JP6828216B2 - Machine-learned model switching system, edge device, machine-learned model switching method, and program - Google Patents

Description

The present invention relates to a machine-learned model switching system, an edge device, a machine-learned model switching method, and a program used in the field of IoT (Internet of Things).

In recent years, advances in AI (artificial intelligence) technology have attracted attention for learning model update technology. For example, Patent Document 1 proposes a technique for updating a model of a machine learning system.

JP-A-2017-120647

In the system described in Patent Document 1, a model of a machine learning system is trained, but the purpose of processing does not change even if the model of the machine learning system is trained. By the way, there are cases where one edge device is desired to be used for different purposes depending on the communication path. For example, when an edge device performs processing using a machine-learned model according to its use, it is necessary to switch the machine-learned model according to a change in the use of the edge device. However, for example, if the edge device does not have sufficient hardware resources, the edge device cannot store a large number of machine-learned models and switch between them depending on the application.

An object of the present invention is to switch the machine-learned model used by an edge device according to a communication path.

According to the embodiment of the present invention, the first machine-learned edge device used is an edge device connected to the server via a communication line and is used according to the communication path information indicating the communication path connecting the server and itself. An edge device is provided that acquires a second machine-learned model having a different purpose from the model from a server and switches the first machine-learned model to the second machine-learned model.
According to the embodiment of the present invention, it is a machine-learned model switching system including a server and an edge device connected to the server via a communication line, and the edge device indicates a communication path connecting the server and itself. According to the communication path information, the second machine-learned model whose purpose is different from the first machine-learned model used is acquired from the server, and the first machine-learned model is switched to the second machine-learned model. A machine-learned model switching system is provided.
According to the embodiment of the present invention, it is a machine-learned model switching method by an edge device connected to a server via a communication line, and the edge device responds to communication path information indicating a communication path connecting the server and itself. Then, the second machine-learned model whose purpose is different from the first machine-learned model used is acquired from the server, and the first machine-learned model is switched to the second machine-learned model. Machine-learned model switching A method is provided.
According to the embodiment of the present invention, the edge device connected to the server via the communication line has the first machine-learned model used according to the communication path information indicating the communication path connecting the server and itself. Is provided with a program for acquiring a second machine-learned model having a different purpose from a server and executing a step of switching the first machine-learned model to the second machine-learned model.
According to the embodiment of the present invention, the machine-learned model switching system is a machine-learned model switching system that switches a machine-learned model having a different purpose to an edge device, and is a communication indicating a communication path of the edge device. An acquisition means for acquiring route information, a control means for controlling the edge device to acquire a machine-learned model associated with the communication route information, and another machine already applied to the edge device. A machine-learned model switching system including a switching means for switching a trained model to the acquired machine-learned model is provided.

According to the embodiment of the present invention, the acquisition means acquires the communication path information that identifies the node with which the edge device directly communicates among the communication paths, and the control means informs the edge device of the communication path. A machine-learned model switching system that controls to acquire the machine-learned model associated with the node identified by information is provided.

According to the embodiment of the present invention, the control means controls the edge device to acquire the machine-learned model associated with the number of the edge devices that directly communicate with the node. A switching system is provided.

According to the embodiment of the present invention, the acquisition means is provided with a machine-learned model switching system that acquires the communication path information when the communication path of the edge device changes.

According to the embodiment of the present invention, the acquisition means acquires environmental information indicating the environment around the edge device, and the control means corresponds to the edge device with the communication path information and the set of the environmental information. A machine-learned model switching system is provided that controls the acquisition of the attached machine-learned model.

According to the embodiment of the present invention, the control means specifies the number of nodes of the communication path indicated by the communication path information, and causes the edge device to acquire the machine-learned model associated with the number of nodes. A machine-learned model switching system is provided.

According to the embodiment of the present invention, the machine learning model is acquired and another machine learning model already applied is acquired according to the control of the machine learning model switching system of the embodiment of the present invention. An edge device is provided to switch to the finished model.

According to the embodiment of the present invention, it is a machine-learned model switching method for switching a machine-learned model having a different purpose for an edge device, and a step of acquiring communication path information indicating a communication path of the edge device. The machine that has acquired a step of controlling the edge device to acquire a machine-learned model associated with the communication path information and another machine-learned model already applied to the edge device. A machine-learned model switching method with a step to switch to a trained model is provided.

