JP7551458B2 - Model data providing method, model data providing system, server device, and computer program - Google Patents

Description

The present invention relates to a method for providing model data for practical application of distributed learning, a system for providing model data, a server device, and a computer program.

Judgment, recognition, etc. using learning models based on deep learning have been put to practical use. It has been confirmed that learning models can be used in a variety of technical fields. Even if a huge amount of training data is used to improve the accuracy of the learning models to a level that can be used in various fields, it takes a long time for the parameters to converge, and there is no guarantee that the accuracy will improve.

Patent Document 1 discloses a method in which multiple replicas of a learning model to be learned are prepared, and these multiple model replicas learn independently and asynchronously. In Patent Document 1, a parameter server is fragmented into multiple parts, and each of the multiple replicas of the learning model asynchronously acquires parameters from the fragmented parameter servers, learns from them, and returns the parameters to each parameter server in a repeated manner. This type of distributed processing is said to allow the parameters of the learning model to converge quickly.

U.S. Pat. No. 8,768,870

Even in the deep learning method using distributed processing as disclosed in Patent Document 1, training data is aggregated. However, data in fields such as medicine, finance, and authentication is personal data and is highly confidential. In order to aggregate data as training data to improve the accuracy of the model, each individual's consent is required to provide the data, and even if consent is obtained, there is always a risk to the security of data management.

The present invention has been made in consideration of the above circumstances, and aims to provide a model data provision method, a model data provision system, a server device, and a computer program that promote the practical application of a learning model based on distributed learning.

A model data provision method according to an embodiment of the present disclosure includes distributing global model data stored in a server to a plurality of nodes, having each node learn a local model based on the global model using local data processed by the plurality of nodes, acquiring the learned local model and version information of the local model from each node, updating the global model data based on the plurality of local model data acquired from each of the plurality of nodes, storing the acquired local model data and the updated global model data in association with the version information, accepting a selection from the global model and the local model associated with different version information, and distributing the selected model data to the plurality of nodes or other devices.

The model data provision system of one embodiment of the present disclosure includes a plurality of nodes each processing local data, and a server connected to the plurality of nodes for communication, the server distributes global model data to the plurality of nodes, the plurality of nodes each proceeds with learning of a local model from the global model using data of the local data to be processed, the processing unit of the server acquires version information of the local model and the local model after learning from the plurality of nodes, updates the data of the global model based on the data of the local model after learning acquired from each of the plurality of nodes, stores the acquired local model data and the updated global model data in association with the version information, accepts a selection from the global model and the local model associated with different version information, and distributes the data of the selected model to the plurality of nodes or other devices.

A server device according to an embodiment of the present disclosure includes a communication unit that communicates with a plurality of nodes and a processing unit for storing global model data, and distributes the stored global model data to the plurality of nodes by the processing unit, acquires local model data learned based on the global model using local data processed by the plurality of nodes in association with version information, updates the global model data based on the learned local model data acquired from each of the plurality of nodes, stores the acquired local model data and the updated global model data in association with version information, accepts a selection from the global model and local models associated with different version information, and distributes the selected model data to the plurality of nodes or other devices.

A computer program according to an embodiment of the present disclosure causes a computer capable of communication with multiple nodes to execute a process of distributing stored global model data to the multiple nodes, acquiring local model data learned based on the global model using local data processed by the multiple nodes in association with version information, updating the global model data based on the learned local model data acquired from each of the multiple nodes, storing the acquired local model data and the updated global model data in association with version information, accepting a selection from a global model and a local model associated with different version information, and distributing the selected model data to the multiple nodes or other devices.

In the model data provision method, model data provision system, server device, and computer program disclosed herein, data of the global model is updated by local models trained at each node, and learning using a large amount of data is realized without aggregating data for learning at the server. The models that can be selected for actual use at a node include not only global models, but also local models trained at other nodes.

The model data provision method of one embodiment of the present disclosure includes a process of having the multiple nodes calculate evaluation values based on the accuracy of a global model and a local model associated with different version information, based on the local data to be processed by each node, and accepting a selection of the global model or the local model based on the calculated evaluation values.

In the model data provision method disclosed herein, evaluation values are calculated and stored for not only global models but also local models, and can be viewed by other nodes. This allows nodes to use models that have been learned by other nodes and are suitable for the node itself.

The model data provision method of one embodiment of the present disclosure accepts the selection of multiple global models and/or local models from global models and local models associated with different version information, and causes each of the global model and/or local model to identify a result for the data based on a comparison of multiple output results obtained by inputting data to be processed by the node.

A computer program according to an embodiment of the present disclosure causes a computer connected to a server to repeatedly perform a process of learning a local model using local data to be processed based on global model data distributed from the server, and a process of transmitting the learned local model data to the server, to acquire global model data and local model data associated with different version information, and to execute a process of inputting data into each of a plurality of global models and/or local models corresponding to the acquired data and identifying a result for the data based on a comparison of a plurality of output results obtained.

In the model data provision method and computer program disclosed herein, multiple specifications are possible from a list of global models and local models, and when data is input for multiple models, the results obtained from each of the multiple models can be compared. As a result of the comparison, it is possible to obtain more accurate results by identifying the results for the data from the results obtained from more models.

The model data provision method of one embodiment of the present disclosure has each of the multiple nodes select a model from among global models and local models associated with different version information, and selects the model selected by a predetermined number or more of nodes.

In the model data provision method disclosed herein, models selected by a predetermined number of nodes or more are selected based on the history of selections from each node. It becomes possible to identify which models are used by many nodes within a model, including local models. The selected models may then be given priority for distribution, or may be targeted for version upgrades and updates.

A data provision method for a model according to an embodiment of the present disclosure stores the data volume, data characteristics, or learning volume of local data used in learning in association with version information of the global model and local model, and outputs options for the global model and local model together with data indicating the data volume, data characteristics, or learning volume, and accepts a selection.

In the model data provision method disclosed herein, not only global models but even local models can be selected for use at each node. By outputting the local data with which each model was trained, an appropriate model for the node can be selected.

A model data provision method according to an embodiment of the present disclosure calculates evaluation values for the global model and the local model during the repeated process of distributing data for the global model, learning a local model, and updating the data for the global model based on the data for the local model, and selects a global model to be distributed and updated from the global model or the local model based on the evaluation values.

In the model data provision method disclosed herein, an appropriate model is selected from among models including not only global models but also local models based on the evaluation value. The data of the selected model may be output so as to be selected preferentially, or may be redistributed to each node as global model data for future updates.

