Associative learning systems, associative learning methods, and associative learning programs

Abstract

To provide a suitable model for each client terminal that participates in federated learning.SOLUTION: A federated learning system has a plurality of client terminals having a learning data set and a server, the learning data set has a data sample including a client ID, a first explanatory variable and a first objective variable, and the system executes calculation processing through the client terminals, and executes first federated learning processing that repeats first learning processing through the client terminals and first integration processing through the server. In the calculation processing, the client terminal inputs the data sample in a similarity calculation mode and calculates similarity between the data sample and a plurality of learning data sets. In the first learning processing, the client terminal learns an individual analysis model on the basis of similarity between a specific learning data set and specific similarity. In the first integration processing, the server integrates first learning results from the plurality of client terminals, and generates first integration information about an integrated individual analysis model.SELECTED DRAWING: Figure 4

Description

[Technical Field]

[0001] The present invention relates to a federated learning system, a federated learning method, and a federated learning program for performing federated learning. [Background technology]

[0002] With the advancement of digitization of healthcare information, secondary utilization of healthcare information managed by local governments, medical institutions, individuals, etc. (hereinafter referred to as "clients") is progressing. In particular, federative learning, which enables model training in a distributed environment without centralized management of information on a server, is attracting attention from the perspective of protecting personal information.

[0003] Patent Document 1 below discloses a concept of associative learning that specifies clients and measures the similarity of training data in a manner more suitable for application in a associative learning environment. The apparatus for associative learning disclosed in Patent Document 1 below receives parameterization updates related to predetermined parameterization of a neural network from multiple clients and performs associative learning of the neural network according to the similarity between the parameterization updates. [Prior art documents] [Patent Documents]

[0004] [Patent Document 1] International Publication No. 2020 / 229684 [Overview of the Initiative] [Problems that the invention aims to solve]

[0005] However, as described above in Patent Document 1, the similarity is determined based on parameters calculated from the client's data, and therefore the similarity of each client's data sample is not considered. Consequently, the device in Patent Document 1 decides whether or not to accept data to be used for integrated learning on a client-by-client basis, and cannot decide whether or not to accept data on a per-data-sample basis for each client, and therefore cannot control the magnitude of the impact on integrated learning on a per-data-sample basis for each client.

[0006] The present invention aims to provide a model suitable for each client terminal participating in federative learning. [Means for solving the problem]

[0007] A federated learning system, which is one aspect of the invention disclosed in this application, comprises a plurality of client terminals, each having a learning dataset, and a server capable of communicating with the plurality of client terminals, and performs federated learning in which each of the plurality of client terminals learns a model using the learning dataset, and the server integrates the models of each of the plurality of client terminals using the learning results, wherein the learning dataset has one or more data samples including a client ID that identifies the client terminal, a first explanatory variable, and a first objective variable, and performs a calculation process by each of the plurality of client terminals, and performs a first federated learning process in which the first learning process by each of the plurality of client terminals and the first integration process by the server are repeated until a first termination condition is met, and in the calculation process, the plurality Each client terminal inputs the data sample into a similarity calculation model that calculates the similarity between the data sample and the multiple training datasets, thereby calculating the similarity between the data sample and the multiple training datasets. In the first learning process, each of the multiple client terminals learns an individual analysis model that calculates a predicted value of the first target variable from the first explanatory variable, the first explanatory variable, the first target variable, and a specific similarity between the individual analysis model and a specific training dataset calculated by each of the multiple client terminals through the calculation process. In the first integration process, the server integrates the multiple first learning results from the multiple client terminals through the first learning process and generates first update information regarding an integrated individual analysis model that integrates the individual analysis models of each of the multiple client terminals. [Effects of the Invention]

[0008] According to a typical embodiment of the present invention, a suitable model can be provided for each client terminal participating in federative learning. Other issues, configurations, and effects not mentioned above will be clarified by the following description of the embodiments. [Brief explanation of the drawing]

[0009] [Figure 1] Figure 1 is an explanatory diagram showing an example of associative learning according to this embodiment. [Figure 2] Figure 2 is an explanatory diagram showing an example of a training dataset. [Figure 3] Figure 3 is an explanatory diagram illustrating an example of federated learning in a similarity calculation model. [Figure 4] Figure 4 is an explanatory diagram illustrating an example of associative learning in an individual analysis model. [Figure 5] Figure 5 is an explanatory diagram illustrating a specific example of the learning weight calculation shown in Figure 4. [Figure 6] Figure 6 is an explanatory diagram illustrating a specific example of the weighted model learning 411t shown in Figure 4. [Figure 7] Figure 7 is an explanatory diagram illustrating a specific example of the weighted model training 412t shown in Figure 4. [Figure 8] Figure 8 is an explanatory diagram illustrating a specific example of the weighted model training 413t shown in Figure 4. [Figure 9] Figure 9 is a sequence diagram of associative learning in an associative learning system. [Figure 10] Figure 10 is an explanatory diagram showing an example of an administration screen. [Figure 11] Figure 11 is an explanatory diagram showing an example of an information display screen. [Figure 12] Figure 12 is a block diagram showing an example of a server and client terminal hardware configuration. [Modes for carrying out the invention]

[0010] <Examples of Associative Learning> Figure 1 is an explanatory diagram illustrating an example of federated learning according to this embodiment. The federated learning system 100 has a server S and multiple client terminals C1 to C3 (three in Figure 1 as an example). When these are not distinguished, they are referred to as client terminals Ck (k=1, 2, 3, ..., K). The number of client terminals Ck is not limited to three; it can be two or more. In this example, K=3. The server S and client terminals Ck are connected via a network such as the Internet, LAN (Local Area Network), or WAN (Wide Area Network) to enable communication.

[0011] Client terminals C1 to C3 each possess training datasets D1 to D3. If not distinguished, these will be referred to as training dataset Dk. Training dataset Dk is a combination of training data (explanatory variables) and ground truth data (target variables). The training dataset Dk is prohibited from being removed from client terminal Ck or the location where client terminal Ck is installed.

[0012] The client terminal Ck is a computer that provides the training dataset Dk to the prediction model, trains it individually, and sends the training results, such as the model parameters or gradients of the trained prediction model, to the server S each time training is performed.

[0013] Server S is a computer that uses the learning results from client terminals Ck to integrate the prediction models for each client terminal Ck to generate an integrated prediction model, which it then sends to client terminal Ck. Client terminal Ck then provides the training dataset Dk to the integrated prediction model from Server S to train the prediction model. Through this repeated training, the associative learning system 100 performs associative learning.

[0014] In this embodiment, the associative learning system 100 performs two types of associative learning. One is associative learning FL1 for the similarity calculation model, and the other is associative learning FL2 for the individual analysis model.

[0015] The federated learning FL1 of the similarity calculation model is a federated learning method that generates an integrated similarity calculation model M1 by integrating the similarity calculation models from each client terminal Ck by performing the above-described federated learning with the similarity calculation model as the prediction model. The similarity calculation model is a prediction model that calculates the similarity between the learning dataset Dk (hereinafter referred to as the learning dataset Dj to distinguish it from Dk) and the i-th data sample (hereinafter referred to as data sample i) of the learning dataset Dj (hereinafter referred to as the learning dataset Dj to distinguish it from Dk), when focusing on the learning dataset Dk of the client terminal Ck (hereinafter referred to as client terminal Cj to distinguish it from Ck) among the client terminals C1 to CK.

[0016] Specifically, for example, the similarity calculation model is a model that calculates a propensity score using a client ID that uniquely identifies the client terminal Ck as the assigned variable.

[0017] From now on, we will use k for any client terminal and its training dataset, but we will use j for the client terminal being trained and its training dataset.

