system according to this embodiment.

[0107] In Figure 6, the processes from step S201 to step S206 are the same as the processes from step S101 to step S106 shown in Figure 2 above, so their explanation is omitted here.

[0108] In step S207, the management server 10a sends a request to generate a machine learning model to the external connection device 40. That is, the learning processing unit 132a of the management server 10a sends a request to generate a machine learning model, which is a request to generate a machine learning model, to the external connection device 40 via the NW communication unit 11.

[0109] Next, the external connection device 40 acquires training data from the storage device 30 (step S208). The model generation unit 42 of the external connection device 40 sends a request to the storage device 30 to transmit training data via the network communication unit 41. In response to the training data transmission request, the storage device 30 transmits the training data stored in the training data storage unit 321 to the external connection device 40 via the network communication unit 31, and the model generation unit 42 acquires the training data.

[0110] Next, the model generation unit 42 executes the machine learning model generation process (step S209). The model generation unit 42 performs a learning process using the acquired training data and generates a new machine learning model.

[0111] Next, the model generation unit 42 transmits the machine learning model to the storage device 30 (step S210). The model generation unit 42 transmits the machine learning model generated by the model generation unit 133 to the storage device 30 via the network communication unit 41.

[0112] Next, the storage device 30 saves the machine learning model (step S211). The storage device 30 stores the machine learning model received via the NW communication unit 31 in the learning model storage unit 322.

[0113] As described above, the information processing system 1a according to this embodiment comprises a management server 10a, an edge device 20, a storage device 30, and an external connection device 40. The external connection device 40 includes a model generation unit 42 and is capable of communicating with the management server 10a. The management server 10a includes a learning processing unit 132a, which sends a generation request to the external connection device 40 to generate a machine learning model. The model generation unit 42 generates a machine learning model in response to the generation request and stores the generated machine learning model in the learning model storage unit 322.

[0114] As a result, the information processing system 1a according to this embodiment achieves the same effects as the first embodiment described above, and can update machine learning models more efficiently. Furthermore, since the information processing system 1a according to this embodiment includes a model generation unit 42 that generates machine learning models in the external connection device 40, the processing load on the management server 10a can be reduced.

[0115] Furthermore, the information processing system 1a according to this embodiment includes a storage device 30. The storage device 30 includes a learning data storage unit 321 and a learning model storage unit 322, and is capable of communicating with the management server 10a and the external connection device 40. The model generation unit 42 acquires learning data from the learning data storage unit 321 of the storage device 30 in response to a request to generate a machine learning model.

[0116] As a result, in the information processing system 1a according to this embodiment, the storage device 30 (learning data storage unit 321) is acquired from the external connection device 40, thereby reducing the processing load on the management server 10a. Therefore, the information processing system 1a according to this embodiment can update machine learning models more efficiently.

[0117] [Third Embodiment] Next, with reference to the drawings, an information processing system 1b according to a third embodiment will be described. In this embodiment, a modified version in which the management server 10b includes the functions of the storage device 30 will be described.

[0118] Figure 7 is a block diagram showing an example of an information processing system 1b according to the third embodiment. As shown in Figure 7, the information processing system 1b comprises a management server 10b, an edge device 20, and an external connection device 40. In Figure 7, components identical to those in the second embodiment shown in Figure 5 are given the same reference numerals, and their descriptions are omitted.

[0119] The management server 10b is a server device that manages the information processing system 1b, which includes the edge device 20. The management server 10b can communicate with the edge device 20 and the external connection device 40 via the network NW1. The management server 10b comprises a network communication unit 11, a server storage unit 12a, and a server control unit 13b.

[0120] The server storage unit 12a stores various information used by the management server 10b. The server storage unit 12a includes, for example, an estimation result storage unit 121, a learning data storage unit 122, and a learning model storage unit 123. In this embodiment, the server storage unit 12a includes a learning data storage unit 122 and a learning model storage unit 123, which is different from the first and second embodiments.

[0121] Since the learning data storage unit 122 and the learning model storage unit 123 have the same functions as the learning data storage unit 321 and the learning model storage unit 322 of the first embodiment described above, their description will be omitted here.

[0122] The server control unit 13b is a processor, such as a CPU, and comprehensively controls the management server 10b. The server control unit 13b comprises an acquisition processing unit 131a, a learning processing unit 132b, and a model supply unit 134a.

