Federated learning system, model learning device, federated learning method, model learning program

Abstract

This federated learning system comprises a plurality of model training devices, each of which is connected to at least one of the other model training devices via a network. The model training devices each comprise a mini-batch extraction unit, a model parameter update unit, a dual-variable calculation / transmission unit, a dual-variable reception unit, and a dual-variable setting unit. The model parameter update unit performs training so as to update model parameters by using dual variables, a step size, a mini-batch of model training data, a constraint parameter, and a coefficient γ based on a pre-defined optimum value η and a pre-defined hyperparameter α. For each of the other model training devices connected to the model training device, the dual-variable calculation / transmission unit calculates dual variables using the coefficient γ and the model parameters updated by the model parameter update unit, and transmits same.

Description

[Technical Field]

[0001] The present invention relates to an associative learning system having a plurality of model learning devices, each of which is connected to one of the other model learning devices via a network, a model learning device, an associative learning method, and a model learning program. [Background technology]

[0002] Utilizing emails, purchase histories, corporate research materials, IoT information, and hospital diagnostic information stored on personal devices promises numerous benefits, including matching, automatic control, and AI-based medical treatment. However, concerns about information security and leaks have prevented this technology from being fully utilized. To prevent the leakage of confidential information, such as personal information, federated learning (or distributed learning) technology analyzes data on end-user devices and enables the utilization of know-how (e.g., dual variables) without disclosing confidential information. Non-Patent Document 1 addresses the drift problem in distributed training of machine learning models by reconstructing the SVR (Stochastic Variance Reduction) gradient correction implicitly by optimally selecting the constraint-strength control parameter in the update process of Edge-Consensus Learning (ECL). Figure 1 shows Algorithm 2 described in Non-Patent Document 1. Non-Patent Document 2 reduces the risk of reproducing the original data by adding noise to the learning model (hereinafter simply referred to as the "model"). FIG. 2 shows Algorithm 1 and Equations 13a and 13b shown in Non-Patent Document 2. [Prior art documents] [Non-patent literature]

[0003] [Non-Patent Document 1] Niwa, K., Zhang, G., Kleijn, WB, Harada, N., Sawada, H., Fujino, A.. (2021). Asynchronous Decentralized Optimization With Implicit Stochastic Variance Reduction. Proceedings of the 38th International Conference on Machine Learning, in Proceedings of Machine Learning Research. [Non-patent document 2] ZonghanH, et al. “Differentially Private ADMM for Convex Distributed Learning: Improved Accuracy via Multi-Step Approximation”, arXiv:2005.07890. Summary of the Invention [Problem to be solved by the invention]

[0004] However, while federated learning allows for distributed and high-speed learning, it poses the risk of reproducing the original data using a user-generated model. It is also unclear how much information is being transmitted during model training. Furthermore, technology that imparts differential privacy to federated learning reduces the risk of reproducing the original data by adding noise to the user-generated model in federated learning. The greater the added noise, the lower the risk of data leakage, but the lower the accuracy of the trained model. With existing methods, the amount of added noise must be determined arbitrarily, leaving open the challenge of simultaneously reducing the risk of data leakage and improving accuracy.

[0005] Furthermore, when learning is performed on each user's device, the distribution of data held by each user varies. For example, the attributes (labels) of the data held by each user may differ; for example, user A may only have blood pressure information, while user B may only have heart rate information. The number of data items held may also differ; for example, user A may hold data for 10 people, while user B holds data for 3 people. Furthermore, the performance of each user's device and the communication environment may differ, so there may be large differences in the computational and communication performance of user devices. Even with the technology in Non-Patent Document 1, there is a problem in that stable federated learning is difficult to perform when the distribution of the distributed data is large and when there is a large difference in the computational and communication performance of user devices.

[0006] Federated learning has several problems. First, the present invention aims to perform stable federated learning even when the variance of distributed data is large and when there is a large difference in the computational and communication performance of users. [Means for solving the problem]

[0007] The federated learning system of the present invention includes multiple model learning devices, each of which is connected to one of the other model learning devices via a network. In the federated learning system of the present invention, the model parameters, dual variables, step size, model learning data, and constraint parameters are set to predetermined initial values. Each model learning device includes a mini-batch extraction unit, a model parameter update unit, a dual variable calculation and transmission unit, a dual variable reception unit, and a dual variable setting unit. The mini-batch extraction unit extracts a predetermined amount of data as a mini-batch from the model learning data. The model parameter update unit performs learning using the dual variables, step size, mini-batch of model learning data, constraint parameters, and a coefficient γ using a predetermined optimal value η and a predetermined high parameter α, and updates the model parameters. The dual variable calculation and transmission unit calculates and transmits dual variables for each of the other model learning devices connected to the model learning device using the model parameters and coefficient γ updated by the model parameter update unit. The dual variable reception unit receives dual variables from the other model learning devices connected to the model learning device. The dual variable setting unit sets the received dual variables as dual variables to be used in the next learning. [Effects of the Invention]

