WO2019171468A1

WO2019171468A1 - Machine learning-trained model switching system, edge device, machine learning-trained model switching method, and program

Info

Publication number: WO2019171468A1
Application number: PCT/JP2018/008580
Authority: WO
Inventors: 将仁谷口
Original assignee: 株式会社ウフル
Priority date: 2018-03-06
Filing date: 2018-03-06
Publication date: 2019-09-12

Abstract

According to the present invention, in the field of Internet of Things (IoT), an edge device is caused to temporally switch between uses of different machine learning-trained models. This machine learning-trained model switching system causes an edge device to switch between machine learning-trained models having different purposes. An acquisition means acquires time information relating to a time. A control means performs control to cause the edge device to acquire a machine learning-trained model associated with the time information.

Description

Machine learning model switching system, edge device, machine learning model switching method, and program

The present invention relates to a technology for switching a machine-learned model used by an edge device, and is used in the field of IoT (Internet of Things).

In recent years, with the advancement of AI (artificial intelligence) technology, renewed technology for learned models has attracted attention. For example, a technique for updating a model of a machine learning system is provided (Patent Document 1).

JP 2017-120647 A

In the system described in Patent Document 1, a machine learning system model is trained. However, the purpose of processing does not change even if the machine learning system model is trained. By the way, there is a case where it is desired to use one edge device for different purposes depending on time. For example, when the edge device performs processing using a machine-learned model corresponding to its application, it is necessary to change the machine-learned model according to a change in the use of the edge device. However, for example, when the edge device does not have sufficient hardware resources, the edge device stores a large number of machine-learned models, and these cannot be switched and used.
An object of the present invention is to switch a machine-learned model used by an edge device according to time.

The present invention is a machine-learned model switching system for switching machine-learned models having different purposes for an edge device, the acquisition means for acquiring time information relating to time, and the edge device corresponding to the time information There is provided a machine learning model switching system comprising: control means for controlling to acquire an attached machine learned model.

Further, the present invention is an edge device that switches and uses different machine-learned models for different purposes, and acquires the machine-learned model according to the control of the machine-learned model switching system according to claim 1, An edge device is provided that replaces other machine-learned models in use with the acquired machine-learned model.

Further, the present invention provides a machine-learned model switching method for switching a machine-learned model having a different purpose for an edge device, the step of acquiring time information related to time, the edge device including the time information A machine learning model switching method comprising: controlling to obtain an associated machine learned model.

Furthermore, the present invention is a program for causing a computer to execute a machine-learned model switching process for switching machine-learned models having different purposes with respect to an edge device, and acquiring time information relating to time; And controlling the edge device to acquire a machine-learned model associated with the time information.

According to the present invention, the machine-learned model used by the edge device can be switched over time.

It is a figure which shows an example of the machine learning completed model switching system 100 which concerns on embodiment. 2 is a diagram illustrating an example of a hardware configuration of a server 110. FIG. It is a figure which shows an example of the time table. 2 is a diagram illustrating an example of a functional configuration of a server 110. FIG. 2 is a diagram illustrating an example of a hardware configuration of an edge device 120. FIG. 2 is a diagram illustrating an example of a functional configuration of an edge device 120. FIG. It is a sequence chart which shows an example of the operation | movement which switches the machine learning completed model of the edge device. 5 is a diagram showing an example of posted information 320. FIG. It is a sequence chart which shows an example of the operation | movement which optimizes the schedule which switches a machine learning completed model.

1. Configuration FIG. 1 is a diagram illustrating an example of a machine-learned model switching system 100 according to the present embodiment. The machine-learned model switching system 100 is a system that switches a machine-learned model used by the edge device 120 according to time.

The machine learning model switching system 100 includes a server 110 and an edge device 120. In addition, the number of each apparatus shown in FIG. 1 is an illustration, and is not limited to this. The server 110 and the edge device 120 are connected via a communication network 130. The communication network 130 includes, for example, the Internet.