[0018] By generating the integrated similarity calculation model M1 prior to the federated learning FL2 of the individual analysis models, the influence of each data sample i on the federated learning in the federated learning FL2 of the individual analysis models can be adjusted.

[0019] The federated learning FL2 of the individual analysis models is a federated learning process that generates individual analysis models M21 to M23 by integrating the individual analysis models obtained for client terminal Cj from each client terminal C1 to C3.

[0020] Specifically, for example, individual analysis model M21 is a predictive model that integrates individual analysis models for client terminal C1 (j=1) from client terminals C1 to C3, individual analysis model M22 is a predictive model that integrates individual analysis models for client terminal C2 (j=2) from client terminals C1 to C3, and individual analysis model M23 is a predictive model that integrates individual analysis models for client terminal C3 (j=3) from client terminals C1 to C3.

[0021] By performing federated learning FL2 on the individual analysis models, appropriate individual analysis models M21 to M23 are generated for each client terminal C.

[0022] <Training dataset Dk> Figure 2 is an explanatory diagram showing an example of a training dataset Dk. The training dataset Dk has the following fields: Client ID (sometimes referred to as "CID" in subsequent figures) 201, Data ID 202, Explanatory variable 203 (sometimes referred to as Explanatory variable X), and Target variable 204 (sometimes referred to as Target variable y). The combination of values ​​for each field in the same row constitutes an entry that defines one data sample. Each of the training datasets D1 to D3 is, for example, a collection of data samples of patient groups for each hospital.

[0023] Client ID 201 is an identifier that uniquely identifies the client terminal Ck. The value of Client ID 201 is represented by Ck. Data ID 202 is an identifier that uniquely identifies a data sample. The value of Data ID 202 is represented by Dki, where i is a unique number for that data sample. Data ID 202 identifies, for example, a patient. Explanatory variable 203 is training data used in the federated learning FL2 of the individual analysis model and contains one or more features x1, x2, ... (if these are not distinguished, they are simply referred to as feature x). Feature x may be, for example, the height, weight, blood pressure, etc., of the patient identified by Data ID 202.

[0024] The dependent variable 204 is the ground truth data used in the federated learning FL2 of the individual analysis model. For example, dependent variable 204 indicates the presence or absence of disease in a patient identified by data ID 202. y=1 indicates the presence of disease, and y=0 indicates the absence of disease.

[0025] <Symbiotic learning FL1 for similarity calculation models> Figure 3 is an explanatory diagram showing an example of federated learning FL1 of the similarity calculation model. In federated learning FL1 of the similarity calculation model, the combination of explanatory variable 203 and target variable 204 of the training dataset Dk becomes explanatory variable 301, and client ID 201 becomes target variable 302.

[0026] Server S has a base similarity calculation model (hereinafter referred to as the base similarity calculation model) M10. The base similarity calculation model M10 may be an untrained neural network, and model parameters such as weights and biases φ k t A pre-trained neural network with the necessary settings may also be used. t is a natural number in ascending order, starting from 1, indicating the number of times the federated learning FL1 of the similarity calculation model is executed. Server S sends the base similarity calculation model M10 to client terminals C1 to C3.

[0027] Furthermore, if the client terminal Ck has an untrained neural network, the server S will use the model parameter φ k t The client terminal Ck receives the model parameter φ from the server S. k t The base similarity calculation model M10 can be constructed by setting it to an untrained neural network.

[0028] The base similarity calculation model M10 becomes the similarity calculation model M11t for the first federated learning FL1 on the client terminal Ck.

[0029] Client terminal Ck executes model learning 31kt (311t~313t) in the t-th federated learning FL1. Specifically, for example, client terminal Ck gives the explanatory variable 301 and the target variable 302 of the learning dataset D1 to the learning target similarity calculation model M1kt (M11t~M13t) and performs individual learning. When client terminal Ck updates the learning target similarity calculation model M1kt, the model parameter φ k t or its gradient gs k t The learning result 32kt (321t~323t) is sent to server S.

[0030] Server S uses the learning result 32kt to execute integrated learning 330t in the t-th federated learning FL1 and generates the next integrated similarity calculation model M1(t + 1). Specifically, for example, server S generates the integrated similarity calculation model M1(t + 1) using the integrated result obtained by integrating the learning results 32kt, and the integrated similarity calculation model M1(t + 1) or its integrated model parameter φ t+1 is sent to client terminal Ck. As a result, the learning target similarity calculation model M1kt in the next federated learning FL1 is set in client terminal Ck.

[0031] In this way, federated learning FL1 is repeatedly executed. When the number of execution times t reaches a predetermined threshold T1, or when the accuracy of the integrated similarity calculation model M1t reaches the target accuracy, server S terminates federated learning FL1, outputs the latest integrated similarity calculation model M1(t + 1) as the integrated similarity calculation model M1, and sends it to client terminals C1~C3.

[0032] The following defines the mathematical formulas used in the federated learning FL1 of the similarity calculation model.

[0033] [p j t =h(X,y;φ j t )···(1)

[0034] ​The above equation (1) is a calculation formula that defines the similarity calculation model and is executed by the model training 31jt of the client terminal Cj. Function h is a combination of explanatory variables 301, which is the combination of explanatory variables 203(X) and the target variable 204(y), and the model parameter φ j t This is the learning target similarity calculation model M1jt (M11t~M13t) defined by [p j ] t This is the predicted probability indicating which training dataset Dk each data sample i of the training dataset Dj on client terminal Cj is similar to in the t-th associative learning FL1.

[0035] The total number of data samples i in the training dataset Dj is N. j Then, the predicted probability [p j ] t is N j It is a matrix of size ×K. That is, the row vector [p ji ] t These are the predicted probabilities showing the similarity between data sample i of training dataset Dj and training dataset D1, the predicted probabilities showing the similarity between data sample i of training dataset Dj and training dataset D2, the predicted probabilities showing the similarity between data sample i of training dataset Dj and training dataset D3, ..., and the predicted probabilities showing the similarity between data sample i of training dataset Dj and training dataset DK.

[0036] Also, the predicted probability [p j ] t The kth column of the matrix N j A column vector [p j k ] t This represents the predicted probability of similarity between the first data sample i (i=1) of the training dataset Dj and the training dataset Dk, the predicted probability of similarity between the second data sample i (i=2) of the training dataset Dj and the training dataset Dk, the predicted probability of similarity between the third data sample i (i=3) of the training dataset Dj and the training dataset Dk, ..., N of the training dataset Dj.j i=N j This is the predicted probability showing the similarity between the data sample i and the training dataset Dk.

[0037]

number

[0038] Equation (2) above is the loss function H(φ) calculated by the model training 31jt of client terminal Cj in the t-th associative learning FL1. j t ) is p ji This is the similarity of data sample i from the training dataset Dj to the training datasets D1-D3. ji For example, it has a range of 0.0 to 1.0, with larger values ​​indicating greater similarity.

[0039] If j=1, then p ji =(1.0,0,0), indicating that data sample i is a data sample from the training dataset D1. If j=2, then p ji =(0,1.0,0), indicating that data sample i is a data sample from the training dataset D2. If j=3, then p ji =(1.0,0,0), indicating that data sample i is a data sample from the training dataset D3.

[0040] [p ji ] t This is the predicted probability [p] in the tth associative learning LF1. j ] t The row vectors in the matrix shown represent the predicted probabilities indicating how similar data sample i is to the training datasets D1-D3.

[0041] N j is the total number of data samples i in the training dataset Dj. The function loss is the error function of data sample i, and the average of the error function loss for each data sample i is the loss function H(φ jt )

[0042]

number

[0043] Equation (3) above represents the model parameter φ j t gradient gs j t This is a formula that defines the gradient g j t The result 32jt is sent to server S.