[0123] The acquisition processing unit 131a is the same as the acquisition processing unit 131 of the first embodiment described above, except that it stores the training data acquired from the edge device 20 in the training data storage unit 122.

[0124] The learning processing unit 132b causes the model generation unit 42 of the external connection device 40 to perform learning processing based on the learning data stored in the learning data storage unit 122 to generate a machine learning model. The learning processing unit 132b retrieves the learning data from the learning data storage unit 321 and sends a generation request including the learning data to the external connection device 40. The learning processing unit 132b also stores the machine learning model received from the external connection device 40 in the learning model storage unit 123 via the NW communication unit 11.

[0125] The model supply unit 134a is the same as the model supply unit 134 of the first embodiment described above, except that it acquires the machine learning model stored in the learning model storage unit 123 and transmits it to the edge device 20.

[0126] Next, the operation of the information processing system 1b according to this embodiment will be described with reference to the drawings. Figure 8 shows an example of the operation of the information processing system 1b according to this embodiment.

[0127] In Figure 8, the processes from step S301 to step S303 are the same as the processes from step S101 to step S103 shown in Figure 2 above, so their explanation is omitted here.

[0128] In step S304, the management server 10b saves the training data. The acquisition processing unit 131a of the management server 10b stores the training data acquired from the edge device 20 in the training data storage unit 122.

[0129] The process in step S305 is the same as the process in step S105 shown in Figure 2 above, so its explanation is omitted here.

[0130] Next, the management server 10b sends a request to generate a machine learning model and training data to the external connection device 40 (step S306). The learning processing unit 132b of the management server 10b acquires the training data from the training data storage unit 321 and sends a request to generate a machine learning model including the training data to the external connection device 40 via the NW communication unit 11.

[0131] Next, the external connection device 40 performs the machine learning model generation process (step S307). The model generation unit 42 of the external connection device 40 performs the learning process based on the learning data received from the management server 10b in response to the machine learning model generation request and generates a machine learning model.

[0132] Next, the model generation unit 42 sends the machine learning model to the management server 10b (step S308). The model generation unit 42 sends the machine learning model to the management server 10b via the network communication unit 41.

[0133] Next, the learning processing unit 132b of the management server 10b stores the machine learning model in the learning model storage unit 123 (step S309). The learning processing unit 132b stores the machine learning model received from the external connection device 40 via the NW communication unit 11 in the learning model storage unit 123.

[0134] As described above, the information processing system 1b according to this embodiment comprises a management server 10b, an edge device 20, and an external connection device 40. The external connection device 40 comprises a model generation unit 42. The management server 10b comprises a learning data storage unit 122, a learning model storage unit 123, and a learning processing unit 132b. The learning processing unit 132b acquires learning data from the learning data storage unit 122 and sends a request to the external connection device 40 to generate a machine learning model including the learning data. The model generation unit 42 of the external connection device 40 generates a machine learning model in response to the request to generate a machine learning model and sends the generated machine learning model to the management server 10b. The learning processing unit 132b stores the machine learning model received from the external connection device 40 in the learning model storage unit 123.

[0135] As a result, the information processing system 1b according to this embodiment achieves the same effects as the first and second embodiments described above, and enables more efficient updating of machine learning models. Furthermore, since the functions of the storage device 30 are built into the management server 10b, the configuration of the information processing system 1b according to this embodiment can be simplified.

[0136] [Fourth Embodiment] Next, with reference to the drawings, an information processing system 1c according to the fourth embodiment will be described. In this embodiment, a modified example will be described in which a machine learning model is supplied to an edge device 20a by switching between LPWA wireless communication (first wireless communication method), which has a low communication speed and low power consumption, and high-speed wireless communication (second wireless communication method), which has a higher communication speed and higher power consumption compared to LPWA.

[0137] In conventional information processing systems, it is common to use low-power wireless communication methods such as LPWA to reduce the power consumption of edge devices. However, in such conventional information processing systems, the slow speed of LPWA wireless communication posed a problem, making it inefficient when updating machine learning models on edge devices.

[0138] Therefore, in the information processing system 1c according to this embodiment, the machine learning model is updated by switching between LPWA wireless communication (first wireless communication method), which has a low communication speed and low power consumption, and high-speed wireless communication (second wireless communication method), which has a higher communication speed and higher power consumption compared to LPWA.