According to the embodiment of the present invention, the computer is made to acquire the communication path information indicating the communication path of the edge device, and the edge device is made to acquire the machine-learned model associated with the communication path information. A program is provided for executing a step of controlling and a step of switching the other machine-learned model already applied to the edge device to the acquired machine-learned model.

According to the present invention, the machine-learned model used by the edge device can be switched according to the communication path.

The figure which shows the structure of the machine-learned model switching system 9 which concerns on this embodiment. The figure which shows the example of the structure of the provision device 1. The figure which shows the example of the model DB 121. The figure which shows the example of the communication path DB 122. The figure which shows the example of the structure of the edge device 2. The figure which shows the functional configuration of the providing device 1 and the edge device 2. The sequence diagram which shows the operation flow of the machine-learned model switching system 9.

1 ... Providing device, 11 ... Control unit, 110 ... Control means, 111 ... Acquisition means, 112 ... Specific means, 113 ... Providing means, 114 ... Update means, 12 ... Storage unit, 121 ... Model DB, 122 ... Communication path DB , 13 ... communication unit, 2 ... edge device, 21 ... control unit, 211 ... transmission means, 212 ... switching means, 213 ... execution means, 22 ... storage unit, 221 ... machine learning model, 222 ... teacher data, 23 ... Communication unit, 24 ... Operation unit, 25 ... Display unit, 27 ... Detection unit, 3 ... Communication line, 31 ... Router, 32 ... Access point, 9 ... Machine learning model switching system.

<Embodiment>
<Overall configuration of machine-learned model switching system>
FIG. 1 is a diagram showing a configuration of a machine-learned model switching system 9 according to the present embodiment. The machine-learned model switching system 9 is a system that switches the machine-learned model used by the edge device 2 according to the communication path in which the edge device 2 connects to the providing device group C. It has an edge device 2 and a communication line 3 for communicably connecting them.

The providing device group C has a plurality of providing devices 1, and the plurality of providing devices 1 communicate with each other and cooperate with each other to share the resources provided for each and to perform a plurality of machine learnings having different purposes. Generate a completed model.

The resource functions of each of the providing devices 1 include, for example, data collection / storage, machine learning using these data, management of machine-learned models obtained by machine learning, and distribution of these various functions. Supervision, management, etc.

The plurality of providing devices 1 constituting the providing device group C may communicate with each other by using, for example, the ad hoc mode in IEEE802.11. Further, each of these providing devices 1 may communicate with each other via the communication line 3.

In the present embodiment, the machine-learned model switching system 9 includes a providing device group C having a plurality of providing devices 1, but the machine-learned model switching system 9 uses a machine-learned model on the edge device 2. The configuration may include only one providing device 1 to be provided.

The plurality of edge devices 2 are terminal devices owned by the user, such as a personal computer, a smartphone, and a tablet PC, each of which is connected to the providing device group C via the communication line 3 from the providing device 1. Receive a machine-learned model. The edge device 2 may be movable.

The communication line 3 is a line that connects the providing device 1 and the edge device 2 so as to be communicable, and is, for example, the Internet. The communication executed by the communication line 3 may be any communication as long as the providing device 1 can acquire the information of the communication path connecting the providing device 1 and the edge device 2, but the following description describes it. It will be described as communication by packet switching.

When the communication line 3 performs communication by packet switching, as shown in FIG. 1, the communication line 3 has a router 31 that relays communication with the providing device 1 or the edge device 2 to control the communication path. Further, the communication line 3 may have an access point 32 for wirelessly connecting to the providing device 1 or the edge device 2.

<Configuration of provided equipment>
FIG. 2 is a diagram showing an example of the configuration of the providing device 1. The providing device 1 has a control unit 11, a storage unit 12, and a communication unit 13.

The control unit 11 has a CPU (Central Processing Unit), a ROM (Read Only Memory), and a RAM (Random Access Memory), and a computer program in which the CPU is stored in the ROM and the storage unit 12 (hereinafter, simply referred to as a program). Is read and executed to control each part of the providing device 1.