According to the model data provision method disclosed herein, not only global models but also local models trained on other nodes can be used if they produce highly accurate results for the node.

FIG. 1 is a schematic diagram of a model providing system according to a first embodiment. FIG. 2 is a block diagram showing a configuration of a node. FIG. 2 is a block diagram showing a configuration of a server. 11 is a flowchart illustrating an example of a model learning processing procedure in the model providing system. 13 is a flowchart illustrating an example of a learning process procedure for a local model based on a distributed global model. 13 is a flowchart illustrating an example of a global model update process in the server. 13 is a flowchart illustrating an example of a processing procedure for selecting a model. 13 shows an example of a model list screen. 1 shows an example of an individual screen of a model. 13 is a flowchart illustrating an example of a model utilization procedure in a node. 13 is a flowchart illustrating an example of a processing procedure for selecting a model according to the second embodiment. 13 is a flowchart illustrating an example of a learning process procedure in the model providing system of the third embodiment. 13 is a flowchart illustrating an example of a procedure for updating a local model according to the third embodiment. 13 is a flowchart illustrating an example of a processing procedure for model selection. 13 shows an example of a model list screen according to the third embodiment. 13 is a flowchart illustrating an example of an update process procedure based on model selection. 13 is a flowchart illustrating an example of an update process procedure based on model selection. 13 is a flowchart illustrating an example of an extraction process procedure based on model selection. 13 shows another example of the model list screen in the third embodiment. FIG. 13 is a schematic diagram of a model providing system according to a fourth embodiment. 23 is a flowchart illustrating an example of a procedure for updating a local model according to the fourth embodiment. 23 shows an example of a model list screen in the fourth embodiment.

This disclosure will be specifically described with reference to drawings showing embodiments thereof.

First Embodiment
1 is a schematic diagram of a model providing system 100 according to a first embodiment. The model providing system 100 includes one or more nodes 1 provided for a storage device 2 that stores data, a server 3, and a communication network N that connects the nodes 1 and the server 3 for communication.

The storage device 2 is capable of inputting and outputting data between devices that input and output data to be learned, such as sensors that measure physical quantities and cameras that take images, and stores this data. The storage device 2 may be connected to a computer for a specific purpose that outputs data according to data input by operation. The storage device 2 may be a storage device of an information terminal used by a user. The storage device 2 may be a storage device used in a server device that collects data from client devices.

When node 1 is input with data of the same type as the data stored in storage device 2, node 1 executes deep learning of the model so as to output a recognition result, a judgment result, or new data based on the data. Server 3 is a computer that provides a model to node 1, and also realizes model providing system 100 that learns the model in cooperation with node 1. In order to learn the model using the data stored in storage device 2 (hereinafter referred to as local data) as training data, it is necessary to be able to access these. In this embodiment, model providing system 100 can proceed with learning in a state where access to the local data from server 3 is not possible.

The server 3 initially obtains a zeroth-order global model 51. The server 3 distributes the zeroth-order global model 51 to the node 1 via the communication network N. The entity (data) of the global model 51 distributed from the server 3 to the node 1 is only the trained parameters, or both the trained parameters and the program. The global model 51 may be definition data (including the network definition, loss, and pre-set hyperparameters) that defines the configuration of the model, and parameters such as the weight coefficients of the learning target.

The model to be trained can have any architecture as long as it is the subject of what is called deep learning. The type of deep learning model should be appropriately selected depending on the contents of the input data and output data. The model to be trained described below may be a classification system, detection system, or generation system using a CNN (Convolutional Neural Network) that includes a convolutional layer, or it may be a RNN (Recurrent Neural Network) that learns by taking into account time series elements.

The communication network N includes a public communication network, such as the Internet, and a carrier network. The communication network N may be a dedicated line for the model providing system 100.

Node 1 can access local data stored in the storage device 2 based on the local network LN between node 1 and storage device 2. Node 1 performs deep learning using the accessible local data. It is preferable that the local data has already been annotated by an operator at the location where node 1 is installed. Node 1 obtains the 0th global model 51 distributed from server 3. Node 1 proceeds with learning based on the 0th global model 51 and the local data as training data, and obtains the 1st local model 52.

Node 1 transmits the first local model 52 to server 3. Because local data is not transmitted to server 3, processing such as abstraction and anonymization of local data is not required.

The server 3 receives multiple first local models 52 from each of the multiple nodes 1, and performs statistical processing on the multiple received first local models 52 to create a first global model 51. The server 3 redistributes the first global model 51 to the multiple nodes 1. The redistributed global model 51 may be only the weight coefficient. The redistributed weight coefficient may be the learning target or may be the entire model. The redistributed global model 51 may correspond to the difference from the previous update.

The model provision system 100 repeats the following steps: distributing an nth-order global model 51 from the server 3 to node 1, learning the nth-order global model 51 using local data at node 1, transmitting the (n+1)th-order local model 52 obtained by learning to the server 3, collecting the (n+1)th-order local model 52 at the server 3, and creating (updating) the (n+1)th-order global model 51.

This enables distributed learning without allowing access to local data from the server 3. In the model providing system 100, in the learning process in which the distribution of the global model 51 and the learning of the local model 52 are repeated, the global model 51 is updated from multiple models including the local model 52 learned at the node 1. In addition, it is possible to select a model to be used at the node 1 or an information processing device 4 outside the system from among the local model 52 learned at the node 1 and the derived global model 51.

The configuration of the model provision system 100 that realizes this model provision method will be described in detail below.

Figure 2 is a block diagram showing the configuration of node 1. Node 1 is a personal computer or a server computer. Node 1 includes a processing unit 10, a memory unit 11, a communication unit 12, a display unit 13, and an operation unit 14.

The processing unit 10 is a processor that uses a CPU (Central Processing Unit) and/or a GPU (Graphics Processing Unit). Based on the node program 1P stored in the memory unit 11, the processing unit 10 executes processing including reading data from the storage device 2, sending and receiving models to and from the server 3, and model learning.

The storage unit 11 uses a non-volatile memory such as a hard disk, a flash memory, or an SSD (Solid State Drive). The storage unit 11 stores data referenced by the processing unit 10. The storage unit 11 stores a node program 1P. The storage unit 11 stores a library 1L for deep learning. The node program 1P and/or the library 1L for deep learning may be a node program 8P and/or a library 8L for deep learning stored in a recording medium 8 that is read by the processing unit 10 and copied to the storage unit 11. The storage unit 11 stores a global model 51 acquired from the server 3 and a local model 52 that is learned using local data.