[0139] Figure 9 is a block diagram showing an example of an information processing system 1c according to this embodiment. As shown in Figure 9, the information processing system 1c comprises a management server 10a, an edge device 20a, a storage device 30, an external connection device 40, an LPWA server 50, and an LPWA access point 51. In Figure 9, components identical to those in the second embodiment shown in Figure 5 are given the same reference numerals, and their descriptions are omitted.

[0140] The LPWA server 50 is a server device for performing LPWA wireless communication with the edge device 20a. The LPWA server 50 performs LPWA wireless communication with the edge device 20a via the LPWA access point 51 and performs relay processing between the network NW1 and the LPWA wireless communication. The LPWA server 50 may be included in the management server 10a.

[0141] The LPWA access point 51 is an access point for performing LPWA wireless communication with the edge device 20a.

[0142] The edge device 20a comprises a wireless communication unit 21a, a sensor unit 22, a device storage unit 23, and a device control unit 24a. The wireless communication unit 21a comprises an LPWA communication unit 211 and a high-speed wireless communication unit 212.

[0143] The LPWA communication unit 211 (an example of the first wireless communication unit) performs communication using the LPWA method (first wireless communication method). The LPWA communication unit 211 can connect to the network NW1 via the LPWA access point 51 and LPWA server 50 described above. The LPWA communication unit 211 performs wireless communication with the LPWA server 50 via the LPWA access point 51.

[0144] The high-speed wireless communication unit 212 (an example of the second wireless communication unit) performs communication using a high-speed wireless communication method (second wireless communication method) that consumes more power and has a faster communication speed than the LPWA method used by the LPWA communication unit 211. The high-speed wireless communication method is, for example, wireless LAN, mobile communication (3G, 4G, LTE, 5G, etc.), and Bluetooth (registered trademark).

[0145] The device control unit 24a is, for example, a processor including a CPU, and comprehensively controls the edge device 20a. The device control unit 24a performs estimation processing based on detection data detected by the sensor unit 22 and a pre-trained machine learning model.

[0146] In normal operation mode, the device control unit 24a transmits the results of the estimation process using the LPWA method provided by the LPWA communication unit 211. In update mode, when updating the machine learning model, the device control unit 24a receives the new machine learning model using the high-speed wireless communication method provided by the high-speed wireless communication unit 212.

[0147] In normal operation mode, the device control unit 24a stops the operation of the high-speed wireless communication unit 212, and in update mode, it activates the high-speed wireless communication unit 212 to perform communication using the high-speed wireless communication method.

[0148] Furthermore, in normal operation mode, the device control unit 24a switches to update mode when it receives a notification requesting an update to the machine learning model using the LPWA method. Furthermore, the device control unit 24a includes a model update processing unit 241a, a detection processing unit 242, and an estimation processing unit 243.

[0149] The model update processing unit 241a executes a process to update the machine learning model. The model update processing unit 241a is basically the same as the model update processing unit 241 of the first and second embodiments described above, except that it uses a high-speed wireless communication method by the high-speed wireless communication unit 212 when executing the process to update the machine learning model. The model update processing unit 241a uses wireless communication via the high-speed wireless communication unit 212 to acquire the machine learning model from the management server 10a and stores the machine learning model in the learning model storage unit 231.

[0150] Next, the operation of the information processing system 1c according to this embodiment will be described with reference to the drawings. Figure 10 is the first diagram showing an example of the operation of the information processing system 1c according to this embodiment. In Figure 10, the update process of the machine learning model in this embodiment, when the storage device 30 is storing a new machine learning model, will be explained.

[0151] As shown in Figure 10, the LPWA server 50 first sends a notification to the edge device 20a of the application of a new machine learning model via the LPWA access point 51 (step S401). The LPWA server 50 sends the notification to the edge device 20a of the application of a new machine learning model via LPWA wireless communication.

[0152] Next, the edge device 20a changes to update mode (step S402). The device control unit 24a of the edge device 20a operates the high-speed wireless communication unit 212 to change to update mode.

[0153] Next, the edge device 20a sends a request to the management server 10a to send a new machine learning model (step S403). The model update processing unit 241a of the edge device 20a sends a request to the management server 10a to send a new machine learning model (model confirmation request) using wireless communication by the high-speed wireless communication unit 212.