The communication unit 13 is a communication circuit connected to the communication line 3 by wire or wirelessly. The communication unit 13 may have a function of directly connecting to another providing device 1 by wire or wirelessly. The providing device 1 exchanges information with the edge device 2 connected to the communication line 3 by the communication unit 13.

The storage unit 12 is a storage means for a solid state drive, a hard disk drive, or the like, and stores various programs, data, and the like read into the CPU of the control unit 11. Further, the storage unit 12 has a model DB 121 and a communication path DB 122.

FIG. 3 is a diagram showing an example of the model DB 121. The model DB 121 is a database that stores a model ID, which is identification information for identifying a machine-learned model, and a machine-learned model identified by the model ID in association with each other.

Machine-learned models have different purposes and uses. For example, the machine-learned model of the model ID “M01” shown in FIG. 3 is a model for prioritizing communication related to work or giving advice regarding work as compared with private use.

On the other hand, for example, the machine-learned model of the model ID "M02" is a model for suppressing communication related to work or giving advice to limit the time spent on the work of the user as compared with private use.

FIG. 4 is a diagram showing an example of the communication path DB 122. The communication path DB 122 is a database that stores the communication path ID and the model ID in association with each other. The communication path ID is identification information that identifies each of the communication paths connecting the providing device 1 and the edge device 2.

In communication by packet switching, a communication path is configured to include one or more relay devices (referred to as nodes). Identification information such as IPv4 (Internet Protocol version 4) and IPv6 (Internet Protocol version 6) is assigned to each node. The communication path information is, for example, identification information of each node included in this communication path. In packet switching, since the identification information of the source and the destination is attached to each packet, the communication path is multiplexed without being occupied by the source and the destination.

The communication path may be identified, for example, by a node that communicates directly with the edge device 2. For example, if the edge device 2 wirelessly connects to the communication line 3, this node is an access point 32 to which the edge device 2 directly connects, and the edge device 2 connects to the communication line 3 by priority. For example, it is a router 31 to which the edge device 2 is directly connected.

The model ID is identification information that identifies a machine-learned model to be applied to the edge device 2 when the edge device 2 communicates with the provider 1 via the communication path identified by the communication path ID described above. Is.

In the communication path DB 122, the same machine-learned model may be associated with a plurality of communication paths, but in the communication path DB 122, any one machine-learned model is always specified for one communication path. It is configured to be.

Further, the machine-learned model identified by the model ID stored in the communication path DB 122 may be stored in the model DB 121, but may not be stored in the storage unit 12 of the providing device 1. In this case, if the communication path DB 122 stores the model ID of the machine-learned model stored in any database accessible by the providing device 1 via the communication line 3 in association with the communication path ID. Good.

<Edge device configuration>
FIG. 5 is a diagram showing an example of the configuration of the edge device 2. The edge device 2 includes a control unit 21, a storage unit 22, a communication unit 23, an operation unit 24, a display unit 25, and a detection unit 27.

The control unit 21 has a CPU, a ROM, and a RAM, and the CPU controls each unit of the edge device 2 by reading and executing a program stored in the ROM and the storage unit 22.

The communication unit 23 is a communication circuit that connects to the communication line 3 by wire or wirelessly. The edge device 2 exchanges information with the providing device 1 connected to the communication line 3 by the communication unit 23.

The operation unit 24 includes operation buttons such as an operation button, a keyboard, and a touch panel for giving various instructions, receives an operation by the user, and sends a signal according to the operation content to the control unit 21.

The display unit 25 has a display screen such as a liquid crystal display, and displays an image under the control of the control unit 21. The transparent touch panel of the operation unit 24 may be superposed on the display screen.

In addition to or in place of the display unit 25, the edge device 2 may include a sound emitting unit such as a speaker that emits sound under the control of the control unit 21. In this case, the control unit 21 of the edge device 2 may output sound from this sound emitting unit.

Further, the edge device 2 may include an interface for connecting to an external device in addition to or in place of the display unit 25. In this case, the control unit 21 of the edge device 2 may control an external device connected through this interface. The external device connected to the interface is, for example, a video / audio device or the like.

The detection unit 27 is a sensor or the like that detects the environment around the edge device 2 and generates information (referred to as environmental information) indicating the detected environment. The detection unit 27 may be, for example, a volume meter for measuring the ambient volume or a thermometer for measuring the ambient temperature. Further, the detection unit 27 may be a vibrometer that detects vibration, or a biological sensor that detects the user's pulse, respiratory rate, and the like.