The communication unit 12 realizes data communication via the communication network N and communication with the storage device 2 via the local network LN. Specifically, the communication unit 12 is, for example, a network card. The processing unit 10 reads data from the storage device 2 via the communication unit 12 and transmits and receives data to and from the server 3.

The display unit 13 is a display such as a liquid crystal display or an organic EL (Electro Luminescence) display. The display unit 13 displays a screen including information based on data stored in the memory unit 11 or data provided from the server 3. The display unit 13 may be a display with a built-in touch panel.

The operation unit 14 is a user interface such as a keyboard and a pointing device that can input and output data to and from the processing unit 10. The operation unit 14 may be a voice input unit. The operation unit 14 may be a touch panel of the display unit 13. The operation unit 14 may be physical buttons. The operation unit 14 notifies the processing unit 10 of operation data by the operator of the node 1.

FIG. 3 is a block diagram showing the configuration of server 3. Server 3 is a server computer. Server 3 includes a processing unit 30, a storage unit 31, and a communication unit 32. In the following description, server 3 is described as being configured as one server computer, but it may also be configured in a manner in which multiple server computers are communicatively connected via a network for distributed processing. Server 3 may be a cloud type that can be communicatively connected from each node 1 via a communication network N, or an on-premise type that is communicatively connected to each node 1 via a virtual private network.

The processing unit 30 is a processor using a CPU and/or a GPU. The processing unit 30 executes the learning process of the global model 51 based on the server program 3P stored in the memory unit 31.

The storage unit 31 uses a non-volatile memory such as a hard disk or SSD. The storage unit 31 stores data referenced by the processing unit 30. The storage unit 31 stores a server program 3P. The server program 3P includes a web server program. The storage unit 31 stores a global model 51 and a local model 52 transmitted from multiple nodes 1. The server program 3P may be a server program 9P stored in a recording medium 9 that is read by the processing unit 30 and copied to the storage unit 31.

The communication unit 32 realizes data communication via the communication network N. Specifically, the communication unit 32 is, for example, a network card. The processing unit 30 transmits and receives data between multiple nodes 1 via the communication unit 32.

The learning process procedure in the model providing system 100 configured in this manner will be described. Figure 4 is a flowchart showing an example of the model learning process procedure in the model providing system 100.

The server 3 acquires an initial (0th order) global model 51 that has been prepared in advance (step S1). The initial global model 51 may be a model created as a 0th order model at a specific node 1, or may be a model learned at a specific location other than node 1, and may be stored in advance in the storage unit 31. The acquisition in step S1 includes reading out the global model 51 that has been stored in advance in the storage unit 31.

The server 3 distributes the acquired zeroth global model 51 to node 1 (step S2).

The server 3 obtains the local model 52 to be learned at the node 1 based on the global model 51 distributed to the node 1 (step S3).

The server 3 performs statistical processing on the acquired local model 52 and updates it to the next-generation global model 51 (step S4). In step S4, the server 3 may increment and store the number of updates (rounds).

The server 3 determines whether the learning completion conditions are met (step S5).

In step S5, the server 3 determines, for example, whether the number of updates has reached a predetermined number. The server 3 may determine that learning is complete when no improvement in the accuracy of the global model 51 is observed compared to the previous update.

If it is determined that the learning completion condition is not met (S5: NO), the server 3 redistributes the updated global model 51 to multiple nodes 1 (step S6) and returns the process to step S3.

In step S6, the server 3 may redistribute only parameters such as weighting coefficients, rather than redistributing the global model 51 as is.

In step S6, the server 3 may also transmit data indicating the number of updates to the global model 51, i.e., the "n" in the nth global model 51.

If it is determined that the learning completion condition is met (S5: YES), the server 3 stores the updated global model 51 as a global model 51 that can be distributed to nodes other than the node 1 in association with the version information (step S7), and ends the process. The version information corresponds to the number of times that learning of the local model 52 and updating of the global model 51 were performed until it was determined in step S5 that the learning completion condition is met. The version information may include round information as minor information.

The server 3 executes the processing procedure shown in the flowchart in Figure 4 at regular intervals, for example once a month. Each time, the version of the global model is upgraded, making it a more practical model.

Figure 5 is a flowchart showing an example of a learning process procedure for a local model 52 based on a distributed global model 51. The process shown in the flowchart in Figure 5 is executed by each of the multiple nodes 1 when the server 3 distributes the global model 51 in step S2 or step S6.

The processing unit 10 of node 1 receives the distributed global model 51 and stores it in the memory unit 11 (step S301).

The processing unit 10 of node 1 loads the stored global model 51 as an instance (step S302). The processing unit 10 acquires the local data stored in the storage device 2 as training data (step S303) and provides this to the global model 51 to perform learning (step S304).

In step S304, the processing unit 10 inputs the input data of the local data to the loaded global model 51. The processing unit 10 calculates the loss for the output data and the result data corresponding to the input data included in the local data. The processing unit 10 learns parameters including weight coefficients in the distributed global model 51 based on the calculated loss.

The processing unit 10 of node 1 determines whether the learning completion condition is met (step S305). In step S305, the processing unit 10 sets the learning completion condition to be that the number of learning (updates) meets a predetermined number (one or more). The processing unit 10 may determine that the learning completion condition is met when the output accuracy of the global model 51 after learning is equal to or greater than a stored predetermined value. The processing unit 10 may determine that the learning completion condition is met when the change in output accuracy falls within a predetermined range and can be determined to have converged.

If it is determined that the learning completion condition is not met (S305: NO), the processing unit 10 returns the process to step S304). This allows learning to continue.

If it is determined that the learning completion condition is met (S305: YES), the processing unit 10 ends the learning, and stores the global model 51 with updated parameters as a local model 52 in association with the version information (step S306). Here, the version information is assumed to match the version information of the global model 51. The version information may be date and time. The latest local model 52 may be overwritten and stored for one version, or a local model 52 may be stored for each different round.

The processing unit 10 of the node 1 transmits the stored local model 52 to the server 3 (step S307) and ends the process. This enables the server 3 to obtain the local model 52 trained with the local data from each of the multiple nodes 1.

In step S307, the processing unit 10 may also transmit data indicating "n" (the number of updates in one upgrade) indicating whether the local model 52 is the nth order or whether the original global model 51 is the nth order. In step S307, if the node 1 obtains round information corresponding to the number of updates until one learning is completed, as described below, the node 1 may transmit the round information in association with the local model 52.