[0154] Next, the management server 10a obtains a new machine learning model from the storage device 30 (step S404). The model supply unit 134 of the management server 10a, via the network communication unit 11, sends a request to the storage device 30 to transmit a new machine learning model in response to a request to transmit a new machine learning model received from the edge device 20a. In response to the request to transmit a new machine learning model, the storage device 30 transmits the machine learning model stored in the learning model storage unit 322 to the management server 10a via the network communication unit 31, and the model supply unit 134 obtains the machine learning model.

[0155] Next, the model supply unit 134 of the management server 10a transmits the new machine learning model acquired from the storage device 30 to the edge device 20a (step S405). The model supply unit 134 transmits the new machine learning model to the edge device 20a using wireless communication by the high-speed wireless communication unit 212.

[0156] Next, the edge device 20a determines whether or not a new machine learning model has been acquired (step S406). The model update processing unit 241a of the edge device 20a determines whether or not a new machine learning model has been acquired, and if a new machine learning model has been acquired (step S406: YES), it proceeds to step S407. If a new machine learning model has not been acquired (step S406: NO), the model update processing unit 241a returns to step S403 and repeats the process until a new machine learning model is acquired.

[0157] In step S407, the model update processing unit 241a updates the machine learning model. The model update processing unit 241a stores the machine learning model received via wireless communication by the high-speed wireless communication unit 212 in the learning model storage unit 231.

[0158] Next, the model update processing unit 241a changes to normal operation mode (step S408). The model update processing unit 241a stops the operation of the high-speed wireless communication unit 212 and changes to normal operation mode.

[0159] Next, the edge device 20a performs detection and estimation processing (step S409). The detection processing unit 242 of the edge device 20a periodically acquires detection data from the sensor unit 22 and instructs the estimation processing unit 243 to perform estimation processing using the detection data and the updated machine learning model.

[0160] Figure 11 is a second diagram showing an example of the operation of the information processing system 1c according to this embodiment. Figure 11 describes the machine learning model update process in this embodiment when the storage device 30 does not have a new machine learning model stored.

[0161] In Figure 11, the processes from step S411 to step S413 are the same as the processes from step S401 to step S403 shown in Figure 10 above, so their explanation is omitted here.

[0162] Next, in step S414, the management server 10a sends a request to send a new machine learning model in order to retrieve the new machine learning model from the storage device 30, but the storage device 30 does not have the new machine learning model stored. Therefore, the storage device 30 sends a notification to the management server 10a that there is no new machine learning model.

[0163] Next, the management server 10a sends a request to the external connection device 40 to generate a machine learning model (step S415). Note that the processing from step S415 to step S419 is the same as the processing from step S207 to step S211 shown in Figure 6 above, so its explanation is omitted here.

[0164] Furthermore, the processes from step S420 to step S425 are the same as the processes from step S403 to step S408 shown in Figure 10 above, so their explanation is omitted here.

[0165] <First modified example of the fourth embodiment> Next, with reference to Figure 12, a modified example (first modified example) of the information processing system 1c according to this embodiment will be described. Figure 12 shows an example of the operation of the first modified version of the information processing system 1c according to this embodiment. Here, as the first modified version of this embodiment, a modification in which the management server 10a sends an update notification of the machine learning model to the external connection device 40 will be described.

[0166] In Figure 12, the processes from step S431 to step S435 are the same as the processes from step S401 to step S405 shown in Figure 10 above, so their explanation is omitted here.

[0167] In step S436, the management server 10a sends a notification of the machine learning model update to the external connection device 40. That is, the model supply unit 134 of the management server 10a notifies the external connection device 40, which is the source of the machine learning model, that the machine learning model has been applied.

[0168] The subsequent processes from step S437 to step S440 are the same as the processes from step S406 to step S409 shown in Figure 10 above, so their explanation will be omitted here.

[0169] <Second modified example of the fourth embodiment> Next, with reference to Figure 13, a modified example (second modified example) of the information processing system 1c according to this embodiment will be described. Figure 13 shows an example of the operation of a second modified version of the information processing system 1c according to this embodiment. Here, as a second modified version of this embodiment, a modification in which the edge device 20a sends an update notification of the machine learning model to the LPWA server 50 will be described.

[0170] In this embodiment, the device control unit 24a of the edge device 20a transmits information indicating that the machine learning model update is complete to, for example, the LPWA server 50, using the LPWA method.