[0044]

number

[0045] Equation (4) above represents the integrated model parameter φ j t Integrated model parameter φ j t+1 This is a calculation formula to update the gradient gs, and is executed by the integrated learning 330t of server S. Server S receives the gradient gs from client terminal Cj. j t The learning result 32jt is received, and equation (4) above is executed. Note that the first term on the right-hand side of equation (4) above is the integrated model parameter φ t These are the integrated model parameters calculated as a result of executing equation (4) above in the previous associative learning FL1.

[0046]

number

[0047] Equation (5) above represents the model parameter φ j t Model parameter φ jt+1 This is a calculation formula for updating the model parameters, and is executed by the integrated learning 330t of server S. It is also executed by the model learning 31jt of client terminal Cj. When applying the above formula (5), client terminal Cj will have the updated model parameters φ j t+1 The result 32jt is sent to server S.

[0048] Equation (6) above is the updated model parameter φ from equation (5) above. j t+1 Using the integrated model parameter φ t+1 This is a calculation formula that is performed by the integrated learning 330t of server S. N is the total number of data samples i of client terminals C1 to CJ.

[0049] The associative learning system 100 uses either the update method according to equation (4) or the update methods according to equations (5) and (6) to update the integrated model parameter φ t+1 The server S calculates the integrated similarity calculation model M1t (or the base similarity calculation model M10 when t=0) using the integrated model parameter φ. t+1 Update and generate the integrated similarity calculation model M1(t+1).

[0050] Server S is the integrated similarity calculation model M1(t+1) or its integrated model parameters φ t+1 By sending this to client terminals C1-C3, the above equation (1) is updated.

[0051] [p j ] t+1 =h(X,y;φ t+1 )···(7)

[0052] Equation (6) above is a calculation formula that defines the integrated similarity calculation model M1(t+1), which is generated by the integrated learning 330 of server S, sent to client terminal Ck, and becomes the learning target similarity calculation model M1kt in the next federated learning FL1.

[0053] [p j]=h(X,y;φ)···(8)

[0054] Equation (8) above is the formula that defines the integrated similarity calculation model M1 when associative learning FL1 is completed. Since it is the same formula as equation (7) above at the end of associative learning FL1, "t+1" has been removed from equation (7).

[0055] <Associative Learning FL2 for Individual Analysis Models> Figure 4 is an explanatory diagram showing an example of federated learning FL2 of individual analysis models. Client terminal C1 performs the learning weight calculation 401. Specifically, for example, client terminal Cj inputs the explanatory variables 301 of the learning dataset Dj into the integrated similarity calculation model M1 (equation (8) above) generated in Figure 3, thereby predicting the probability [p] that each data sample i of the learning dataset Dj is similar to which learning dataset Dk. j The client terminal Cj calculates the calculated predicted probability [p j Using ], the learning weight w j Calculate.

[0056] Server S has a base individual analysis model (hereinafter referred to as the base individual analysis model) M20. The base individual analysis model M20 may be an untrained neural network, and model parameters such as weights and biases θ t A pre-trained neural network with the necessary settings may also be used. Server S sends the base individual analysis model M20 to client terminal Ck.

[0057] Furthermore, if the client terminal Ck has an untrained neural network, the server S will use the model parameter θ. t The client terminal Ck receives the model parameter θ from the server S. t The base individual analysis model M20 can be constructed by setting it to an untrained neural network.

[0058] The base individual analysis model M20 becomes the learning target individual analysis model M2j-kt in the first federated learning FL2 on the client terminal Ck. t is a natural number starting from 1 in ascending order indicating the number of executions of the federated learning FL2 of the individual analysis model.

[0059] When j = 1, the client terminal C1 holds the learning target individual analysis models M21-1t to M21-3t. When j = 2, the client terminal C2 holds the learning target individual analysis models M22-1t to M22-3t. When j = 3, the client terminal C3 holds the learning target individual analysis models M23-1t to M23-3t.

[0060] The client terminal Cj performs model learning 41jt by weighting. Specifically, for example, the client terminal Cj inputs the explanatory variable 203 of the learning dataset Dj into each of the learning target individual analysis models M2j-kt, and for each learning target individual analysis model M2j-kt, calculates the predicted value [y j k of the target variable 204. The client terminal Cj calculates the loss function F(θ j with the learning weight w, the target variable y, and the predicted value [y j k and the model parameter θ j kt for each learning target individual analysis model M2j-kt. j kt )

[0061] Then, the client terminal Cj calculates the gradient ga j kt of the model parameter θ j [[ID=3l]] kt as the individual analysis model learning result 42j-kt, and updates the model parameter θ j kt to the model parameter θ j k(t+1) . The model parameter θ j k(t+1) is also included in the individual analysis model learning result 42j-kt.

[0062] When j=1, client terminal C1 generates the individual analysis model training results 421-1t to 421-3t. The individual analysis model training result 421-1t is the model parameter θ1 updated in the training individual analysis model M21-1t. 1t or its gradient ga1 1t The individual analysis model learning result 421-2t is obtained by updating the model parameter θ1 in the individual analysis model M21-2t that was the target of the learning. 2t or its gradient ga1 2t The individual analysis model learning result 421-3t is obtained by updating the model parameter θ1 in the individual analysis model M21-3t that was the target of the learning. 3t or its gradient ga1 3t That is the case.

[0063] When j=2, client terminal C2 generates the individual analysis model training results 422-1t to 422-3t. The individual analysis model training result 422-1t is the model parameter θ2 updated in the training individual analysis model M22-1t. 1t or its gradient ga2 1t The individual analysis model learning result 422-2t is obtained by updating the model parameter θ2 in the individual analysis model M22-2t that was the target of the learning. 2t or its gradient ga2 2t The individual analysis model learning result 422-3t is obtained by updating the model parameter θ2 in the individual analysis model M22-3t that was the target of the learning. 3t or its gradient ga2 3t That is the case.

[0064] When j=3, client terminal C3 generates the individual analysis model training results 423-1t to 423-3t. The individual analysis model training result 423-1t is the model parameter θ3 updated in the training individual analysis model M23-1t. 1t or its gradient ga3 1t The individual analysis model learning result 423-2t is obtained by updating the model parameter θ3 in the individual analysis model M23-2t that was the target of the learning. 2t or its gradient ga3 2tThe individual analysis model learning result 423-3t is obtained by updating the model parameter θ3 in the individual analysis model M23-3t that was the target of the learning. 3t or its gradient ga3 3t That is the case.

[0065] Server S receives individual analysis model training results 42j-kt from client terminal Ck. Specifically, for example, Server S receives individual analysis model training results 421-1t~421-3t from client terminal C1, individual analysis model training results 422-1t~422-3t from client terminal C2, and individual analysis model training results 423-1t~423-3t from client terminal C3.

[0066] Server S performs integrated learning 430t in the t-th associative learning FL2 and generates the individual analysis model M2k(t+1) for the next iteration. Specifically, for example, Server S integrates the individual analysis model learning results 421-1t, 422-1t, and 423-1t to generate individual analysis model M21(t+1). Server S also integrates the individual analysis model learning results 421-2t, 422-2t, and 423-2t to generate individual analysis model M22(t+1). Furthermore, Server S integrates the individual analysis model learning results 421-3t, 422-3t, and 423-3t to generate individual analysis model M23(t+1).

[0067] Server S is the individual analysis model M2j(t+1) or its integrated model parameter θ j t+1 This is sent to the client terminal Ck. As a result, the client terminal Ck is configured with the individual analysis model M2j-kt, which will be the target of learning in the next associative learning FL2.

[0068] Specifically, for example, server S is the individual analysis model M21(t+1) or its integrated model parameter θ1 t+1This information is sent to client terminals C1 to C3. Using this received information, client terminal C1 sets up the individual analysis model M21-1t for the next associative learning FL2 session, client terminal C2 sets up the individual analysis model M22-1t for the next associative learning FL2 session, and client terminal C3 sets up the individual analysis model M23-1t for the next associative learning FL2 session.