Further, the detection unit 27 may have, for example, a receiver that receives (detects) navigation signals from a plurality of navigation satellites based on the Global Navigation Satellite System (GNSS). The detection unit 27 may have a processor that calculates the position of the edge device 2 based on the received navigation signal. Further, the control unit 21 may calculate the position of the edge device 2 using the plurality of navigation signals received by the detection unit 27.

The storage unit 22 is a storage means for a solid state drive, a hard disk drive, or the like, and stores various programs, data, and the like read into the CPU of the control unit 21. The storage unit 22 stores the machine-learned model 221. Only one machine-learned model 221 can be stored by the storage unit 22.

Further, the storage unit 22 may have the teacher data 222. The teacher data 222 is data in which the control unit 21 of the edge device 2 performs processing such as calculation and inference using the machine-learned model 221 to give an evaluation to the output obtained.

The evaluation of the output of the edge device 2 may be performed based on an operation performed each time the output is performed by the user who owns the edge device 2, or the detection result of the detection unit 27 is stored in the storage unit 22 in advance as a reference. It may be done by comparing with a value or the like.

<Functional configuration of providing device and edge device>
FIG. 6 is a diagram showing a functional configuration of the providing device 1 and the edge device 2. In FIG. 6, the communication unit 13, the communication unit 23, and the communication line 3 are omitted. The control unit 11 of the providing device 1 functions as the acquiring means 111, the specifying means 112, the providing means 113, and the updating means 114 by reading and executing the program stored in the storage unit 12.

These functions may be realized by the control unit 11 of one providing device 1, but the control units 11 of a plurality of providing devices 1 constituting the providing device group C communicate with each other and perform mutual calculation. It may be realized by sharing resources such as functions and storage contents of the storage unit 12.

The control unit 21 of the edge device 2 functions as the transmission means 211, the switching means 212, and the execution means 213 by reading and executing the program stored in the storage unit 22.

When the communication unit 23 communicates with any of the providing devices 1 via the communication line 3 in the edge device 2, the transmitting means 211 transmits a request for establishing a communication path between itself and the providing device 1. .. At this time, the transmitting means 211 may request communication with, for example, a specific providing device 1, but requests communication from the providing device 1 that satisfies the condition among the providing devices 1 belonging to the providing device group C. You may.

In the communication line 3, a plurality of routers 31 and access points 32 establish a communication path in response to this request and start communication between any one of the providing devices 1 and the edge device 2.

When the communication path with the edge device 2 is established, the acquisition means 111 acquires the communication path information indicating the communication path in the providing device 1 and delivers it to the specific means 112.

The identification means 112 identifies the machine-learned model according to the communication path as indicated by the communication path information acquired by the acquisition means 111.

The identification means 112 identifies the communication path indicated by the communication path information, and refers to the communication path DB 122 to specify the machine-learned model associated with the communication path. That is, the identification means 112 identifies the machine-learned model according to the communication path established by the providing device 1 with the edge device 2.

The providing means 113 reads the machine-learned model specified by the specifying means 112 from the model DB 121 and provides it to the edge device 2.

The provision of the machine-learned model by the providing means 113 may be a so-called push-type provision in which the machine-learned model is transmitted from the providing device 1 to the edge device 2, or may be provided to the edge device 2. It may be a so-called pull-type offer that allows access to obtain a machine-learned model.

In short, as shown in FIG. 6, the specifying means 112 and the providing means 113 serve as the control means 110 that controls the edge device 2 to acquire the machine-learned model associated with the communication path information of the edge device 2. It just works.

Further, when the machine-learned model specified by the specific means 112 is stored in a database other than the model DB 121, the providing means 113 acquires the above-mentioned machine-learned model from the database and provides the providing device. It may be provided to the edge device 2 that has sent the communication path information to 1. In this case, the specific means 112 uses machine learning stored in a database capable of connecting the machine-learned model associated with the communication path information between the providing device 1 and the edge device 2 via the communication line 3. You can specify it from the completed models.

The providing means 113 may provide a machine-learned model 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 well-known encrypted communication method such as SSL (Secure Sockets Layer) or TLS (Transport Layer Security) may be used.