FIG. 2 is a diagram illustrating an example of a hardware configuration of the server 110. The server 110 is an apparatus that provides a plurality of machine learned models with different purposes to the edge device 120. The server 110 is a computer including a processor 111, a memory 112, a storage 113, and a communication unit 114, for example. These devices are connected via a bus 115.

The processor 111 executes various processes by reading the program into the memory 112 and executing it. The processor 111 may be configured by, for example, a CPU (Central Processing 例えば Unit). The memory 112 stores a program executed by the processor 111. The memory 112 may be configured by, for example, a ROM (Read Only Memory) or a RAM (Random Access Memory). The storage 113 stores various data and programs. The storage 113 may be configured by, for example, a hard disk drive or a flash memory. The storage 113 stores a plurality of machine learned models provided to the edge device 120 and a time table 310 indicating a schedule for switching the machine learned models used by the edge device 120. The communication unit 114 is a communication interface connected to the communication network 130. The communication unit 114 performs data communication via the communication network 130.

FIG. 3 is a diagram illustrating an example of the time table 310. In this example, the edge device 120 performs a watching process and a security process. The watching process is a process of observing whether the user is in a safe state. For example, in the watching process, a user abnormality is detected, and detection information is notified to a predetermined notification destination. In the watching process, a dedicated machine-learned model (hereinafter referred to as “watched machine-learned model”) for accurately performing the watching process is used. On the other hand, the crime prevention process is a process for monitoring whether or not a suspicious person has entered. For example, in the crime prevention process, intrusion of a suspicious person is detected, and detection information is notified to a predetermined notification destination. In the security process, a dedicated machine-learned model (hereinafter referred to as a “machine-learned model for crime prevention”) for accurately performing the security process is used.

In this example, in the time table 310, identification information for identifying a machine-learned model for watching is stored in association with time information indicating the time from 6:00 to 22:00. This indicates that the edge device 120 performs the watching process using the machine-learned model for watching from 6:00 to 22:00. Further, the time table 310 stores identification information for identifying a machine-learned model for crime prevention in association with time information indicating the time from 22:00 to next 6:00. This indicates that the edge device 120 performs the crime prevention process using the machine-learned model for crime prevention from 22:00 to next 6 o'clock. The correspondence between the time information and the machine-learned model may be determined in advance by a user or an administrator of the server 110, for example.

FIG. 4 is a diagram illustrating an example of a functional configuration of the server 110. The server 110 includes an update unit 211, a first acquisition unit 212, a posting unit 213, a control unit 214, a second acquisition unit 215, an estimation unit 216, and a change unit 217. These functions are realized by the processor 111 performing calculations or controlling communication by the communication unit 114 in cooperation with the server program stored in the memory 112 or the storage 113 and the processor 111 executing the server program. Is done.

The update means 211 updates the machine learned model stored in the storage 113 by machine learning of learning data using a learning device. This “update” means to modify at least a part. For example, the machine-learned model after update has higher accuracy than the machine-learned model before update. The learning data is data used for machine learning. The learning data may be image data indicating an image captured by the imaging unit 125 of the edge device 120, for example. In this case, the image data is transmitted from the edge device 120 to the server 110 at a predetermined time interval.

The first acquisition unit 212 acquires time information from the time table 310. This time information is used to determine whether or not it is time to switch the machine-learned model.

When the posting unit 213 determines that it is time to switch the machine-learned model used by the edge device 120 based on the time information acquired by the first acquisition unit 212, the posting unit 213 indicates that the machine-learned model is switched. Post it. This “posting” means presenting information. This posting may be realized, for example, by storing in the storage 113 information indicating that a machine-learned model is switched.

The control unit 214 controls the edge device 120 to acquire the machine learned model and replace the machine learned model used by the edge device 120. This control may be realized, for example, by acquiring a machine-learned model and transmitting a control signal for replacing the machine-learned model.