[0069] Furthermore, Server S is the individual analysis model M22(t+1) or its integrated model parameter θ2 t+1 This information is sent to client terminals C1 to C3. Using this received information, client terminal C1 sets up the individual analysis model M21-2t for the next associative learning FL2 session, client terminal C2 sets up the individual analysis model M22-2t for the next associative learning FL2 session, and client terminal C3 sets up the individual analysis model M23-2t for the next associative learning FL2 session.

[0070] Furthermore, Server S is the individual analysis model M23(t+1) or its integrated model parameter θ3 t+1 This information is sent to client terminals C1 to C3. Using this received information, client terminal C1 sets up the individual analysis models M21-3t for the next associative learning FL2 session, client terminal C2 sets up the individual analysis models M22-3t for the next associative learning FL2 session, and client terminal C3 sets up the individual analysis models M23-3t for the next associative learning FL2 session.

[0071] In this way, federated learning FL2 is repeatedly executed. When the number of executions t reaches a predetermined threshold T2, or when the accuracy of all individual analysis models M21(t+1) to M2K(t+1) reaches the target accuracy, the server S terminates federated learning FL2, outputs the latest individual analysis model M2k(t+1) as individual analysis model M2k, and sends it to the client terminal Ck.

[0072] Specifically, for example, server S outputs individual analysis model M21(t+1) as individual analysis model M21 and sends it to client terminal C1. Server S also outputs individual analysis model M22(t+1) as individual analysis model M22 and sends it to client terminal C2. Server S also outputs individual analysis model M23(t+1) as individual analysis model M23 and sends it to client terminal C3.

[0073] <Calculation of learning weights: 40k> Figure 5 is an explanatory diagram illustrating a specific example of the learning weight calculation 40k (401-403) shown in Figure 4. In Figure 5, the learning dataset D1 has data samples D11 and D12, the learning dataset D2 has data samples D21 and D22, and the learning dataset D3 has data samples D31 and D32.

[0074] The client terminal Cj inputs the explanatory variable 301 of the training dataset Dj into the integrated similarity calculation model M1 and predicts the probability [p j Calculate the predicted probability [p j The column with CID=k in ] represents the predicted probability [p], which indicates how similar all data samples i in the training dataset Dj are to the training dataset Dk. j k This is a column vector representing ]. Also, the predicted probability [p ji ] is a row vector that shows the predicted probability of how similar a given data sample i in the training dataset Dj is to each training dataset Dk.

[0075] Specifically, for example, when j=1, client terminal C1 inputs explanatory variable 301 from training dataset D1 into integrated similarity calculation model M1 to calculate the predicted probability [p1]. By inputting explanatory variable 301 from data sample D11 into integrated similarity calculation model M1, the predicted probability [p 11 The following is calculated, and by inputting the explanatory variable 301 of data sample D12 into the integrated similarity calculation model M1, the predicted probability [p 12The following is calculated: The column CID=1~3) of the predicted probability [p1] shows how similar each data sample i of the training dataset D1 is to the training datasets D1~D3. 1 ], [p1 2 ], [p1 3 ]. Client terminal C1 predicts probability [p1 1 ], [p1 2 ], [p1 3 For each of ], the learning weight w1 1 w1 2 w1 3 Calculate.

[0076] Furthermore, when j=2, client terminal C2 inputs explanatory variable 301 from the training dataset D2 into the integrated similarity calculation model M1 to calculate the predicted probability [p2]. By inputting explanatory variable 301 from data sample D21 into the integrated similarity calculation model M1, the predicted probability [p 21 The following is calculated, and by inputting the explanatory variable 301 of data sample D22 into the integrated similarity calculation model M1, the predicted probability [p 22 The following is calculated: The column CID=1~3) in the predicted probability [p2] shows how similar each data sample i of the training dataset D2 is to the training datasets D1~D3. 1 ], [p2 2 ], [p2 3 ]. Client terminal C2 predicts the probability [p2 1 ], [p2 2 ], [p2 3 For each of ], the learning weight w2 1 w2 2 w2 3 Calculate.

[0077] Furthermore, when j=3, client terminal C3 inputs explanatory variable 301 from training dataset D33 into integrated similarity calculation model M1 to calculate the predicted probability [p3]. By inputting explanatory variable 301 from data sample D31 into integrated similarity calculation model M1, the predicted probability [p31 The following is calculated, and by inputting the explanatory variable 301 of data sample D32 into the integrated similarity calculation model M1, the predicted probability [p 32 The following is calculated: The column CID=1~3) of the predicted probability [p3] shows how similar each data sample i of the training dataset D3 is to the training datasets D1~D3. 1 ], [p3 2 ], [p3 3 ]. Client terminal C3 predicts probability [p3 1 ], [p3 2 ], [p3 3 For each of ], the learning weight w3 1 w3 2 w3 3 Calculate.

[0078] Below are the learning weights w in the associative learning FL2 of the individual analysis model. j k This defines the formula used in the calculation of [the value].

[0079]

number

[0080] w on the left side ji k [p] is the training weight applied to the similarity relationship between data sample i of the training dataset Dj and the training dataset Dk. ji k ] is the predicted probability indicating how similar data sample i of the training dataset Dj is to the training dataset Dk. The training weight w of all data sample i of the training dataset Dj ji k The set of these is the learning weights w j k That is the case.

[0081] <Weighted model learning 41kt> Next, we will explain a specific example of the weighted model training 41kt (411t~413t) shown in Figure 4.

[0082] Figure 6 is an explanatory diagram showing a specific example of model training 411t using the weighting method shown in Figure 4. Client terminal C1 has explanatory variables 203 and target variable 204 of the training dataset D1, and training weights w1 1 Using this, the individual analysis model M21-1t is updated, and the individual analysis model learning result 421-1t is output.

[0083] Client terminal C2 uses the explanatory variables 203 and the target variable 204 of the training dataset D2, and the training weights w2 1 Using this, the individual analysis model M22-1t is updated, and the individual analysis model learning result 422-1t is output.

[0084] Client terminal C3 uses the explanatory variables 203 and the target variable 204 of the training dataset D3, and the training weights w3 1 Using this, the individual analysis model M23-1t is updated, and the individual analysis model learning result 423-1t is output.

[0085] Server S integrates the individual analysis model learning results 421-1t, 422-1t, and 423-1t in integrated learning 430t to generate the next individual analysis model M21(t+1). Server S then generates the individual analysis model M21(t+1) or its integrated model parameter θ1 t+1 This information is sent to client terminals C1 to C3. Using this received information, client terminal C1 sets the individual analysis model M21-1t for the next associative learning FL2 session, client terminal C2 sets the individual analysis model M22-1t for the next associative learning FL2 session, and client terminal C3 sets the individual analysis model M23-1t for the next associative learning FL2 session. Individual analysis model M21 is the individual analysis model of client terminal C1 that was determined at the end of associative learning FL2.

[0086] Figure 7 is an explanatory diagram showing a specific example of model training 412t using the weighting method shown in Figure 4. Client terminal C1 receives explanatory variables 203 and target variable 204 from the training dataset D1, and training weights w1 2 Using this, the individual analysis model M21-2t is updated, and the individual analysis model learning result 421-2t is output.

[0087] Client terminal C2 uses the explanatory variables 203 and the target variable 204 of the training dataset D2, and the training weights w2 2 Using this, the individual analysis model M22-2t is updated, and the individual analysis model learning result 422-2t is output.

[0088] Client terminal C3 uses the explanatory variables 203 and the target variable 204 of the training dataset D3, and the training weights w3 2 Using this, the individual analysis model M23-2t is updated, and the individual analysis model learning result 423-2t is output.