When the machine-learned model is provided to the edge device 2, the switching means 212 switches the old machine-learned model 221 stored in the storage unit 22 by the provided new machine-learned model.

That is, the edge device 2 acquires the machine-learned model under the control of the machine-learned model switching system 9, and switches to the acquired machine-learned model by acquiring another machine-learned model to which it applies.

Further, the providing means 113 provides the edge device 2 with the machine-learned model specified by the specific means 112, and the machine-learned model switches another machine-learned model applied to the edge device 2. ..

Switching the old machine-learned model with the new machine-learned model includes replacing all of the old machine-learned model with the new machine-learned model, but updating at least a part of it.

The execution means 213 reads the machine-learned model from the storage unit 22 and executes a process using the machine-learned model. The executing means 213 causes the display unit 25 to display the output obtained by this processing, for example.

When the edge device 2 includes the above-mentioned sound emitting unit in addition to or in place of the display unit 25, the executing means 213 sends the output obtained by the above-mentioned processing to the sound emitting unit to this output. The sound corresponding to may be emitted.

Further, when the edge device 2 includes the above-mentioned interface in addition to or in place of the display unit 25, the executing means 213 may send the output obtained by the above-mentioned processing to an external device via this interface. .. In this case, the external device is controlled according to the transmitted output, and for example, the reproduction of music by the sound device is controlled.

Further, the executing means 213 evaluates the output obtained by the processing using the machine-learned model, generates the teacher data 222 according to the evaluation, and stores it in the storage unit 22.

The evaluation of this output may be performed in association with the environmental information detected by the detection unit 27. Further, the evaluation of this output may be performed according to the content input by the user by operating the operation unit 24.

The teacher data read from the teacher data 222 is transmitted to the providing device 1 by the transmitting means 211 when a predetermined condition is satisfied.

When the providing device 1 receives the teacher data transmitted from each of the edge devices 2, the acquisition means 111 acquires these teacher data. Then, for example, when the aggregated teacher data satisfies a predetermined condition, the update means 114 receives the aggregated teacher data from the acquisition means 111, performs machine learning based on the teacher data, and has completed new machine learning. The machine-learned model is updated by generating a model and storing it in the model DB 121.

In this case, the machine learning algorithm performed by the updating means 114 is a so-called supervised learning algorithm that generates a model suitable for the teacher data.

<Operation of machine-learned model switching system>
FIG. 7 is a sequence diagram showing an operation flow of the machine-learned model switching system 9. The edge device 2 transmits a communication request to any of the providing devices 1 (step S101).

Upon request from the edge device 2, the communication line 3 selects one of the providing devices 1 from the providing device group C, and establishes a communication path between the providing device 1 and the edge device 2.

When the communication path is established, the providing device 1 acquires the communication path information indicating the communication path established with the edge device 2 and identifies the machine-learned model associated with the communication path. (Step S102). Then, the providing device 1 provides the specified machine-learned model to the edge device 2 (step S103).

When the edge device 2 receives the machine-learned model from the providing device 1, the machine-learned model stored in the storage unit 22 is switched by the received machine-learned model (step S104), and the machine after switching is switched. The determined process is executed using the trained model (step S105).

When the process is executed and an output is obtained, the edge device 2 evaluates the output and generates teacher data based on the evaluation. The edge device 2 stores the generated teacher data (step S106).

For example, when the predetermined condition is satisfied, the edge device 2 transmits the stored teacher data to the providing device 1 (step S107). Upon receiving this teacher data, the providing device 1 performs learning based on the teacher data and updates the machine-learned model stored by itself (step S108).

By the above operation, the edge device 2 acquires a machine-learned model having a different purpose according to the communication path established with the providing device 1, and executes the process using the machine-learned model, so that the process according to the communication path is performed. It can be carried out.

For example, according to the machine-learned model switching system 9 of the present invention, it is possible to determine for each requirement or source an email that is permitted to be received or transmitted when the edge device 2 is at work and at home. You can change the content such as music and video to be played.

Specifically, when the user connects the edge device 2 to the access point 32 in the workplace, the edge device 2 is provided with a machine-learned model for the workplace, and the model is switched. Therefore, the user is permitted by the machine-learned model switching system 9 to work with the edge device 2.