It should be noted that the process shown in the flowchart of Figure 5 does not result in local data being sent from node 1 to server 3. No anonymization of the local data is performed. The data sent from node 1 is the model itself. The characteristics of the local data are reflected, but no data is sent.

Figure 6 is a flowchart showing an example of a process for updating the global model 51 in the server 3. The process procedure shown in the flowchart in Figure 6 corresponds to the details of step S4 in the process procedure shown in the flowchart in Figure 4.

The processing unit 30 of the server 3 acquires the local model 52 transmitted from the node 1 in association with the version information (step S401), and stores the local model 52 in association with the identification data and version information of the node 1 (step S402). In step S401, the processing unit 30 of the server 3 acquires the local model 52 transmitted asynchronously from each node 1. In step S402, the processing unit 30 of the server 3 stores the latest local model 52 from each node 1 until it is determined in step S5 of the flowchart in FIG. 4 that the learning completion condition is met. When round information is associated with each round, the processing unit 30 may store the local model 52 for each round until learning is completed.

The processing unit 30 determines whether the global model 51 should be updated with the acquired local model 52 (step S403). In step S403, the processing unit 30 may determine that the global model 51 should be updated if local models 52 can be acquired from all of the nodes 1 to which the global model 51 has been distributed. In step S403, the processing unit 30 may determine that the global model 51 should be updated if local models 52 can be acquired from a number of representative nodes 1 that have been determined in advance.

If it is determined that an update is not necessary (S403: NO), the processing unit 30 of the server 3 returns the process to step S401. The processing unit 30 acquires and aggregates the local models 52 sent from each node 1 until it is determined that an update is necessary.

If it is determined that an update is necessary (S403: YES), the processing unit 30 of the server 3 calculates the average of the local models 52 from the multiple nodes 1 (step S404). The processing unit 30 updates the average as a new global model 51 (step S405). In step S404, the processing unit 30 calculates, for example, the average of each weighting coefficient included in the local model 52.

The processing unit 30 stores the updated global model 51 in association with round information, for example, data "n" indicating the number of updates (nth) (step S406), and ends the update process for the global model 51. As a result, the (n-1)th global model 51 is updated to the nth global model 51.

In step S406, the processing unit 30 stores the global model 51 of each generation in the memory unit 31 so that it can be read back by going back in time by the number of updates.

As described above, in the model providing system 100 of the first embodiment, the local data is not sent to the server 3, and the global model 51 is put into practical use based on the learning results using the local data stored in each location. Even in cases where the amount of data is insufficient using only the local data from each location, it is possible to provide a model that is more accurate and can be put into practical use sooner than learning using a large amount of data aggregated in one location.

In the first embodiment, the server 3 stores the global model 51 and the local model 52 taken from the node 1 in association with round information and version information during the process of learning (updating) and revising (upgrading) the global model 51. The server 3 accepts a selection from the global model 51 and the local model 52 of each version for the node 1 for each version and each round.

Figure 7 is a flowchart showing an example of a processing procedure for selecting a model. The server 3 accepts a selection from among the stored global models 51 and local models 52 in response to a request from the node 1, by the following processing procedure based on the server program 3P. The node 1 accepts a selection operation from the operator and acquires the model to be used in the node 1, by the following processing procedure based on the node program 1P.

The operator of node 1 attempts to access a web page provided by server 3. The processing unit 10 of node 1 sends a request to server 3 based on a web browser program included in node program 1P (step S201). Hereinafter, processing based on the web browser program may be executed in the background of the web browser.

The server 3 receives the request (step S601). The processing unit 30 of the server 3 sends a web page including a list of global models 51 and local models 52 corresponding to the account to the node 1 according to the account information of the node 1 included in the request (step S602).

Each administrator of node 1 is issued an account for accessing server 3. The account includes an account ID, authentication information, and authority data. Server 3 stores the account data in the storage unit 31. The account for the administrator of node 1 is associated with authority to select and download global models 51 and local models 52. The global models 51 and local models 52 that can be downloaded may be set for each account.

At node 1, the processing unit 10 receives the list of global models 51 and local models 52 (step S202) and displays them on the display unit 13 (step S203). The processing unit 10 accepts the selection of one or more models from the list of global models 51 and local models 52 (step S204). The processing unit 10 transmits a designation of the selected models to the server 3 (step S205).

Server 3 receives the model specification (step S603), sends the specified model (global model 51 or local model 52) to node 1 (step S604), and ends the processing of server 3.

The node 1 acquires the specified global model 51 or local model 52 from the server 3 (step S206), and calculates an accuracy evaluation value for the received global model 51 or local model 52 using the local data stored in the storage device 2 (step S207).

The processing unit 10 of node 1 displays the calculated evaluation value (step S208) and accepts a selection of whether to use it or not (step S209). If it is selected to use it (S209: YES), the processing unit 10 stores the target model in the storage unit 11 (step S210). The stored global model 51 or local model 52 can be used by node 1 thereafter.

If it is selected not to use (S209: NO), or after S210, the processing unit 10 determines whether to acquire another model (step S211), and if it is determined not to acquire another model (S211: NO), the processing unit 10 ends the processing.

If it is determined that the information should be acquired (S211: YES), the process returns to step S203.

Of the processing steps shown in the flowchart of FIG. 7, steps S207-S211 are not essential, and the global model 51 or local model 52 specified in step S204 may be stored and used as is.

Figure 8 shows an example of a model list screen 131. The list screen 131 in Figure 8 shows an example (S203) of a page including a list of global models 51 and local models 52 sent from the server 3. The list includes text indicating the model names and version information of the global models 51 and local models 52. The version information may be divided into different rounds. The text indicating the version information includes links to individual pages and is selectable. In Figure 8, the global model 51 and the local model 52 are distinguished. The model name of the local model 52 includes data identifying the node that trained the model. For example, in the list in Figure 8, the local model 52 with the model name suffixed with "_ndA" was trained in "node A". When the text indicating the model name and version name of the local model 52 is selected, the model designation is accepted by the processing unit 10 of the node 1.