The second acquisition unit 215 acquires image data indicating an image captured by the imaging unit 125 of the edge device 120. This image data is used as learning data.

The estimation means 216 analyzes the image data acquired by the second acquisition means 215 to estimate the time that satisfies the condition for starting the process using the machine-learned model. This analysis may include, for example, image recognition processing. This condition is predetermined for each process. For example, when processing is performed when the user is in a certain state, the condition for starting this processing may be that the user has entered that state.

The changing unit 217 changes the time information included in the time table 310 to time information indicating the time estimated by the estimating unit 216.

FIG. 5 is a diagram illustrating an example of a hardware configuration of the edge device 120. The edge device 120 is, for example, a surveillance camera, and performs predetermined processing by inputting image data indicating a captured image to a machine-learned model. The edge device 120 performs different processing depending on time.

The edge device 120 is a computer including a processor 121, a memory 122, a storage 123, a communication unit 124, and an imaging unit 125, for example. These devices are connected via a bus 126. Since the processor 121, the memory 122, the storage 123, the communication unit 124, and the bus 126 are the same as the processor 111, the memory 112, the storage 113, the communication unit 114, and the bus 115 described above, description thereof is omitted. However, the storage 123 does not have a storage capacity for storing many machine-learned models. For example, the storage 123 stores only one machine-learned model used for processing currently performed by the edge device 120. The imaging unit 125 captures an image. The imaging unit 125 shoots an image by forming an image on the imaging element using, for example, an optical system. This image may be a still image or a moving image.

FIG. 6 is a diagram illustrating an example of a functional configuration of the edge device 120. The edge device 120 includes a confirmation unit 221, an acquisition unit 222, a replacement unit 223, and a switching unit 224. These functions are realized by the processor 121 performing calculations or controlling communication by the communication unit 124 in cooperation with the client program stored in the memory 122 or the storage 123 and the processor 121 that executes the program program. Is done.

The confirmation unit 221 confirms whether there is a switch of the machine-learned model based on the posted contents of the server 110. This confirmation may be performed by polling the bulletin information 320 stored in the storage 113 of the server 110, for example.

The acquisition unit 222 acquires a machine learned model from the server 110. More specifically, acquisition of the machine-learned model is performed by pull distribution. That is, the machine-learned model is transmitted in response to a request from the acquisition unit 222. At this time, the machine-learned model may be transmitted using an encrypted communication method. This encrypted communication method refers to a communication method in which data is encrypted for communication. As the encrypted communication method, a known encrypted communication method such as SSL (Secure Sockets Layer) or TLS (Transport Layer Security) may be used.

The replacement unit 223 replaces the machine-learned model stored in the storage 123 (currently used) with the machine-learned model acquired by the acquisition unit 222 according to the control of the control unit 214 of the server 110. This replacement means making a new machine-learned model usable. The replacement may be realized, for example, by storing a new machine learned model in the storage 123 and setting the new machine learned model to be used instead of the old machine learned model.

When the machine-learned model is replaced, the switching unit 224 switches the process performed by the edge device 120 from the process using the old machine-learned model to the process using the new machine-learned model. These processes are performed at different times.

2. Operation 2.1 Switch of Machine Learned Model FIG. 7 is a sequence chart illustrating an example of an operation of switching the machine learned model of the edge device 120. Here, it is assumed that the storage 113 of the server 110 stores a machine-learned model for watching and a machine-learned model for crime prevention. These machine-learned models are updated as needed by the updating unit 211. Further, it is assumed that the machine learning completed model for watching is stored in the storage 123 of the edge device 120, and the edge device 120 performs the watching process using the machine learning completed model for watching.