[0171] In Figure 13, the processes from step S451 to step S458 are the same as the processes from step S401 to step S408 shown in Figure 10 above, so their explanation is omitted here.

[0172] In step S459, the edge device 20a sends a machine learning model update notification to the LPWA server 50 via the LPWA access point 51. The device control unit 24a of the edge device 20a uses wireless communication via the LPWA communication unit 211 to send a machine learning model update notification to the LPWA server 50 indicating that the machine learning model update is complete.

[0173] Next, the edge device 20a performs detection processing and estimation processing (step S460). The detection processing unit 242 of the edge device 20a periodically acquires detection data from the sensor unit 22 and instructs the estimation processing unit 243 to perform estimation processing using the detection data and the updated machine learning model.

[0174] As described above, the information processing system 1c according to this embodiment comprises a management server 10a, an edge device 20a, a storage device 30, an external connection device 40, and an LPWA server 50. The edge device 20a comprises an LPWA communication unit 211 (first wireless communication unit), a high-speed wireless communication unit 212 (second wireless communication unit), and a device control unit 24a. The LPWA communication unit 211 performs communication using the LPWA method (first wireless communication method). The high-speed wireless communication unit 212 performs communication using the high-speed wireless communication method (second wireless communication method), which consumes more power and has a faster communication speed than the LPWA method used by the LPWA communication unit 211. In the normal operation mode, the device control unit 24a transmits the results of estimation processing using the LPWA method used by the LPWA communication unit 211, and in the update mode, which updates the machine learning model, it receives the new machine learning model using the high-speed wireless communication method used by the high-speed wireless communication unit 212.

[0175] As a result, the information processing system 1c according to this embodiment uses a low-power LPWA method (first wireless communication method) for communication in normal operation mode, and a high-speed wireless communication method (second wireless communication method) for communication in update mode when updating machine learning models. Therefore, the information processing system 1c according to this embodiment can maintain low power consumption in normal operation while efficiently communicating at high speed when updating machine learning models. Thus, the information processing system 1c according to this embodiment can update machine learning models more efficiently.

[0176] Furthermore, in this embodiment, the device control unit 24a stops the operation of the high-speed wireless communication unit 212 in the normal operation mode, and operates the high-speed wireless communication unit 212 in the update mode to perform communication using the high-speed wireless communication method.

[0177] As a result, the information processing system 1c according to this embodiment can further reduce power consumption because it stops the operation of the high-speed wireless communication unit 212 in the normal operating mode.

[0178] Furthermore, in this embodiment, the device control unit 24a, in normal operation mode, changes to update mode when it receives a notification requesting an update to the machine learning model using the LPWA method. As a result, the information processing system 1c according to this embodiment can switch to high-speed wireless communication when updating the machine learning model.

[0179] In this embodiment, the device control unit 24a transmits information indicating that the machine learning model update has been completed using the LPWA method. As a result, the information processing system 1c according to this embodiment can notify an external party that the machine learning model update is complete, with low power consumption.

[0180] Although the first and second modifications of this embodiment described above were explained as being implemented individually, the invention is not limited to this, and the first and second modifications of this embodiment may also be implemented in combination.

[0181] [Fifth Embodiment] Next, with reference to the drawings, an information processing system 1d according to a fifth embodiment will be described. In this embodiment, a modification will be described in which the edge device 20b switches to normal operation mode when it is unable to acquire a machine learning model.

[0182] Figure 14 is a block diagram showing an example of an information processing system 1d according to the fifth embodiment. As shown in Figure 14, the information processing system 1d comprises a management server 10a, an edge device 20b, a storage device 30, an external connection device 40, an LPWA server 50, and an LPWA access point 51. In Figure 14, components identical to those in the fourth embodiment shown in Figure 9 are given the same reference numerals, and their descriptions are omitted.

[0183] The edge device 20b comprises a wireless communication unit 21a, a sensor unit 22, a device storage unit 23, and a device control unit 24b. The wireless communication unit 21a comprises an LPWA communication unit 211 and a high-speed wireless communication unit 212.

[0184] The device control unit 24b is, for example, a processor including a CPU, and comprehensively controls the edge device 20b. The basic functions of the device control unit 24b are the same as those of the device control unit 24a in the fourth embodiment described above. The device control unit 24b differs from the device control unit 24a in the fourth embodiment in that it switches to normal operation mode when it is unable to acquire a machine learning model.