Figure 9 shows an example of an individual model screen 132. The individual model screen 132 in Figure 9 is displayed when any version name text is selected on the list screen 131 in Figure 8 and a model is specified. The individual model screen 132 in Figure 9 includes an evaluation value calculated for the specified model. The individual model screen 132 includes an interface 133 that accepts a selection of whether or not to use the model. When the interface 133 is selected, use is accepted (S209: YES), the selected model is stored in the storage unit 11, and the model is used thereafter. When the interface 134 for returning to the full screen is selected, it is determined that another model is to be acquired (S211: YES), and the display returns to the model list screen 131 in Figure 8.

As shown in the first embodiment, the model providing system 100 can provide to other nodes 1 not only the global model 51 updated by aggregating the local models 52 learned at each node 1, but also the local models 52 aggregated from other nodes 1.

The following describes the use of a global model 51 and a local model 52 selected by node 1 according to the processing procedure shown in the flowchart of Figure 7. As described in Figures 7 to 9, node 1 obtains multiple global models 51 and local models 52. For the same model "model-image2segmentation", node 1 can obtain global models 51 of different versions, "ver. 4.0" and "ver. 3.0", as well as a local model 52 with node identification data "_ndA" added to the end of the model name.

The operator of node 1 may use either model for the same input data and output data. Node 1 may use different versions of global model 51 and local model 52 and automatically take a majority vote for the output data from each.

Figure 10 is a flowchart showing an example of a model usage procedure in node 1. Based on node program 1P, node 1 transmits to server 3 a specification of a global model 51 and a local model 52 of different versions for one model (step S221).

Server 3 sends a global model 51, which is a different version of the specified model, and a local model 52 to node 1.

Node 1 receives and acquires the global model 51 and local model 52, which are different versions of the specified model (step S222).

The processing unit 10 of node 1 provides the local data stored in the storage device 2 as input data to each of the acquired different versions of the global model 51 and the local model 52, and obtains multiple output data (step S223).

The processing unit 10 compares the results of the multiple output data obtained (step S224) and identifies the result obtained from the largest amount of output data (step S225).

The processing unit 10 outputs the identified results to the display unit for display (step S226), stores them (step S227), and ends the processing.

In step S225, for example, if the model is a model that outputs a judgment result, the processing unit 10 takes a majority vote on the judgment result. If the model is a model that outputs an image segmentation result, calculations such as average, AND, and OR may be performed. When identifying the result in step S225, the processing unit 10 of node 1 may use an evaluation value that can be calculated for each of the multiple models obtained in step S222, and may use the accuracy as a weight so that output data from a model with a high evaluation value is more likely to affect the result.

Second Embodiment
In the second embodiment, the local model 51 stored in the server 3 may include one optimized in the node 1. The configuration of the model providing system 100 in the second embodiment is similar to that of the model providing system 100 in the first embodiment except for the details of the processing, and therefore the common configuration is denoted by the same reference numerals and detailed description thereof will be omitted.

Figure 11 is a flowchart showing an example of a processing procedure for model selection in the second embodiment. Among the processing procedures shown in the flowchart in Figure 11, the steps common to the processing procedures shown in the flowchart in Figure 7 of the first embodiment are given the same step numbers and detailed descriptions are omitted.

In the second embodiment, when the acquired model is selected for use (S209: YES), the node 1 performs re-learning using local data stored in the accessible storage device 2 (step S212), and then stores the model in association with the version information (S210).

The processing unit 10 of node 1 transmits the re-learned global model 51 or local model 52 to the server 3 in association with the version information (step S213), and proceeds to step S211.

The server 3 receives the re-learned global model 51 or local model 52 and the version information (step S605), and stores them together with the identification data and version information of the node 1 (step S606). After this, the re-learned global model 51 or local model 52 at the node 1 is also included in the list sent in step S602.

Third Embodiment
The model provision system 100 of the third embodiment calculates evaluation values based on local data for the global model 51 and the local model 52 at node 1, and outputs both evaluation values to node 1 so that the operator of node 1 can refer to them at the time of selection. The configuration of the model provision system 100 in the third embodiment is similar to that of the model provision system 100 of the first embodiment except for the details of the above-mentioned processing, and therefore the common configuration is denoted by the same reference numerals and detailed description will be omitted.

Figure 12 is a flowchart showing an example of a learning process procedure in the model provision system 100 of the third embodiment. Among the process procedures shown in the flowchart of Figure 12, the steps common to the process procedures shown in the flowchart of Figure 4 of the first embodiment are given the same step numbers and detailed explanations are omitted.

In the second embodiment, the server 3 updates the global model 51 based on the aggregated local model 52 (S4), and then transmits the updated global model 51 to each node 1 to have an evaluation value calculated (step S501).

Each node calculates the evaluation value given to the updated global model 51 using the local data and transmits it to the server 3.

The server 3 acquires the evaluation values transmitted from each node 1 (step S502), calculates a total evaluation value based on the acquired evaluation values (step S503), and stores the acquired evaluation values and the total evaluation value in association with the round information of the updated global model 51 (step S504).

This allows each global model 51 to store the evaluation value at node 1 for each version, making it possible to include the evaluation value in the list of global models 51.

Figure 13 is a flowchart showing an example of a procedure for updating the local model 52 in the third embodiment. Among the processing procedures shown in the flowchart of Figure 13, the steps common to the processing procedures shown in the flowchart of Figure 5 are assigned the same step numbers and detailed descriptions are omitted.

Before learning the distributed global model 51, the processing unit 10 of node 1 calculates and stores the amount of data (number of data items) of the acquired local data (step S311).

When the learning completion condition is satisfied (S305: YES), the processing unit 10 calculates an evaluation value of the local model 52 obtained after the learning process (step S312). In step S312, the processing unit 10 calculates, as the evaluation value, an average value of accuracy when local data is input, or the like. The processing unit 10 may derive a judgment result of A/B/C as the evaluation value.

The processing unit 10 transmits the amount of local data used to train the local model 52 stored in step S311, together with the trained local model 52, and the evaluation value calculated in step S312 to the server 3 (step S313), and ends the process.

This makes it possible to refer to the amount of local data that was the basis for learning the local model 52 as an index for evaluating the local model 52.

Figure 14 is a flowchart showing an example of a processing procedure for model selection. Among the processing procedures shown in the flowchart of Figure 14, the steps common to the processing procedures shown in the flowchart of Figure 7 of the first embodiment are given the same step numbers and detailed explanations are omitted.

In the third embodiment, when a request is sent from node 1 (S201), server 3 receives the request (S601). The processing unit 30 of server 3 extracts models whose evaluation values and total evaluation values are equal to or greater than a predetermined value from among the global models 51 and local models 52 corresponding to the account according to the account information of node 1 (step S612). The processing unit 30 transmits to node 1 a web page of a list including models and evaluation values for preferentially outputting the extracted models (step S613).