[0089] Server S integrates the individual analysis model learning results 421-2t, 422-2t, and 423-2t in integrated learning 430t to generate the next individual analysis model M22(t+1). Server S then generates the individual analysis model M22(t+1) or its integrated model parameter θ2 t+1 This information is sent to client terminals C1 to C3. Using this received information, client terminal C1 sets the individual analysis model M21-2t for the next associative learning FL2 session, client terminal C2 sets the individual analysis model M22-2t for the next associative learning FL2 session, and client terminal C3 sets the individual analysis model M23-2t for the next associative learning FL2 session. Individual analysis model M22 is the individual analysis model of client terminal C2 that was determined at the end of associative learning FL2.

[0090] Figure 8 is an explanatory diagram showing a specific example of model training 413t using the weighting method shown in Figure 4. Client terminal C1 has explanatory variables 203 and target variable 204 of the training dataset D1, and training weights w1 3Using this, the individual analysis model M21-3t is updated, and the individual analysis model learning result 421-3t is output.

[0091] Client terminal C2 uses the explanatory variables 203 and the target variable 204 of the training dataset D2, and the training weights w2 3 Using this, the individual analysis model M22-3t is updated, and the individual analysis model learning result 422-3t is output.

[0092] Client terminal C3 uses the explanatory variables 203 and the target variable 204 of the training dataset D3, and the training weights w3 3 Using this, the individual analysis model M23-3t is updated, and the individual analysis model learning result 423-3t is output.

[0093] Server S integrates the individual analysis model learning results 421-3t, 422-3t, and 423-3t in integrated learning 430t to generate the next individual analysis model M23(t+1). Server S then generates the individual analysis model M23(t+1) or its integrated model parameter θ3. t+1 This information is sent to client terminals C1 to C3. Using this received information, client terminal C1 sets the individual analysis models M21-3t for the next associative learning FL2 session, client terminal C2 sets the individual analysis models M22-3t for the next associative learning FL2 session, and client terminal C3 sets the individual analysis models M23-3t for the next associative learning FL2 session. Individual analysis model M23 is the individual analysis model of client terminal C3 that was determined at the end of associative learning FL2.

[0094] The following defines the formulas used in the associative learning FL2 of the individual analysis model.

[0095] [y j ]=f(x j ;θ j )···(10)

[0096] Equation (10) above defines the individual analysis model M2j. xj These are the 203 explanatory variables of the training dataset Dj, and θ j [y] is the integrated model parameter of the individual analysis model M2j. j ] represents the predicted values ​​of the target variable 204 in the training dataset Dj.

[0097]

number

[0098] Equation (11) above represents the loss function F(θ) in the individual analysis model M2j-kt at the tth associative learning FL2. j kt ) defines y ji The target variable is 204 in a data sample i of the training dataset Dj, and [y ji ] is its predicted value. [y ji ] is [y j ] is. w ji k This is the learning weight calculated using equation (9) above.

[0099]

number

[0100] Equation (12) above represents the model parameter θ in the tth associative learning FL2. j kt The gradient ga j kt This defines the gradient ga j kt This is calculated on the client terminal Cj.

[0101]

number

[0102] Equation (13) above represents each model parameter θ in the tth associative learning FL2. j ktThis is the calculation formula that performs the update, and is executed in the model training 41jt with weighting of client terminal Cj. j t As shown in equation (14) above, in the t-th associative learning FL2, the P of client terminal Ck j kt It is the sum. P j kt As shown in equation (15) above, in the t-th associative learning FL2, the predicted probability at client terminal Cj [p ji k ] t It is the sum of [the numbers].

[0103] The client terminal Cj has updated model parameters θ. j 1(t+1) ~θ j 3(t+1) The individual analysis model learning results 42j-1t, 42j-2t, and 42j1-3t are sent to server S. Server S, in integrated learning 430t, sends the individual analysis model learning results 421-1t (updated model parameters θ1 1(t+1) ), Individual analysis model learning results 422-1t (updated model parameters θ2 1(t+1) ), Individual analysis model learning results 423-1t (updated model parameters θ3 1(t+1) ) can be averaged and integrated, for example, to update the integrated model parameter θ1 for client terminal C1. (t+1) This is calculated, and the individual analysis model M21(t+1) is generated.

[0104] Furthermore, in integrated learning 430t, server S obtained the individual analysis model learning results 421-2t (updated model parameters θ2) 1(t+1) ), Individual analysis model learning results 422-2t (updated model parameters θ2 2(t+1) ), Individual analysis model learning results 423-2t (updated model parameters θ3 2(t+1) ) can be averaged and integrated, for example, to update the integrated model parameter θ2 for client terminal C2. (t+1) This is calculated, and the individual analysis model M22(t+1) is generated.

[0105] Furthermore, in integrated learning 430t, server S obtained the individual analysis model learning results 421-3t (updated model parameters θ3) 1(t+1) ), Individual analysis model learning results 422-3t (updated model parameters θ2 3(t+1) ), Individual analysis model learning results 423-3t (updated model parameters θ3 3(t+1) By, for example, averaging and integrating, the updated integrated model parameter θ3 for client terminal C2 can be obtained. (t+1) This is calculated, and the individual analysis model M23(t+1) is generated.

[0106]

number

[0107] Equation (16) above calculates the gradient ga from the client terminal Cj as the learning result 42j-kt of the individual analysis model. j kt and predicted probability [p ji k ] t When a signal is received, each model parameter θ in the tth associative learning FL2 is considered. j kt This is the calculation formula that performs the update, and it is executed on server S's integrated learning 430t.

[0108] Equation (17) above calculates the gradient ga from the client terminal Cj as the learning result 42j-kt of the individual analysis model. j kt and predicted probability [p ji k ] t When a signal is received, the integrated model parameter θ is used in the tth associative learning FL2. j t This is the calculation formula that performs the update of the integrated model parameter θ1 on server S, and is executed in integrated learning 430t on server S. Specifically, for example, server S will update the integrated model parameter θ1 t+1 The individual analysis model M21(t+1) is generated using this. Server S is the updated integrated model parameter θ2 t+1The individual analysis model M22(t+1) is generated using this. Server S is the updated integrated model parameter θ3 t+1 The individual analysis model M23(t+1) is generated using this method.

[0109] <Associative Learning Sequence> Figure 9 is a sequence diagram of federative learning in the federative learning system 100. First, prior to the federative learning FL1 and FL2, the server S sends client ID 201 to each client terminal Ck (step S901). The client terminal Ck associates client ID 201 with the training dataset Dk.

[0110] Next, the associative learning system 100 performs associative learning FL1 of the similarity calculation model (step S902) and associative learning FL2 of the individual analysis model (step S903).

[0111] In the federated learning FL1 (step S902) of the similarity calculation model, the server S sends similarity calculation model information (integrated similarity calculation model M1(t+1) or its integrated model parameters φ) to the client terminal Ck. t+1 Send (step S921).

[0112] The client terminal Ck trains the target similarity calculation model using the training dataset Dj and similarity calculation model information (step S922).

[0113] The client terminal Ck sends the learning result 32jt from step S922 to the server S (step S923).

[0114] Server S performs integrated training 330t of the similarity calculation model (step S924). Server S determines whether the termination conditions for integrated training 330t of the similarity calculation model (step S924) are met (step S925).

[0115] If the termination conditions are not met, the server S sends the updated similarity calculation model information to the client terminal Ck (step S921).

[0116] If the termination conditions are met, the server S sends the updated similarity calculation model information to the client terminal Ck (step S926). As a result, the client terminal Ck sets up the integrated similarity calculation model M1 using the updated similarity calculation model information and terminates the federated learning FL1 of the similarity calculation model (step S902).