The server 110 reads the time information from the time table 310 by the first acquisition unit 212, and determines whether or not the time that is the starting point indicated by any of the time information included in the time table 310 is reached based on the time information. (Step 401). This determination may be performed based on, for example, a built-in clock of the server 110. If it is not the start time (NO at Step 401), this determination is repeated until the start time is reached. On the other hand, for example, at 22:00, which is the starting point of the time from 22:00 indicated by the time information included in the time table 310 illustrated in FIG. 3 (the determination in step 401 is YES), the server 110 determines that the machine-learned model The posting information 320 indicating that there is switching is stored in the storage 113 by the posting means 213 (step 402).

FIG. 8 is a diagram illustrating an example of the posting information 320. In the time table 310 shown in FIG. 3, identification information of a machine-learned model for crime prevention is stored in association with time information indicating the time from 22:00 to next 6:00. For example, at 22:00, the device ID “001” for identifying the edge device 120, the information “Yes” indicating that there is a switch of the machine-learned model, and the acquisition destination of the machine-learned model for crime prevention, In the storage 113, an address “http://www.example.com/MCP” indicating the location where the machine-learned model for crime prevention is stored is stored in association with it.

The edge device 120 polls the posting information 320 stored in the storage 113 at predetermined time intervals by the confirmation unit 221 (step 403), and determines whether or not the machine-learned model is switched by the confirmation unit 221. (Step 404). For example, in the bulletin information 320, when information “None” indicating that there is no switching of the machine-learned model is stored in association with the device ID “001”, it is determined that there is no switching of the learned model. (NO at step 404). In this case, the process returns to step 403 described above, and the posting information 320 is polled again after a predetermined time interval. On the other hand, as shown in FIG. 8, in the bulletin information 320, when the information “Yes” indicating that there is a switch of the machine-learned model is stored in association with the device ID “001”, the machine learning has been completed. It is determined that there is a model change (YES in step 404). In this case, the process proceeds to the next step.

In accordance with step 403 described above, the server 110 obtains the next machine learned model using the encrypted communication method, and controls the control signal for replacing the machine learned model stored in the storage 123. The means 214 transmits the result to the edge device 120 whose device ID is “001” (step 405). This control signal includes information specifying the encrypted communication method.

The edge device 120 uses the encrypted communication method in accordance with the received control signal, and the storage indicated by the address “http://www.example.com/MCP” included in the bulletin information 320 shown in FIG. A machine-learned model for crime prevention is acquired from the location in 113 by the acquisition means 222 (step 406). Specifically, the acquisition unit 222 transmits a signal requesting a machine-learned model for crime prevention to the server 110, and receives the machine-learned model for crime prevention transmitted from the server 110 in response to the request for this signal. .

The edge device 120 replaces the machine-learned model for watching stored in the storage 123 with the machine-learned model for crime prevention by the replacement unit 223 according to the received control signal (step 407). For example, after a machine-learned model for watching stored in the storage 123 is deleted, a machine-learned model for crime prevention is stored, and a machine-learned model for crime prevention is used instead of the machine-learned model for watching To be set.

When the machine-learned model for watching is replaced with the machine-learned model for security, the edge device 120 switches the watching process to the security process by the switching unit 224 (step 408). As a result, the edge device 120 performs crime prevention processing using the machine-learned model for crime prevention from 22:00 to next 6 o'clock. At the next 6 o'clock, the same processing as steps 402 to 408 described above is performed, and the machine-learned model for crime prevention stored in the storage 123 is replaced with the machine-learned model for watching, and the crime prevention process is watched over. Switch to processing. Thereby, the edge device 120 performs the watching process from 6 o'clock to 22:00 using the machine learning completed model for watching.

2.2 Optimization of Schedule FIG. 9 is a sequence chart showing an example of an operation for optimizing a schedule for switching machine-learned models. Here, it is assumed that the imaging unit 125 is shooting a place where the user lives.