[0185] Furthermore, in this embodiment, the external connection device 40 sends a notification of machine learning model update to the LPWA server 50 when a new machine learning model is generated. Furthermore, in this embodiment, the LPWA server 50 sends a notification to the edge device 20b to apply the new machine learning model in response to receiving a notification of an update to the machine learning model.

[0186] Next, with reference to Figure 15, the operation of the information processing system 1d according to this embodiment will be described. Figure 15 is the first diagram showing an example of the operation of the information processing system 1d according to this embodiment. Figure 15 describes the machine learning model update process in this embodiment when the storage device 30 does not have a new machine learning model stored.

[0187] In Figure 15, the processes from step S501 to step S509 are the same as the processes from step S411 to step S419 shown in Figure 11 above, so their explanation is omitted here.

[0188] In step S510, the external connection device 40 sends a notification of machine learning model update to the LPWA server 50. The model generation unit 42 of the external connection device 40 sends the notification of machine learning model update to the LPWA server 50 via the NW communication unit 41 and the network NW1.

[0189] Furthermore, in step S511, the edge device 20b switches to normal operation mode. The model update processing unit 241b of the edge device 20b stops the operation of the high-speed wireless communication unit 212 and switches to normal operation mode if it is unable to acquire a machine learning model.

[0190] Furthermore, the processes from step S512 to step S519 are the same as the processes from step S431 to step S439 shown in Figure 12 above, so their explanation is omitted here.

[0191] As described above, the information processing system 1d according to this embodiment comprises a management server 10a, an edge device 20b, a storage device 30, an external connection device 40, an LPWA server 50, and an LPWA access point 51. The device control unit 24b switches to normal operation mode when it is unable to acquire a machine learning model. In addition, the external connection device 40 in this embodiment sends a machine learning model update notification to the LPWA server 50 when it generates a new machine learning model, and the LPWA server 50, upon receiving the machine learning model update notification, sends a notification to the edge device 20b that it has applied the new machine learning model. As a result, the information processing system 1d according to this embodiment achieves the same effects as the fourth embodiment described above, and can update machine learning models more efficiently.

[0192] It should be noted that the present invention is not limited to the embodiments described above, and can be modified without departing from the spirit of the invention. For example, in each of the embodiments described above, the management server 10 (10a, 10b) is described as being composed of a single device, but it is not limited to this and may be composed of multiple server devices.

[0193] Furthermore, in the fourth and fifth embodiments described above, the LPWA server 50 may be configured as being included in the management server 10a. In addition, the LPWA server 50 may be connected to edge devices 20a (20b) via an LPWA repeater, in addition to the LPWA access point 51.

[0194] Furthermore, although the fourth and fifth embodiments described above illustrate examples of application to the second embodiment, they are not limited to these and may also be applied to other embodiments, such as the first or third embodiment.

[0195] Furthermore, each component of the information processing system 1 (1a to 1d) described above has a computer system inside. The processing in each component of the information processing system 1 (1a to 1d) described above may be performed by recording a program for realizing the functions of each component of the information processing system 1 (1a to 1d) onto a computer-readable recording medium, loading the program recorded on this recording medium into the computer system, and executing it. Here, "loading the program recorded on the recording medium into the computer system and executing it" includes installing the program into the computer system. Here, "computer system" includes hardware such as the OS and peripheral devices.

[0196] Furthermore, "computer system" may include multiple computer devices connected via a network, including communication lines such as the Internet, WAN, LAN, and dedicated lines. "Computer-readable recording medium" refers to portable media such as flexible disks, magneto-optical disks, ROMs, and CD-ROMs, as well as storage devices such as hard disks built into computer systems. Thus, the recording medium storing the program may be a non-transient recording medium such as a CD-ROM.

[0197] Furthermore, the recording medium also includes internal or external recording media accessible from the distribution server for distributing the program. The program may be divided into multiple parts, downloaded at different times, and then combined in each configuration of the information processing system 1 (1a-1d), and each divided program may be distributed by a different distribution server. Additionally, "computer-readable recording medium" includes volatile memory (RAM) within computer systems that act as servers or clients when a program is transmitted over a network, which retains the program for a certain period of time. Moreover, the program may be intended to implement only a portion of the functions described above. Furthermore, the program may be a so-called differential file (differential program) that can implement the functions described above in combination with a program already recorded in the computer system.