Further, when the user connects the edge device 2 to the access point 32 at home, the edge device 2 is provided with a machine-learned model for home use, and the model is switched. Therefore, the user is prohibited by the machine-learned model switching system 9 from doing more than a certain amount of work using the edge device 2. This prevents the user from overworking.

Further, according to the machine-learned model switching system 9 of the present invention, the edge device 2 does not have to have a storage capacity sufficient to store a large number of machine-learned models, and thus requires necessary hardware resources. Can be reduced.

Further, according to the machine-learned model switching system 9 of the present invention, since any of the providing devices 1 constituting the providing device group C generates a new machine-learned model, the edge device 2 has been machine-learned. You don't need the processing power to generate a model.

Further, when the machine-learned model switching system 9 of the present invention exchanges a machine-learned model by using an encrypted communication method, a malicious third party is unlikely to steal the machine-learned model.

<Modification example>
The above is the description of the embodiment, but the content of this embodiment can be modified as follows. Moreover, the following modification examples may be combined.

<Modification example 1>
In the above-described embodiment, the transmitting means 211 has transmitted a request for establishing a communication path between itself and the providing device 1 to the providing device 1, but in addition to this, the environmental information generated by the detection unit 27 has been transmitted. May be transmitted to the providing device 1. The acquisition means 111 acquires the communication route information and the environment information and delivers them to the specific means 112, and the specific means 112 specifies the machine-learned model according to the set of the delivered communication route information and the environment information. Good.

In this case, the communication path DB 122 stored in the storage unit 12 of the providing device 1 may be configured so that the correspondence table between the communication path ID and the model ID can be selected according to the environmental information. The communication path DB 122 may classify the values of environmental information such as the detected volume and temperature into a plurality of grades, and may be associated with different correspondence tables for each grade.

<Modification 2>
Further, when the communication path is identified by a node that directly communicates with the edge device 2, the identifying means 112 may specify the machine-learned model according to the number of edge devices 2 that directly communicate with this node. In this case, the acquisition means 111 acquires communication route information indicating the communication routes established with the plurality of edge devices 2. Then, the identification means 112 identifies the nodes with which the edge devices 2 directly communicate from the acquired communication path information, and monitors the number of edge devices 2 connected to the nodes for each of the nodes.

The communication path DB 122 may classify the number of edge devices 2 that directly communicate with a common node into a plurality of grades, and store the model IDs in association with each classification. The specific means 112 monitors the operating rate obtained by dividing this number by the processing capacity of the node, etc., instead of the number of edge devices 2 that directly communicate with the common node, and has completed machine learning according to this operating rate. The model may be specified.

<Modification example 3>
In the above-described embodiment, the transmission means 211 has transmitted a request for establishing a communication path between itself and the providing device 1 to the providing device 1, but the communication path changes after the communication path is established. At that time, communication route information indicating the communication route may be transmitted. In this case, in the providing device 1, the acquisition means 111 acquires the communication route information indicating the changed communication route.

<Modification example 4>
In the above-described embodiment, the updating means 114 receives the teacher data transmitted from each of the edge devices 2, performs machine learning based on the teacher data, and generates a new machine-learned model. Although the trained model was updated, machine learning may be performed using other teacher data.

The machine learning algorithm performed by the update means 114 is a supervised learning algorithm, but is not limited to this, and is an algorithm for machine learning such as unsupervised learning, semi-supervised learning, enhanced learning, and expression learning. It may be.

Further, the algorithm for machine learning may include other algorithms for learning such as data mining and deep learning. These learning algorithms include those using various technologies such as decision tree learning, correlation rule learning, neural networks, genetic programming, inductive logic programming, support vector machines, clustering, and Bayesian networks. .. In short, the algorithm for machine learning may be processed together with some data provided by the data provider, and as a result of the processing, the information desired by the user may be output.

<Modification 5>
Further, the providing device 1 specifies the number of nodes included in the communication path indicated by the communication path information, and controls the edge device 2 to acquire the machine-learned model associated with the number of nodes. May be good.