[0117] In the federated learning FL2 of the individual analysis model (step S903), the client terminal Cj performs the learning weight calculation 40j using the training dataset Dj and the integrated similarity calculation model M1 to calculate the learning weight wj (step S930).

[0118] Furthermore, server S sends individual analysis model information (individual analysis model M2j(t+1) or its integrated model parameter θ) to client terminal Ck. j t+1 Send (step S931).

[0119] The client terminal Cj performs weighted model training 41jt and sends the individual analysis model training results 42j-kt to the server S (step S933).

[0120] Server S performs integrated learning 430t of the individual analysis models using the individual analysis model learning results 42j-kt (step S934).

[0121] Server S determines whether the termination conditions for the integrated learning 430t of the individual analysis model (step S934) are met (step S925).

[0122] If the termination conditions are not met, server S sends the updated individual analysis model information to client terminal Ck (step S931).

[0123] If the termination conditions are met, the server S sends the updated individual analysis model information to the client terminal Ck (step S936). As a result, the client terminal Ck generates the individual analysis model M2j using the updated individual analysis model information and terminates the federated learning FL2 of the individual analysis model (step S903).

[0124] <Display screen example> Figure 10 is an explanatory diagram showing an example of an administration screen. Administration screen 1000 is displayed on server S. Administration screen 1000 displays the trained model list 1001, the model parameter list 1002, and the client learning contribution list 1003. The trained model list 1001 is a table data that associates client ID 201 with the ID of the individual analysis model M2j (individual analysis model ID).

[0125] Model parameter list 1002 is the model parameter θ of the individual analysis model M2j. j This is tabular data defining the values. The client learning contribution list 1003 is tabular data that associates client ID 201 with the contribution 1030 of client terminal Cj. The contribution 1030 is calculated by the following formula. The contribution 1030 is an index value that shows how much the learning dataset Dk of client terminal Ck contributes to the learning of the individual analysis model.

[0126]

number

[0127] R on the left side of equation (18) above j k The contribution of client terminal Cj is 1030. The P on the right side... j This is the P of the client terminal Ck at the end of associative learning FL2. j k This is the sum of (see equation (14) above). P j k This is the predicted probability [p] in client terminal Cj at the end of associative learning FL2. j kThis is the sum of ] (see formula (14) above). The contribution of client terminal Cj, 1030, is calculated by server S.

[0128] Figure 11 is an explanatory diagram showing an example of an information display screen. The information display screen 1100 is a screen that displays individual analysis model information for client terminal Cj, and is displayed on server S or client terminal Cj. Figure 11 shows an example where j=1.

[0129] The information display screen 1100 displays the client learning contribution list 1003, the data sample i specified from the learning dataset D1 of client terminal C1 (D11 in Figure 11 as an example), and the predicted probability distribution 1101 between the specified data sample D11 of client terminal C1 and the data sample i. The data sample i is specified by the server S or the user of client terminal C1.

[0130] The horizontal axis of the prediction probability distribution 1101 represents client ID 201, and the vertical axis represents the predicted probability between the specified data sample D11 and the training dataset Dk identified by client ID 201. The circular dots represent data sample i within the training dataset D1 of client terminal C1, and indicate the predicted probability with client terminal Ck. Note that the data sample i specified by cursor 1102 is data sample D11, and indicates the predicted probability with the training dataset D1.

[0131] <Example hardware configuration for server S and client terminal Ck> Figure 12 is a block diagram showing an example of the hardware configuration of a server S and a client terminal Ck (hereinafter referred to as computer 1200). Computer 1200 includes a processor 1201, a storage device 1202, an input device 1203, an output device 1204, and a communication interface (communication IF) 1205. The processor 1201, storage device 1202, input device 1203, output device 1204, and communication IF 1205 are connected by a bus 1206. The processor 1201 controls computer 1200. The storage device 1202 serves as the work area for the processor 1201. The storage device 1202 is also a non-temporary or temporary recording medium that stores various programs and data. Examples of storage devices 1202 include ROM (Read Only Memory), RAM (Random Access Memory), HDD (Hard Disk Drive), and flash memory. The input device 1203 receives data. Input devices 1203 include, for example, a keyboard, mouse, touch panel, numeric keypad, scanner, microphone, and sensor. Output devices 1204 output data. Output devices 1204 include, for example, a display, printer, and speaker. Communication IF 1205 connects to the network and sends and receives data.

[0132] Thus, according to this embodiment, it is possible to provide an individual analysis model M2k suitable for each client terminal Ck participating in federated learning. Furthermore, when generating the integrated similarity calculation model M1, the client terminal Ck can predict its similarity with other training datasets Dk without passing its data sample i to other client terminals Ck or the server S, thus preventing the leakage of the data sample i itself.

[0133] In the above embodiment, server S participated in federated learning FL1 and FL2, but it may also participate only in federated learning FL2 and not in federated learning FL1. In this case, client terminal Ck can obtain the results of federated learning FL1 (integrated similarity calculation model M1 or training weights wk) from the other server in which the other server participated. Conversely, server S may participate in federated learning FL1 but not in federated learning FL2. In this case, client terminal Ck obtains the results of federated learning FL1 (integrated similarity calculation model M1 or training weights wk) from server S. In this case, integrated learning 430t will be executed by the other server.

[0134] Furthermore, server S can send a program, which is to be executed by client terminal Ck, to client terminal Ck as a container, for example.

[0135] Furthermore, the federative learning system 100 according to the above-described embodiment can also be configured as shown in [1] to

[11] below.

[0136] [1] The federated learning system 100 comprises a plurality of client terminals C1 to C3, each having a learning dataset D1 to D3, and a server S capable of communicating with the plurality of client terminals C1 to C3. The system performs federated learning in which each of the plurality of client terminals C1 to C3 learns a model using the learning dataset D1 to D3, and the server S uses the learning results to integrate the models of each of the plurality of client terminals C1 to C3, repeating this process.

[0137] The training dataset Dk has one or more data samples i, which include a client ID 201 that identifies the client terminal Ck, a first explanatory variable 203, and a first objective variable 204.

[0138] The federated learning system 100 performs calculation processes (learning weight calculations 401 to 403) by each of the multiple client terminals C1 to C3, and also performs a first federated learning process (FL2) in which the first learning process (model learning by weighting 411t to 413t) by each of the multiple client terminals C1 to C3 and the first update process (integrated learning 330t) by the server S are repeated until the first termination condition is met.

[0139] In the calculation process (learning weight calculation 401-403), each of the multiple client terminals C1-C3 inputs the data sample i into a similarity calculation model M1 that calculates the similarity between the data sample i and the multiple learning datasets D1-D3, thereby calculating the similarity (predicted probability [p ji k ] t Calculate ).

[0140] In the first learning process (weighted model learning 411t~413t), each of the multiple client terminals C1~C3 calculates a specific similarity (prediction probability [p]) between the first explanatory variable 203, the first objective variable 204, and a specific learning dataset (e.g., D1) calculated by each of the multiple client terminals C1~C3 through the calculation process. ji 1 ] t Based on the above, the individual analysis models (individual analysis models to be learned M21-1t, M22-1t, M23-1t) are trained.

[0141] In the first integration process (integrated learning 330t), the server S integrates the multiple first learning results (421-1t, 422-1t, 423-1t) from the first learning process (weighted model learning 411t-413t) from the multiple client terminals C1-C3, and provides first integrated information (integrated individual analysis model M21(t+1) or integrated model parameter θ1) relating to the integrated individual analysis model M21(t+1) which is formed by integrating the individual analysis models (learning target individual analysis models M21-1t, M22-1t, M23-1t) from each of the multiple client terminals C1-C3. t+1 ) generates.