The edge device 120 transmits image data indicating an image captured by the imaging unit 125 to the server 110 at predetermined time intervals (step 501). When the image data is received by the second acquisition unit 215, the server 110 stores the image data in the storage 113. The server 110 analyzes the image data stored in the storage 113 to estimate a time suitable for starting the watching process and the security process by the estimation unit 216 (step 502). For example, the estimation unit 216 recognizes the user and the user's behavior by performing image recognition processing on the image data. The watching process is preferably performed while the user is awake. In this case, the estimation unit 216 estimates the time when the user wakes up by machine learning of the result of image recognition, and uses it as a time suitable for watching the estimated time and starting the processing. On the other hand, the security process is preferably performed while the user is sleeping. In this case, the estimating unit 216 estimates the time when the user goes to bed by machine learning the result of the image recognition, and uses the estimated time as a time suitable for starting the crime prevention process. The server 110 changes the time table 310 by the changing means 217 based on the estimated time (step 503). For example, it is assumed that 21:00 is estimated as a time suitable for starting the crime prevention process, and 5 o'clock is estimated as a time suitable for starting the watching process. In this case, in the time table 310 shown in FIG. 3, the time information indicating from 22:00 to next 6:00 is changed to time information indicating from 21:00 to next 5:00. Also, the time information indicating 6 o'clock to 22:00 is changed to time information indicating 5 o'clock to 21 o'clock.

In the above-described embodiment, “first acquisition unit 212”, “acquisition unit 222”, “storage 113”, “storage 123”, “watched machine-learned model”, “crime-proof machine-learned model” ”,“ Watching process ”,“ crime prevention process ”are respectively“ acquisition means ”,“ second acquisition means ”,“ first storage means ”,“ second storage means ”,“ first machine learning ”according to the present invention. Used model ”,“ second machine learned model ”,“ first process ”, and“ second process ”.

According to the embodiment described above, when it is time to switch the machine-learned model, a machine-learned model whose purpose is different from the machine-learned model used by the edge device 120 is provided from the server 110 to the edge device 120. Therefore, the machine-learned model used by the edge device 120 can be switched according to time. In addition, since the edge device 120 does not have to have a storage capacity for storing a large number of machine-learned models, the hardware resources necessary for the edge device 120 can be reduced. In addition, although high processing capability is required for updating the machine-learned model, in the above-described embodiment, since the server 110 updates the new machine-learned model, the edge device 120 has high processing capability. It does not have to be.

Furthermore, in the above-described embodiment, the bulletin information 320 and the machine-learned model are transmitted in response to a pull distribution from the edge device 120, that is, in response to a request from the edge device 120. If the server 110 adopts a push-type delivery, that is, a configuration in which the server 110 transmits the data without requesting it from the edge device 120, the edge device 120 impersonates the server 110 as well as the server 110. It is also possible to receive information transmitted by push-side distribution from other devices. In this case, there is a risk that damage is caused by information transmitted by another malicious device. On the other hand, when such data is transmitted from the server 110 in response to a request from the edge device 120 as in the above-described embodiment, such a risk is reduced, so that information security is high. Become.

Furthermore, according to the above-described embodiment, since the machine-learned model is acquired using the encrypted communication method, it is possible to prevent the machine-learned model from being acquired by a malicious third party. Furthermore, according to the above-described embodiment, since the schedule for switching the machine-learned model is optimized according to the user state, the edge device 120 can switch the process according to the schedule according to the user state. .

3. The present invention is not limited to the above-described embodiment. Various modifications may be made to the above-described embodiment. Moreover, the following modifications may be implemented in combination.

In the above-described embodiment, the edge device 120 is not limited to the monitoring camera. The edge device 120 may be any device as long as it performs processing using a machine-learned model. The edge device 120 may include a sensor, and sensor data output from the sensor may be used as learning data.

In the embodiment described above, the number of machine learned models that the edge device 120 switches is not limited to two. The edge device 120 may switch between three or more machine learned models. The machine-learned model is not limited to the machine-learned model for watching and the machine-learned model for crime prevention. For example, the machine-learned model may include a dedicated machine-learned model for accurately performing the process of observing the weather. In short, the plurality of machine-learned models may be machine-learned models used for any process as long as they are machine-learned models having different purposes.