In this case, the specifying means 112 specifies the number of nodes (referred to as the number of nodes) included in the communication path from the providing device 1 to the edge device 2 from the communication route information acquired by the acquiring means 111, and responds to the number of nodes. Identify the machine-learned model. The communication path DB 122 may classify the number of nodes of the communication path into a plurality of grades, and store the model ID in association with each classification.

Further, the providing device 1 specifies the communication speed (communication speed) in the communication path indicated by the communication path information, and controls the edge device 2 to acquire the machine-learned model associated with the communication speed. You may.

In this case, the specifying means 112 specifies the communication speed between the providing device 1 and the edge device 2 from the communication path information acquired by the acquiring means 111, and specifies the machine-learned model according to the communication speed. The communication path DB 122 may classify the communication speeds in the communication path into a plurality of grades, and store the model IDs in association with each of the classifications.

<Modification 6>
In the above-described embodiment, the communication executed by the communication line 3 has been described as communication by packet switching, but for example, data communication in which the providing device 1 and the edge device 2 occupy a dedicated line may be used. In short, if the providing device 1 can acquire the communication path information indicating the communication path of the edge device, the communication performed on this communication path does not have to be the communication by packet switching.

<Modification 7>
The process executed by the control unit 11 of the providing device 1 can be conceived as a machine-learned model switching method for providing the machine-learned model to the edge device 2 that executes the process using the stored machine-learned model. That is, the present invention is a machine-learned model switching method for switching a machine-learned model having a different purpose for the edge device 2, and includes a step of acquiring communication path information indicating a communication path of the edge device 2 and an edge. The step of controlling the device 2 to acquire the machine-learned model associated with the communication path information, and the acquired machine-learned model of another machine-learned model already applied to the edge device 2. It may be provided as a machine-learned model switching method having a step of switching to.

<Modification 8>
The programs executed by the providing device 1 and the edge device 2 are recordings readable by a computer device such as a magnetic recording medium such as a magnetic tape and a magnetic disk, an optical recording medium such as an optical disk, an optical magnetic recording medium, and a semiconductor memory. It can be provided as stored on a medium. Further, this program may be downloaded via a communication line such as the Internet. In addition to the CPU, various devices may be applied as the control means described above, and for example, a dedicated processor or the like is used.

Claims (9)

An edge device connected to the server via a communication line
A second machine-learned model having a purpose different from that of the first machine-learned model used is acquired from the server according to the communication path information indicating the communication path connecting the server and itself, and the first machine-learned model is obtained. An edge device that switches a machine-learned model to the second machine-learned model. Among the communication paths indicated by the communication path information, the second machine-learned model is acquired from the server according to the node with which it directly communicates, and the first machine-learned model is obtained by the second machine learning. The edge device according to claim 1, which switches to a finished model. According to claim 2, the second machine-learned model is acquired from the server and the first machine-learned model is switched to the second machine-learned model according to the number of edge devices that directly communicate with the node. Described edge device. A claim that acquires the second machine-learned model from the server and switches the first machine-learned model to the second machine-learned model according to the communication path information when the communication path changes. The edge device according to 1. The second machine-learned model is acquired from the server and the first machine-learned model is switched to the second machine-learned model according to the communication path information and the environment information indicating the surrounding environment. The edge device according to any one of claims 1 to 4. The second machine-learned model is acquired from the server and the first machine-learned model is switched to the second machine-learned model according to the number of nodes of the communication path shown in the communication path information. The edge device according to any one of claims 1 to 5. A machine-learned model switching system that includes a server and an edge device connected to the server via a communication line.
The edge device acquires a second machine-learned model having a different purpose from the first machine-learned model used from the server according to the communication path information indicating the communication path connecting the server and itself. A machine-learned model switching system that switches the first machine-learned model to the second machine-learned model. A machine-learned model switching method using an edge device connected to a server via a communication line.
The edge device acquires a second machine-learned model having a different purpose from the first machine-learned model used from the server according to the communication route information indicating the communication path connecting the server and itself. Then, the machine-learned model switching method for switching the first machine-learned model to the second machine-learned model. To an edge device connected to the server via a communication line,
A second machine-learned model having a purpose different from that of the first machine-learned model used is acquired from the server according to the communication path information indicating the communication path connecting the server and itself, and the first machine-learned model is obtained. A program for executing a step of switching a machine-learned model to the second machine-learned model.

Images