[0142] [2] In the federated learning system 100 described in [1] above, in the first integration process (integrated learning 330t), the server S continues to process the first integrated information (integrated individual analysis model M21(t+1) or integrated model parameter θ1 until the first termination condition is met. t+1 ) is sent to each of the aforementioned client terminals C1 to C3.

[0143] In the first learning process (weighted model learning 411t~413t), each of the multiple client terminals C1~C3 receives the first integrated information (integrated individual analysis model M21(t+1) or integrated model parameter θ1 t+1 The individual analysis models (learning target individual analysis models M21-1t, M22-1t, M23-1t) are updated with the first explanatory variable 203, the first dependent variable 204, and the specific similarity (predicted probability [p ji 1 ] t Based on the above, the updated individual analysis models (individual analysis models to be learned M21-1t, M22-1t, M23-1t) are trained.

[0144] [3] In the federated learning system 100 described in [1] above, in the first integration process (integrated learning 330t), if the first termination condition is met, the server S provides the first integrated information (integrated individual analysis model M21(t+1) or integrated model parameter θ1 t+1 ) is sent to a specific client terminal having the specific training dataset.

[0145] [4] In the federated learning system 100 described in [1] above, in the calculation process (learning weight calculation 401-403), each of the multiple client terminals C1-C3 inputs the combination of the first explanatory variable 203 and the first target variable 204 of the data sample i as the second explanatory variable 301 into the similarity calculation model (learning target similarity calculation model M11t-M13t), thereby calculating the similarity (predicted probability [p ji k ] t Calculate ).

[0146] [5] In the federated learning system 100 described in [1] above, the calculation process (learning weight calculation 401-403) is performed on each of the multiple client terminals C1-C3, and the specific similarity (prediction probability [p ji 1 ] t Learning weights (w1) according to ) 1 w2 1 w3 1 Calculate ).

[0147] In the first learning process (weighted model learning 411t~413t), each of the multiple client terminals C1~C3 receives the individual analysis model (learning target individual analysis model M21-1t, M22-1t, M23-1t), the first explanatory variable 203, the first objective variable 204, and a specific similarity (predicted probability [p]) calculated by each of the multiple client terminals C1~C3 through the calculation process (learning weight calculation 401~403). ji 1 ] t Learning weights (w1) according to ) 1 w2 1 w3 1 Based on the above, the individual analysis models (individual analysis models to be learned M21-1t, M22-1t, M23-1t) are trained.

[0148] [6] Prior to the first federated learning process (FL2), the federated learning system 100 described in [1] above executes a second federated learning process (FL1) in which each of the plurality of client terminals C1 to C3 performs a second learning process (model learning 311t to 313t) and the server S performs a second integration process (integrated learning 330t), and this process is repeated until the second termination condition is met.

[0149] In the second learning process (model learning 311t to 313t), each of the multiple client terminals C1 to C3 learns the learning target similarity calculation model M11t to M13t, using the combination of the first explanatory variable 203 and the first target variable 204 as the second explanatory variable 301, and the client ID 201 as the second target variable 302.

[0150] In the second integration process (integrated learning 330t), the server S integrates the second learning results 321t to 323t of the learning target similarity calculation models M11t to M13t from the multiple client terminals C1 to C3 obtained in the second learning process (model learning 311t to 313t), and provides second integration information (integrated similarity calculation model M1(t+1) or its integrated model parameters φ) relating to the integrated similarity calculation model M1(t+1) which is formed by integrating the learning target similarity calculation models M11t to M13t from the multiple client terminals C1 to C3. t+1 ) generates.

[0151] [7] In the federated learning system 100 described in [6] above, in the second integration process (integrated learning 330t), the server S provides the second integrated information (integrated similarity calculation model M1(t+1) or its integrated model parameters φ t+1 ) is sent to each of the aforementioned client terminals C1 to C3.

[0152] In the second learning process (model learning 311t~313t), each of the multiple client terminals C1~C3 receives the second integrated information (integrated similarity calculation model M1(t+1) or its integrated model parameters φ t+1The learning target similarity calculation models M11t to M13t are updated using the above method, and the updated learning target similarity calculation models M11t to M13t are trained based on the second explanatory variable 301 and the second objective variable 302.

[0153] [8] In the federated learning system 100 described in [6] above, in the second learning process (model learning 311t to 313t), each of the multiple client terminals C1 to C3 sets the updated learning target similarity calculation models M11t to M13t to the similarity calculation model M1 if the second termination condition is met.

[0154] [9] In the federative learning system 100 described in [3] above, the server S provides the first integrated information (integrated model parameter θ) at the end of the first federative learning process (FL2). j Outputs (Model parameter list 1002).

[0155]

[10] In the federative learning system 100 described in [3] above, the server S determines the specific similarity (prediction probability [p ji 1 ]) and the aforementioned similarity (predicted probability [p ji k Based on ]), a contribution score of 1030 is calculated to indicate how much the specific training dataset D1 contributes to the training of the integrated individual analysis model M21.

[0156]

[11] The federated learning system 100 comprises a plurality of client terminals C1 to C3, each having a learning dataset D1 to D3, and a server S capable of communicating with the plurality of client terminals C1 to C3. The system performs federated learning in which each of the plurality of client terminals C1 to C3 learns a model using the learning dataset D1 to D3, and the server S uses the learning results to integrate the models of each of the plurality of client terminals C1 to C3, repeating this process.

[0157] The training dataset Dk has one or more data samples i, which include a client ID 201 that identifies the client terminal Ck, a first explanatory variable 203, and a first objective variable 204.

[0158] The federated learning system 100 executes a second federated learning process (FL1) which repeats the second learning process (model learning 311t to 313t) by each of the multiple client terminals C1 to C3 and the second integration process (integrated learning 330t) by the server S until the second termination condition is met.

[0159] In the second learning process (model learning 311t to 313t), each of the multiple client terminals C1 to C3 learns the learning target similarity calculation model M11t to M13t, using the combination of the first explanatory variable 203 and the first target variable 204 as the second explanatory variable 301, and the client ID 201 as the second target variable 302.

[0160] In the second integration process (integrated learning 330t), the server S integrates the second learning results 321t to 323t of the learning target similarity calculation models M11t to M13t from the multiple client terminals C1 to C3 obtained in the second learning process (model learning 311t to 313t), and provides second integration information (integrated similarity calculation model M1(t+1) or its integrated model parameters φ) relating to the integrated similarity calculation model M1(t+1) which is formed by integrating the learning target similarity calculation models M11t to M13t from the multiple client terminals C1 to C3. t+1 ) generates.

[0161] It should be noted that the present invention is not limited to the embodiments described above, but includes various modifications and equivalent configurations within the spirit of the attached claims. For example, the embodiments described above are described in detail to make the present invention easier to understand, and the present invention is not necessarily limited to having all of the described configurations. Furthermore, some of the configurations of one embodiment may be replaced with those of another embodiment. Furthermore, some of the configurations of one embodiment may be added to those of another embodiment. Furthermore, some of the configurations of each embodiment may be added, deleted, or replaced with other configurations.

[0162] Furthermore, each of the above-described configurations, functions, processing units, processing means, etc. may be implemented in hardware by designing part or all of them, for example, by using an integrated circuit, or may be implemented in software by a processor interpreting and executing a program that realizes each function.

[0163] Information such as programs, tables, and files that realize each function can be stored in a storage device such as a memory, a hard disk, an SSD (Solid State Drive), or a recording medium such as an IC (Integrated Circuit) card, an SD card, or a DVD (Digital Versatile Disc).

[0164] Also, the control lines and information lines show those considered necessary for explanation, and do not necessarily show all the control lines and information lines required for implementation. In reality, it can be considered that almost all components are interconnected.