At node 1, the processing unit 10 receives and displays a list of global models 51 and local models 52 (S202, S203). In step S203, the processing unit 10 displays the evaluation value and total evaluation value associated with each model, along with the model name and version name of each model. The processing unit 10 accepts a selection from the list (S204) and transmits a designation of the selected model to the server 3 (S205). In step S204, the processing unit 10 can accept a selection of the global model 51 or local model 52 based on the evaluation value.

The processing unit 30 of the server 3 receives the model specification (S603), transmits the specified model itself (S604), stores the specification history for the specified global model 51 or local model 52 (step S614), and terminates the process. As a result, the history of the global model 51 or local model 52 selected by the operator at node 1 is stored in the storage unit 31.

In the model providing system 100 of the third embodiment, as shown in the flowchart of FIG. 14, the specified global model 51 or local model 52 is stored (S614). After the quantity of specified models, the number of times they have been specified, etc. are accumulated, the server 3 can select the specified global model 51 or local model 52 from a predetermined number or more of nodes 1. It can be said that voting for the global model 51 and the local model 52 becomes possible from each node 1.

Figure 15 shows an example of a model list screen 131 in the third embodiment. As shown in Figure 15, the model list screen 131 displays the total evaluation value calculated for the global model 51 in association with the global model, and the evaluation value is also output to the local model 52. As shown in Figure 15, the model list screen 131 displays the amount of data transmitted together with the local model 52 in association with the global model. An operator checking the model list screen 131 can select a model to use based on these evaluation values or the amount of data used to train the local model 52.

In the third embodiment, the calculated evaluation value or the history of the selected model is stored, and the server 3 can select one of the models from among past versions including the local model 52 based on the evaluation value or the model selection history. The server 3 may select the global model 51 and the local model 52 for which an evaluation value equal to or greater than a predetermined value has been calculated, or may select the global model 51 and the local model 52 for which the number of votes, i.e., the number of selected nodes 1, is equal to or greater than a predetermined number.

The model may be selected as follows. As shown in the flowchart of FIG. 7 of the first embodiment, the local data of each node 1 is used to calculate evaluation values for the global model 51 and the local models 52 learned by the other nodes 1, and the selected model is then transmitted to the server 3 together with the evaluation value. The server 3 may then store the model selection history (S614) together with the evaluation value, and based on the evaluation value and selection history, select one model with the highest evaluation value from among the global models 51 and local models 52 selected by a predetermined number or more of nodes 1.

The selected global model 51 or local model 52 may be preferentially displayed on the model list screen 131 thereafter. The selected global model 51 or local model 52 may then be selected as the global model 51 that distributes the model selected at each node 1 to the node 1 and continues to update it. In this case, in step S1 of the learning processing procedure shown in the flowchart of FIG. 4, the selected global model 51 or local model 52 is obtained, and thereafter, the processes of S2 to S7 are executed.

Figures 16 and 17 are flowcharts showing an example of an update processing procedure based on model selection. The flowcharts in Figures 16 and 17 show the processing procedure when the server 3 selects a model to distribute to node 1 and continue updating. Among the processing procedures shown in the flowcharts in Figures 16 and 17, steps common to the processing procedures shown in the flowchart in Figure 12 are given the same reference numerals and detailed explanations are omitted.

The processing unit 30 of the server 3 refers to the selection history of a global model 51 with the same model name but a different version, and a local model 52 corresponding to that global model 51 (step S101).

The processing unit 30 extracts global models 51 and local models 52 that have the same model name and are selected by a predetermined number or more of nodes 1 (step S102).

The processing unit 30 selects the global model 51 or local model 52 with the highest evaluation value from among the extracted global models 51 or local models 52 (step S103). Through steps S102 and S103, the global model 51 or local model 52 supported by more nodes 1 is selected.

The processing unit 30 distributes the selected global model 51 or local model 52 to each node 1 (step S104). The processing unit 30 obtains and stores an evaluation value for the distributed global model 51 or local model 52 (step S105).

For the processing of steps S104 and S105, each node 1 calculates an evaluation value and learns a local model 52 based on the distributed global model 51 or local model 52.

The processing unit 30 acquires the local models learned at each node 1 (S3) and performs statistical processing on the acquired local models to update the global model 51 (S4).

The processing unit 30 transmits the updated global model 51 to each node 1 to calculate the evaluation value (S501), and obtains the calculated evaluation value (S502).

For each of the multiple nodes 1, the processing unit 30 selects the model for which the evaluation value is calculated higher by more nodes 1 from the global model 51 (or local model 52) before and after the update (step S513). In step S513, there are nodes 1 that calculate a higher evaluation value in the updated global model 51 than before the update, and nodes 1 that calculate a higher evaluation value in the pre-update global model 51 (or local model 52) than after the update. In step S513, the processing unit 30 takes a majority vote of those nodes 1.

The processing unit 30 determines whether the selected model satisfies the learning completion condition (S5).

If it is determined that the learning completion condition is not met (S5: NO), the processing unit 30 distributes the global model 51 selected in step S513 to the node (step S516) and returns the process to step S3.

Only one of the selection of a model based on the selection history in steps S101-S103 and the majority vote based on the evaluation values calculated for the models before and after the update in step S513 may be performed.

Figure 18 is a flowchart showing an example of an extraction processing procedure based on model selection. The server 3 executes the following processing at regular intervals for the global models 51 and local models 52 stored therein that have the same model name and the corresponding local models 52. The processing unit 30 of the server 3 executes this processing, for example, in step S612 of the processing shown in the flowchart of Figure 14.

The processing unit 30 of the server 3 refers to the selection history of global models 51 with the same model name but different versions (step S701). The processing unit 30 extracts global models 51 or local models 52 that have been selected by a predetermined number or more of nodes 1 from among the global models 51 with the same model name and the corresponding local models 52 (step S702).

The processing unit 30 selects the global model 51 or local model 52 with the highest evaluation value from among the extracted global models 51 or local models 52 (step S703). Through steps S702 and S703, the global model 51 or local model 52 supported by more nodes 1 is selected.

The processing unit 30 stores the selected global model 51 or local model 52 (step S704) and ends the processing.