[0198] Furthermore, some or all of the above-mentioned functions may be implemented as integrated circuits such as LSIs (Large Scale Integrations). Each of the above-mentioned functions may be implemented as an individual processor, or some or all of them may be integrated into a single processor. In addition, the method of implementing integrated circuits is not limited to LSIs; they may also be implemented using dedicated circuits or general-purpose processors. Furthermore, if advances in semiconductor technology lead to the emergence of integrated circuit technologies that can replace LSIs, integrated circuits using such technologies may be used. [Explanation of symbols]

[0199] 1, 1a, 1b, 1c, 1d... Information processing system 10, 10a, 10b... Management Server 11, 31, 41…NW Communications Department 12,12a... Server storage 13, 13a, 13b... Server Control Unit 21,21a…Wireless communication department 22...Resensing unit 23...Device storage unit 24, 24a, 24b... Device control unit 30…Storage device 32...Storage section 40…External connection device 50…LPWA Server 51…LPWA access point 121…Estimation result storage unit 131... Acquisition Processing Unit 132, 132a, 132b... Learning Processing Unit 133,42...Model generation unit 134...Model Supply Department 211...LPWA Communications Department 212... High-speed wireless communication section 231,322,123…Learning model memory unit 232...Detection data storage unit 241, 241a, 241b... Model update processing unit 242...Detection Processing Unit 243… Estimation Processing Unit 321,122…Learning data storage unit NW1…Network

Claims

1. An edge device comprising a device control unit that performs estimation processing based on detection data detected by a sensor unit and a pre-trained machine learning model, A management server that can communicate with the edge device using wireless communication Equipped with, The aforementioned management server An acquisition processing unit that acquires training data including the detection data from the edge device and stores the training data in a training data storage unit, A learning processing unit causes a model generation unit to perform a learning process based on the learning data stored in the learning data storage unit to generate a machine learning model, and also causes a learning model storage unit to store the machine learning model. A model supply unit transmits the machine learning model stored in the learning model storage unit to the edge device using wireless communication. Equipped with, The edge device is A first wireless communication unit that performs communication using the first wireless communication method, A second wireless communication unit performs communication using a second wireless communication method which consumes more power and has a faster communication speed than the first wireless communication method used by the first wireless communication unit. Equipped with, In the normal operating mode, the device control unit transmits the results of the estimation process using the first wireless communication method by the first wireless communication unit, and in the update mode for updating the machine learning model, it receives the new machine learning model using the second wireless communication method by the second wireless communication unit. Information processing system.

2. The first wireless communication method is an LPWA (Low Power Wide Area) wireless communication method, The device control unit, In the normal operation mode, the operation of the second wireless communication unit is stopped, and the result of the estimation process is transmitted using the LPWA method. In the update mode, the second wireless communication unit is operated to perform communication using the second wireless communication method. The information processing system according to claim 1.

3. The acquisition processing unit sends the request to transmit the learning data to the edge device. The edge device transmits the training data to the management server in response to the request to transmit the training data. The information processing system according to claim 1 or claim 2.

4. The acquisition processing unit determines whether the generation of the training data is permitted in response to the training data generation permission request transmitted from the edge device, and if the generation of the training data is permitted, it sends a training data transmission request to the edge device. The information processing system according to claim 3.

5. An external connection device comprising the model generation unit, further comprising an external connection device capable of communicating with the management server, The learning processing unit sends a generation request to the external connection device to generate the machine learning model. The information processing system according to claim 1 or 2, wherein the model generation unit of the external connection device generates the machine learning model in response to the generation request and stores the generated machine learning model in the learning model storage unit.

6. A storage device comprising the learning data storage unit and the learning model storage unit, wherein the storage device is capable of communicating with the management server and the external connection device, The model generation unit acquires the training data from the training data storage unit of the storage device in response to the generation request. The information processing system according to claim 5.

7. The management server comprises the learning data storage unit and the learning model storage unit, The learning processing unit acquires the learning data from the learning data storage unit and transmits the generation request including the learning data to the external connection device. The model generation unit generates the machine learning model in response to the generation request and transmits the generated machine learning model to the management server. The learning processing unit causes the machine learning model received from the external connection device to be stored in the learning model storage unit. The information processing system according to claim 5.