Explanation of Reference Numerals

[0165] 100 Federated Learning System 203 Explanatory Variable 204 Objective Variable 301 Explanatory Variable 302 Objective Variable 330t Integrated Learning 411t~413t Model Update 430t Integrated Learning 1000 Management Screen 1100 Information Display Screen C1~C3 Client Terminals D1~D3 Learning Datasets M1 Integrated Similarity Calculation Model M21~M23 Individual Analysis Models

Claims

1. A federated learning system comprising: a plurality of client terminals, each having a training dataset; and a server capable of communicating with the plurality of client terminals, wherein each of the plurality of client terminals trains a model using the training dataset, and the server integrates the models of each of the plurality of client terminals using the training results, and this process is repeated in the process of federated learning, The aforementioned training dataset has one or more data samples including a client ID that identifies the client terminal, a first explanatory variable, and a first objective variable. The calculation process is performed by each of the multiple client terminals, and the first combined learning process is performed, which is repeated by each of the multiple client terminals and the first integration process by the server until the first termination condition is met. In the calculation process described above, each of the multiple client terminals inputs the data sample into a similarity calculation model that calculates the similarity between the data sample and the multiple training datasets, thereby calculating the similarity between the data sample and the multiple training datasets. In the first learning process, each of the plurality of client terminals learns the individual analysis model based on the first explanatory variable, the first explanatory variable, the first objective variable, and a specific similarity between the first explanatory variable and a specific learning dataset calculated by each of the plurality of client terminals through the calculation process. In the first integration process, the server integrates the multiple first learning results from the multiple client terminals obtained through the first learning process and generates first integrated information relating to an integrated individual analysis model that integrates the individual analysis models of each of the multiple client terminals. An associative learning system characterized by the following:

2. The associative learning system according to claim 1, In the first integration process, the server transmits the first integration information to each of the multiple client terminals until the first termination condition is met. In the first learning process, each of the multiple client terminals updates the individual analysis model with the first integrated information and learns the updated individual analysis model based on the first explanatory variable, the first target variable, and the specific similarity. An associative learning system characterized by the following:

3. The associative learning system according to claim 1, In the first integration process, if the first termination condition is met, the server transmits the first integration information to a specific client terminal having the specific learning dataset. An associative learning system characterized by the following:

4. The associative learning system according to claim 1, In the calculation process, each of the multiple client terminals inputs the combination of the first dependent variable and the first independent variable of the data sample as a second independent variable into the similarity calculation model to calculate the similarity. An associative learning system characterized by the following:

5. The associative learning system according to claim 1, In the calculation process described above, each of the multiple client terminals calculates a learning weight corresponding to the specific similarity, In the first learning process, each of the multiple client terminals learns the individual analysis model based on the individual analysis model, the first explanatory variable, the first target variable, and the learning weights corresponding to a specific similarity calculated by each of the multiple client terminals through the calculation process. An associative learning system characterized by the following:

6. The associative learning system according to claim 1, Prior to the first federated learning process, a second federated learning process is executed, in which the second learning process by each of the plurality of client terminals and the second integration process by the server are repeated until the second termination condition is met. In the second learning process, each of the multiple client terminals learns a similarity calculation model for the target to be learned, using the combination of the first explanatory variable and the first target variable as the second explanatory variable and the client ID as the second target variable. In the second integration process, the server integrates the second learning results of the learning target similarity calculation model from the multiple client terminals obtained by the second learning process, and generates second integration information relating to the integrated similarity calculation model obtained by integrating the similarity calculation models of the multiple client terminals. An associative learning system characterized by the following:

7. The associative learning system according to claim 6, In the second integration process, the server transmits the second integration information to each of the multiple client terminals. In the second learning process, each of the multiple client terminals updates the learning target similarity calculation model with the second integrated information and learns the updated learning target similarity calculation model based on the second explanatory variable and the second target variable. An associative learning system characterized by the following:

8. The associative learning system according to claim 6, In the second learning process, if each of the multiple client terminals satisfies the second termination condition, it sets the updated learning target similarity calculation model to the similarity calculation model. An associative learning system characterized by the following:

9. The associative learning system according to claim 3, The aforementioned server, Output the first integrated information at the end of the first associative learning process. An associative learning system characterized by the following:

10. The associative learning system according to claim 3, The server calculates a contribution score, based on the specific similarity score and the similarity score, which indicates how much the specific training dataset contributes to the training of the integrated individual analysis model. An associative learning system characterized by the following:

11. A federated learning system comprising: a plurality of client terminals, each having a training dataset; and a server capable of communicating with the plurality of client terminals, wherein each of the plurality of client terminals trains a model using the training dataset, and the server integrates the models of each of the plurality of client terminals using the training results, and this process is repeated in the process of federated learning, The aforementioned training dataset has one or more data samples including a client ID that identifies the client terminal, a first explanatory variable, and a first objective variable. The second federated learning process is executed by repeating the second learning process by each of the multiple client terminals and the second integration process by the server until the second termination condition is met. In the second learning process, each of the multiple client terminals learns a similarity calculation model for the target to be learned, using the combination of the first explanatory variable and the first target variable as the first explanatory variable and the client ID as the second target variable. In the second integration process, the server integrates the second learning results of the learning target similarity calculation model from the multiple client terminals obtained by the second learning process, and generates second integration information relating to the integrated similarity calculation model obtained by integrating the learning target similarity calculation models of the multiple client terminals. An associative learning system characterized by the following:

12. A federated learning system comprising a plurality of client terminals, each having a training dataset, and a server capable of communicating with the plurality of client terminals, performs federated learning in which each of the plurality of client terminals trains a model using the training dataset, and the server integrates the models of each of the plurality of client terminals using the training results, and this process is repeated, The aforementioned training dataset has one or more data samples including a client ID that identifies the client terminal, a first explanatory variable, and a first objective variable. The calculation process is performed by each of the multiple client terminals, and the first combined learning process is performed, which is repeated by each of the multiple client terminals and the first integration process by the server until the first termination condition is met. In the calculation process described above, each of the multiple client terminals inputs the data sample into a similarity calculation model that calculates the similarity between the data sample and the multiple training datasets, thereby calculating the similarity between the data sample and the multiple training datasets. In the first learning process, each of the plurality of client terminals learns the individual analysis model based on the first explanatory variable, the first explanatory variable, the first objective variable, and a specific similarity between the first explanatory variable and a specific learning dataset calculated by each of the plurality of client terminals through the calculation process. In the first integration process, the server integrates the multiple first learning results from the multiple client terminals obtained through the first learning process and generates first integrated information relating to an integrated individual analysis model that integrates the individual analysis models of each of the multiple client terminals. A characteristic associative learning method.

13. A federated learning program to be executed by the processor of a client terminal in a federated learning system having a plurality of client terminals, each having a training dataset, and a server capable of communicating with the plurality of client terminals, wherein each of the plurality of client terminals trains a model using the training dataset, and the server integrates the models of each of the plurality of client terminals using the training results, and this process is repeated. The aforementioned training dataset has one or more data samples including a client ID that identifies the client terminal, a first explanatory variable, and a first objective variable. The aforementioned processor, A similarity calculation model that calculates the similarity between the data sample and the multiple training datasets is input to the data sample, and a calculation process is performed to calculate the similarity between the data sample and the multiple training datasets. An individual analysis model is provided that calculates a predicted value of the first target variable from the first explanatory variable. A first learning process is performed to train the individual analysis model based on the first explanatory variable, the first target variable, and a specific similarity between the first explanatory variable and a specific learning dataset calculated on each of the multiple client terminals by the calculation process. When the server integrates the multiple first learning results from the multiple client terminals through the first learning process and receives first integrated information from the server regarding the integrated individual analysis model which integrates the individual analysis models of each of the multiple client terminals, it updates the individual analysis model with the first integrated information and repeatedly executes the calculation process and the first learning process. A collaborative learning program characterized by the following features.

Images