In the embodiment described above, the machine learning algorithm generates a model based on the result of machine learning from given data, for example, on a computer, and inputs new input data to the model. A so-called supervised learning algorithm for outputting an event predicted from the input data may be used. However, the machine learning algorithm is not limited to a so-called supervised learning algorithm, and may be an algorithm for machine learning such as unsupervised learning, semi-supervised learning, reinforcement learning, and expression learning. Further, the algorithm for machine learning may include other learning algorithms such as data mining and deep learning. Note that these learning algorithms use various techniques or techniques such as decision tree learning, correlation rule learning, neural network, genetic programming, inductive logic programming, support vector machine, clustering, and Bayesian network. included. In short, any machine learning algorithm may be used as long as it is processed together with some data provided by the data provider and outputs information desired by the user as a result of the processing.

In the above-described embodiment, the location where the machine-learned model is stored is not limited to the storage 113 of the server 110. For example, when a storage device is connected to the communication network 130, the machine-learned model may be stored in this storage device instead of the storage 113. In this case, the posting information 320 includes an address indicating a location in the storage apparatus. Then, the edge device 120 acquires a machine learned model from the storage device based on this address.

In the above-described embodiment, the posting information 320 may be updated in response to the machine-learned model acquired by the edge device 120. The determination as to whether or not a machine-learned model has been acquired may be made based on whether or not information indicating that a new machine-learned model has been acquired from the edge device 120, for example. When a machine-learned model is acquired by the edge device 120 with the device ID “001”, whether or not the machine-learned model stored in association with the device ID “001” in the bulletin information 320 illustrated in FIG. Is changed from “Yes” to “No”. The machine learning model is changed to “-” indicating that the acquisition destination of the model is not applicable. Thereby, it is possible to prevent the edge device 120 from trying to acquire the acquired machine-learned model again.

The subject that realizes the function of the machine learning model switching system 100 is not limited to the example described in the above embodiment. For example, the edge device 120 or an external device may have a part of the function of the server 110. For example, an external device used by a provider such as an AI (Artificial Intelligent) vendor may have the updating unit 211 instead of the server 110. In this case, the external device updates the machine learned model using the learning data, and transmits the updated machine learned model to the server 110. As described above, the machine-learned model may be updated on the provider side and registered in the server 110. In another example, the edge device 120 may include the estimation unit 216 instead of the server 110. In this case, time information indicating the time estimated by the estimating means 216 is transmitted to the server 110. Further, the server 110 or an external device may have a part of the function of the edge device 120.

The steps of processing performed in the machine learning completed model switching system 100 are not limited to the example described in the above embodiment. The steps of this process may be interchanged as long as there is no contradiction. In addition, the present invention may be provided as a machine-learned model switching method performed in the machine-learned model switching system 100.

The present invention may be provided as a program executed on the server 110 or the edge device 120. This program may be downloaded via a communication line such as the Internet, or a computer such as a magnetic recording medium (magnetic tape, magnetic disk, etc.), an optical recording medium (optical disk, etc.), a magneto-optical recording medium, or a semiconductor memory. May be provided in a state of being recorded on a readable recording medium.

In the embodiment described above, the timing for switching the machine-learned model may be determined by analyzing image data indicating an image captured by the imaging unit 125. In this case, the edge device 120 determines whether or not the condition for starting the process using the machine-learned model is satisfied by analyzing the image data indicating the image captured by the imaging unit 125. When this condition is satisfied, the edge device 120 transmits information indicating that fact to the server 110. Since this information is transmitted when the condition for starting the process using the machine-learned model is satisfied, it is information about time. Therefore, in this modification, this information is used as time information. Upon receiving this information, the server 110 may control the edge device 120 to acquire the corresponding machine learned model. For example, when the user goes to bed, the conditions for starting the crime prevention process are satisfied. Therefore, when it is determined that the user has gone to bed by analyzing the image data, the edge device 120 may be controlled so as to acquire a machine-learned model for crime prevention used for the crime prevention process. . According to this modification, since the machine-learned model is switched according to the user's state, the edge device 120 can switch processing at a timing according to the user's state.