Figure 19 shows another example of the model list screen 131 in the third embodiment. In the model list screen 131 in Figure 19, compared to the model list screen 131 in Figure 15, the selected global model 51 or local model 52 is selected, and therefore one of each of the same models is displayed. An operator checking the model list screen 131 can select a model to use based on these selection records.

Fourth Embodiment
The model providing system 100 can be applied to deep learning models including neural networks. For example, the input data is image data, and the output data is a detection result of a subject appearing in the image of the image data, or a judgment result for the subject. In another example, the input data is cash receipts and withdrawals data, and the output data is a judgment result such as an evaluation, or a predicted value regarding a change in a company's business condition (growth or deterioration) or a predicted value regarding an economic forecast. In another example, the input data is measurement data from various sensors installed in a factory or production equipment, and the output data is data regarding production management including abnormality/normality. In another example, the input data is text data, and the output data is a judgment result or predicted data.

In the fourth embodiment, an example will be described in which the model providing system 100 is applied to learning a model that outputs diagnostic support data that supports the diagnosis of whether or not a patient is experiencing symptoms of a specific disease, based on medical data obtained about the patient at a medical facility. In the following description, the medical data is, for example, image data of a fundus photograph taken during an examination.

Figure 20 is a schematic diagram of the model providing system 100 of the fourth embodiment. The configuration of the model providing system 100 in the fourth embodiment is basically the same as that of the model providing system 100 of the first embodiment. Among the configuration of the model providing system 100 of the fourth embodiment, the configuration common to the model providing system 100 of the first embodiment is given the same reference numerals and detailed description is omitted.

In the fourth embodiment, the node 1 and the information processing device are provided in a medical facility. The node 1 and the information processing device can acquire image data obtained from an imaging device that takes fundus photographs of patients. In the medical facility where the node 1 is provided, the imaging device outputs image data to the storage device 2. The imaging device includes different types of devices.

In the fourth embodiment, the global model 51, the local model 52, and the model stored as distributable are models that are trained to output data and accuracy that assists in the diagnosis of glaucoma when image data is input. The local data used as training data is image data as input data, and image data resulting from segmenting the disc and cup portions of a fundus photograph as output data. The output data may also be a data set including the results of a doctor or technician's determination of whether or not symptoms are present. The image data that is input data for the local data is associated with device data that indicates the type of imaging device. The device data may be a model number or data that identifies the device manufacturer.

In the fourth embodiment of the model learning system 100, the server 3 acquires an initial global model (zeroth-order global model) 51 that has been created in advance with multiple different architectures at a specific medical facility. The server 3 distributes the zeroth-order global models 51 with different architectures to each of the nodes 1 of the medical facilities that cooperate in the training.

Each node 1 receives the distributed multiple 0th order global models 51, and proceeds with learning based on different 0th order global models 51, each using local data as training data, to obtain multiple first order local models 52. Node 1 transmits the first order local models 52 trained with different architectures to server 3.

The server 3 creates a first global model 51 by weighting the local models 52 acquired from each node 1 for each different architecture. Here, the server 3 calculates the weighted average of the weight coefficients of each local model 52. The server 3 redistributes the created first global model 51 to multiple nodes 1.

The server 3 repeatedly obtains the (n+1)th order local model 52 created from the distributed nth order global model 51 and updates the (n+1)th order global model 51 from the (n+1)th order local model 52 for each different architecture.

The server 3 compares the global models 51 obtained repeatedly for each different architecture, selects the global model 51 of the architecture with higher accuracy, and provides it to the node 1 and the information processing device 4 as a distributable model.

This makes it possible to learn using a large amount of data across different medical facilities without consolidating image data of test results, which is personal information itself, on the server 3. It also makes it possible for each node 1 to select a model that is useful for node 1 regarding the diagnostic assistance data.

The process of model learning by the server 3 and node 1 in the fourth embodiment is similar to the process procedure shown in the first to third embodiments. However, in the fourth embodiment, each node 1 transmits the characteristics of the local data to the server 3 when updating the local model 52.

Figure 21 is a flowchart showing an example of a procedure for updating the local model 52 in the fourth embodiment. Among the processing procedures shown in the flowchart in Figure 21, the steps common to the processing procedures shown in the flowchart in Figure 5 of the first embodiment are given the same step numbers and detailed descriptions are omitted.

Before executing learning of the distributed global model 51, the processing unit 10 of node 1 calculates and stores the amount of data (number of data items) of the acquired local data (step S321). The processing unit 10 acquires and stores device data of the imaging device that is stored in advance in the storage unit 11 or associated with the image data in the storage device 2 (step S322). In step S322, the processing unit 10 may acquire not only the type of device but also characteristics such as annotation accuracy of the local data as characteristics of the local data.

When the processing unit 10 satisfies the learning completion condition (S305: YES) and stores the learned local model 52 (S306), the processing unit 10 calculates an evaluation value of the local model 52 obtained after the learning process (step S323). In step S323, the processing unit 10 calculates, as the evaluation value, an average value of accuracy when local data is input, or the like. The processing unit 10 may derive a judgment result of A/B/C as the evaluation value.

The processing unit 10 transmits the amount of local data stored in step S321, the device data stored in step S322, and the evaluation value calculated in step S323 together with the learned local model 52 to the server 3 (step S324), and ends the process.

As a result, the amount of data, characteristics, etc. of the local data that was the basis for learning the local model 52 are stored in the server 3 in association with the local model 52. When obtaining a list of models from the server 3, the operator of node 1 can refer to these characteristics as an index for evaluating the local models 52 of other nodes 1.

Figure 22 shows an example of the model list screen 131 in the fourth embodiment. As shown in Figure 22, among the models, in addition to the evaluation value, the data characteristics of the local data used in learning are output to the local model 52, which can be referenced at the time of selection. In the example shown in Figure 22, the data characteristics include the specification data (model number) of the imaging device and the number of data items. Based on these data characteristics, the operator can select the model to be used, including the local model 52.

The embodiments disclosed above are illustrative in all respects and are not restrictive. The scope of the present invention is defined by the claims, and includes all modifications within the meaning and scope of the claims.