8. When the edge device receives a new machine learning model from the management server, it performs the estimation process using both the existing machine learning model and the new machine learning model, compares the results of both estimation processes, and determines which of the existing machine learning model or the new machine learning model to adopt based on the comparison results. The information processing system according to claim 1 or claim 2.

9. The edge device transmits information based on the comparison results to the management server. The information processing system according to claim 8.

10. The device control unit, In the normal operating mode, when a notification requesting an update of the machine learning model is received using the first wireless communication method, the system switches to the update mode. The information processing system according to claim 2.

11. The device control unit transmits information indicating that the machine learning model update has been completed using the first wireless communication method. The information processing system according to claim 1 or claim 2.

12. An edge device comprising a device control unit that performs estimation processing based on detection data detected by a sensor unit and a pre-trained machine learning model, and a management server that can communicate using wireless communication, An acquisition processing unit that acquires training data including the detection data from the edge device and stores the training data in a training data storage unit, A learning processing unit causes a model generation unit to perform a learning process based on the learning data stored in the learning data storage unit to generate a machine learning model, and also causes a learning model storage unit to store the machine learning model. A model supply unit transmits the machine learning model stored in the learning model storage unit to the edge device using wireless communication. Equipped with, The edge device is A first wireless communication unit that performs communication using the first wireless communication method, A second wireless communication unit performs communication using a second wireless communication method which consumes more power and has a faster communication speed than the first wireless communication method used by the first wireless communication unit. It is equipped with, The device control unit receives the results of the estimation process transmitted from the edge device using the first wireless communication method in the normal operating mode. In the update mode for updating the machine learning model, the model supply unit transmits the new machine learning model to the edge device using the second wireless communication method by the second wireless communication unit. Management server.

13. An edge device comprising a device control unit that performs estimation processing based on detection data detected by a sensor unit and a pre-trained machine learning model, and a method for updating the machine learning model by a communicationable management server using wireless communication, The management server acquires the learning data, including the detection data, from the edge device and stores the learning data in the learning data storage unit; The management server performs a learning process step in which it causes the model generation unit to execute a learning process based on the learning data stored in the learning data storage unit to generate a machine learning model, and also causes the learning model storage unit to store the machine learning model. The management server performs a model supply step in which it transmits the machine learning model stored in the learning model storage unit to the edge device using wireless communication. Includes, The edge device is A first wireless communication unit that performs communication using the first wireless communication method, A second wireless communication unit performs communication using a second wireless communication method which consumes more power and has a faster communication speed than the first wireless communication method used by the first wireless communication unit. It is equipped with, The edge device receives the results of the estimation process transmitted by the device control unit in the normal operating mode using the first wireless communication method. In the model supply step, the management server, in the update mode in which the machine learning model is updated, transmits the new machine learning model to the edge device using the second wireless communication method by the second wireless communication unit. How to update.

14. An edge device equipped with a device control unit that performs estimation processing based on detection data detected by a sensor unit and a pre-trained machine learning model, communicates wirelessly with a computer on a management server, The acquisition process step includes acquiring training data, including the detection data, from the edge device and storing the training data in the training data storage unit, A learning processing step involves causing the model generation unit to perform a learning process based on the learning data stored in the learning data storage unit to generate a machine learning model, and storing the machine learning model in the learning model storage unit. A model supply step of transmitting the machine learning model stored in the learning model storage unit to the edge device using wireless communication. This is a program to execute, The edge device is A first wireless communication unit that performs communication using the first wireless communication method, A second wireless communication unit performs communication using a second wireless communication method which consumes more power and has a faster communication speed than the first wireless communication method used by the first wireless communication unit. It is equipped with, The edge device receives the results of the estimation process transmitted by the device control unit in the normal operating mode using the first wireless communication method. In the model supply step, in the update mode for updating the machine learning model, the second wireless communication unit performs a process to transmit the new machine learning model to the edge device using the second wireless communication method. program.

Citation Information

Patent Citations

  • Edge server and ai learning model generation and operation method in edge server

    JP2019139734A

  • Learning method, learning program and learning device

    JP2020177344A

  • WIRELESS COMMUNICATION DEVICE, IoT DEVICE, CONTROL METHOD, AND PROGRAM

    JP2020202462A

  • Trained model update system, trained model update method, and program

    JP6759472B2