100: Model switching system 110: Server 120: Edge device 211: Update unit 212: First acquisition unit 213: Posting unit 214: Control unit 215: Second acquisition unit 216: Estimation unit 217 : Change means, 221: confirmation means, 222: acquisition means, 223: replacement means, 224: switching means

Claims

A machine-learned model switching system for switching machine-learned models with different purposes for edge devices,
An acquisition means for acquiring time information relating to time;
Control means for controlling the edge device to acquire a machine-learned model associated with the time information;
Machine learning completed model switching system.
The machine learning model switching system according to claim 1, wherein the control unit controls the edge device to acquire the machine learned model using an encrypted communication method.
Update means for updating the machine-learned model by machine learning of learning data using a learning device;
The machine learning model switching system according to claim 1, wherein the control unit controls the edge device to acquire the updated machine learned model.
The acquisition means is a first acquisition means,
Storage means for storing time information indicating a plurality of times and identification information for identifying a plurality of machine-learned models having different purposes from each other;
Second acquisition means for acquiring sensor data output from a sensor of the edge device;
By estimating the sensor data acquired, an estimation unit that estimates a time that satisfies a condition for starting a plurality of processes using the plurality of machine-learned models;
Changing means for changing the time information stored in the storage means to time information indicating the estimated time;
When it is determined that the starting point of any one of the plurality of times is based on the time information stored in the storage unit, the control unit determines the time among the plurality of machine-learned models. The machine-learned model switching according to any one of claims 1 to 3, wherein control is performed to cause the edge device to acquire a machine-learned model identified by the identification information stored in association with time information to be displayed. system.
First storage means for storing time information indicating a plurality of times in association with identification information for identifying a plurality of machine-learned models having different purposes;
When it is determined based on the time information stored in the first storage means that any one of the plurality of times has started, there is a switch of the first machine learned model used by the edge device. And an address indicating a location where the second machine learned model identified by the identification information stored in association with the time information indicating the time among the plurality of machine learned models is stored. And posting means for posting
The acquisition means is a first acquisition means,
The edge device is
Second storage means for storing the first machine-learned model;
Confirmation means for confirming whether there is a switch of the first machine-learned model by polling the posted information;
When there is a switch of the first machine learned model, a second acquisition unit that acquires the second machine learned model from the location indicated by the posted address according to the control of the control unit;
Replacement means for replacing the first machine learned model stored in the second storage means with the acquired second machine learned model;
When the first machine learned model is replaced with the second machine learned model, the processing performed by the edge device is changed from the first process using the first machine learned model to the second machine learned model. The machine learning-completed model switching system according to any one of claims 1 to 4, further comprising switching means for switching to the second process used.
In response to determining that the condition for starting the process using the machine-learned model has been satisfied by analyzing the sensor data output from the sensor of the edge device, the control means, The machine-learned model switching system according to claim 1, wherein control is performed to cause the edge device to acquire the machine-learned model.
An edge device that switches between different machine-learned models for different purposes,
An edge device that acquires the machine-learned model according to the control of the machine-learned model switching system according to claim 1 and replaces the other machine-learned model currently used with the acquired machine-learned model.
A machine-learned model switching method for switching machine-learned models with different purposes for edge devices,
Obtaining time information about time;
Controlling the edge device to acquire a machine-learned model associated with the time information;
A machine-learned model switching method comprising:
A program for causing a computer to execute a machine-learned model switching process for switching a machine-learned model with a different target for an edge device,
Obtaining time information about time;
Controlling the edge device to acquire a machine-learned model associated with the time information;
A program for running