REFERENCE SIGNS LIST 1 Node 10 Processing section 11 Storage section 2 Storage device 3 Server 30 Processing section 31 Storage section 3P Server program 51 Global model 52 Local model

Claims (11)

Distributing global model data stored in the server to multiple nodes;
By using local data processed by the plurality of nodes, each node learns a local model based on the global model;
Obtaining the local model after training and version information of each local model from each node;
updating data of a global model based on data of a plurality of local models acquired from each of the plurality of nodes;
The acquired local model data and the updated global model data are stored in association with version information,
causing the plurality of nodes to calculate evaluation values based on accuracy of a global model and a local model associated with different version information, based on local data to be processed by each node;
Calculating a total evaluation value for each of the global model and the local model based on the evaluation values calculated for the plurality of nodes;
The acquired evaluation value and total evaluation value are stored in association with the global model and the local model;
Accepting a selection from a global model and a local model in which different version information is associated with each of the evaluation values and the total evaluation value;
distributing data of a selected model to the plurality of nodes or other devices. Distributing global model data stored in the server to multiple nodes;
By using local data processed by the plurality of nodes, each node learns a local model based on the global model;
Obtaining the local model after training and version information of the local model from each node;
updating data of a global model based on data of a plurality of local models acquired from each of the plurality of nodes;
The acquired local model data and the updated global model data are stored in association with version information,
accepting a selection of a plurality of global models and/or local models from among global models and local models associated with different version information;
Distributing data of the selected models to the nodes or other devices;
inputting data that can be processed by the plurality of nodes or other devices into each of the plurality of distributed models;
A result for the data is determined by a majority vote of a plurality of output results output from the plurality of distributed models.
How the model provides data, including processing. having each of the plurality of nodes select a model from among a global model and a local model to which different version information is associated;
The method for providing model data according to claim 1 or 2 , further comprising the step of selecting a model selected by a predetermined number of nodes or more. storing a data amount, a data characteristic, or a learning amount of the local data used in the learning in association with version information of the global model and the local model;
The method for providing model data according to claim 1 , further comprising the steps of : outputting options for the global model and the local model together with the data indicating the data amount, the data characteristics, or the learning amount, and accepting a selection. calculating evaluation values of the global model and the local model in a process of repeating distribution of data of the global model, learning of the local model, and updating of the data of the global model based on the data of the local model;
The method for providing model data according to claim 1 , further comprising the step of selecting a global model to be distributed or updated from the global models or the local models based on the evaluation value. Distributing global model data stored in the server to multiple nodes;
By using local data processed by the plurality of nodes, each node learns a local model based on the global model;
Obtaining the local model after training and version information of each local model from each node;
updating data of a global model based on data of a plurality of local models acquired from each of the plurality of nodes;
The acquired local model data and the updated global model data are stored in association with version information,
having each of the plurality of nodes select a model from among a global model and a local model to which different version information is associated;
Select a model that is selected by a predetermined number of nodes or more;
Distributing data of the selected model to the plurality of nodes or other devices.
How the model provides data, including processing. The system includes a plurality of nodes each processing local data, and a server connected to the plurality of nodes for communication;
The server distributes data of a global model to the plurality of nodes;
each of the plurality of nodes learns a local model from the global model based on the local data to be processed;
The processing unit included in the server includes:
acquiring the local model after learning and version information of each of the local models from the plurality of nodes;
updating data of a global model based on data of a local model after learning acquired from each of the plurality of nodes;
The acquired local model data and the updated global model data are stored in association with version information,
causing the plurality of nodes to calculate evaluation values based on accuracy of a global model and a local model associated with different version information, based on local data to be processed by each node;
Calculating a total evaluation value for each of the global model and the local model based on the evaluation values calculated for the plurality of nodes;
The acquired evaluation value and total evaluation value are stored in association with the global model and the local model;
Accepting a selection from a global model and a local model in which different version information is associated with each of the evaluation values and the total evaluation value;
A model data providing system that distributes data of a selected model to the plurality of nodes or other devices. The system includes a plurality of nodes each processing local data, and a server connected to the plurality of nodes for communication;
The server distributes data of a global model to the plurality of nodes;
each of the plurality of nodes learns a local model from the global model based on the local data to be processed;
The processing unit included in the server includes:
acquiring the local model after learning and version information of each of the local models from the plurality of nodes;
updating data of a global model based on data of a local model after learning acquired from each of the plurality of nodes;
The acquired local model data and the updated global model data are stored in association with version information,
accepting a selection of a plurality of global models and/or local models from among global models and local models associated with different version information;
Distributing data of the selected models to the nodes or other devices;
inputting data that can be processed by the plurality of nodes or other devices into each of the plurality of distributed models;
A result for the data is determined by a majority vote of a plurality of output results output from the plurality of distributed models.
Data provision system for the model. A communication unit that communicates with a plurality of nodes;
a processing unit for storing data of a global model;
The processing unit
Distributing the stored global model data to the plurality of nodes;
acquiring data of a local model learned based on the global model using local data processed by the plurality of nodes in association with version information ;
updating data of a global model based on data of a local model after learning acquired from each of the plurality of nodes;
The acquired local model data and the updated global model data are stored in association with version information,
causing the plurality of nodes to calculate evaluation values based on accuracy of a global model and a local model associated with different version information, based on local data to be processed by each node;
Calculating a total evaluation value for each of the global model and the local model based on the evaluation values calculated for the plurality of nodes;
The acquired evaluation value and total evaluation value are stored in association with the global model and the local model;
Accepting a selection from a global model and a local model in which different version information is associated with each of the evaluation values and the total evaluation value;
A server device that distributes data of the selected model to the plurality of nodes or other devices. A computer that can communicate with multiple nodes.
Distributing the stored global model data to the plurality of nodes;
acquiring data of each local model learned based on the global model using local data processed by the plurality of nodes in association with version information;
updating data of a global model based on data of a local model after learning acquired from each of the plurality of nodes;
The acquired local model data and the updated global model data are stored in association with version information,
causing the plurality of nodes to calculate evaluation values based on accuracy of a global model and a local model associated with different version information, based on local data to be processed by each node;
Calculating a total evaluation value for each of the global model and the local model based on the evaluation values calculated for the plurality of nodes;
The acquired evaluation value and total evaluation value are stored in association with the global model and the local model;
Accepting a selection from a global model and a local model in which different version information is associated with each of the evaluation values and the total evaluation value;
and distributing data of the selected model to the plurality of nodes or other devices. The computer that is connected to the server
repeating a process of learning a local model using the local data to be processed based on the data of a global model distributed from the server, and a process of transmitting the data of the learned local model to the server;
Acquire a plurality of global model data and local model data associated with different version information,
inputting data to be processed into a plurality of global models and/or local models corresponding to the acquired data;
A computer program that executes a process of determining a result for the data by majority vote of a plurality of output results output from the plurality of global models and/or local models .

Images