[0008] According to the federated learning system of the present invention, the model parameter update unit of the model learning device updates the model parameters using a coefficient γ that uses a predetermined optimal value η and a predetermined high parameter α. Furthermore, the dual variable calculation and transmission unit calculates and transmits dual variables for each of the other model learning devices connected to the model learning device using the updated model parameters and coefficient γ. Because the degree of updating can be adjusted using coefficient γ, stable federated learning can be performed even when the variance of the distributed and stored data is large or when there is a large difference in the computational and communication performance of users. [Brief explanation of the drawings]

[0009] [Figure 1] FIG. 1 is a diagram showing Algorithm 2 shown in Non-Patent Document 1. [Figure 2]FIG. 1 is a diagram showing Algorithm 1 and Equations 13a and 13b shown in Non-Patent Document 2. [Figure 3] FIG. 1 is a diagram showing an example of the configuration of a federated learning system according to a first embodiment. [Figure 4] FIG. 1 is a diagram showing an example in which model learning devices of the present invention are connected in a ring shape. [Figure 5] FIG. 10 is a diagram showing an example in which model learning devices of the present invention are randomly connected. [Figure 6] FIG. 2 is a diagram showing an example of a processing flow of the federated learning system according to the first embodiment. [Figure 7] FIG. 10 is a diagram showing an example of the configuration of a federated learning system according to a second embodiment. [Figure 8] FIG. 10 is a diagram showing an example of a processing flow of the federated learning system according to the second embodiment. [Figure 9] FIG. 10 is a diagram showing the algorithm of the federated learning system 20 of the second embodiment in the same manner as in Non-Patent Document 1. [Figure 10] FIG. 2 is a diagram showing an example of the functional configuration of a computer. DETAILED DESCRIPTION OF THE INVENTION

[0010] Hereinafter, embodiments of the present invention will be described in detail. Components having the same functions are given the same numbers, and duplicated explanations will be omitted. [Example]

[0011] FIG. 3 shows an example of the configuration of a federated learning system according to the first embodiment. FIG. 4 shows an example of model learning devices of the present invention connected in a ring configuration, and FIG. 5 shows an example of model learning devices of the present invention connected randomly. FIG. 6 is a diagram showing an example of the processing flow of the federated learning system according to the first embodiment. The federated learning system 10 according to the first embodiment comprises N model learning devices 1001, ..., 100 N Each model learning device 100 i Other model learning device 100 j Here, N is an integer of 2 or more, i is an integer of 1 or more and N or less, and j is an integer of 1 or more and N or less other than i. Each model learning device 100 iThe system includes an initial setting unit 110, a mini-batch extraction unit 130, a model parameter update unit 140, a dual variable calculation and transmission unit 150, a dual variable reception unit 160, and a dual variable setting unit 170.

[0012] Each model learning device 1001, ..., 100 N In the initial setting section 110, the model parameters w i , dual variable z i|j , step size μ, model training data x i , constraint parameter A i|j are set to predetermined initial values ​​(S110). For example, w1,...,w N are set to the same value, and z i|j =0.

[0013] Each model learning device 1001, ..., 100 N The mini-batch extraction unit 130 extracts the model training data x i A predetermined amount of data is divided into mini-batches ξ i r,k For example, when there are 10,000 pieces of learning data, 500 or 1,000 mini-patches can be arbitrarily extracted. The i at the bottom right of ξ indicates the number of the model learning device. The k at the top right of ξ indicates the number of each model learning device 1001, ..., 100 Nis an integer indicating the number of iterations of the iterative process (inner loop process) performed by ξ. The inner loop process is iterated K times. K is an integer greater than or equal to 2, and k is an integer greater than or equal to 1 and less than or equal to K. The r in the upper right corner of ξ is an integer indicating the number of iterations of the iterative process of the inner loop process (outer loop process) for the entire associative learning system. The outer loop process is iterated R times. R is an integer greater than or equal to 2, and r is an integer greater than or equal to 1 and less than or equal to R. In the following explanation, the i and j in the lower right corner of the symbol indicate the number of the model learning device, the r in the upper right corner of the symbol indicates the number of iterations in the outer loop process, and the k in the upper right corner indicates the number of iterations in the inner loop process. Note that in the text of the specification, the symbols in the lower right and the upper right cannot be written in the same horizontal position, so they are written offset from each other. On the other hand, in the formulas and figures in the specification, they are written in the same horizontal position because it is possible to write them.

[0014] Each model learning device 1001, ..., 100 N The model parameter update unit 140 performs learning using the dual variables, step size, mini-batch of model learning data, constraint parameters, and coefficient γ using a predetermined optimal value η and a predetermined high parameter α, and updates the model parameters (S140). Specifically, the model parameter update unit 140 updates the model parameters as shown in the following equation, and updates the model parameters w to be used in the next processing. i r,k+1 Ask for.

number

[0015] The predetermined optimum value η is 1 / (μKE i ) can be used. E i The model learning device 100 i The number of other model learning devices connected to the model learning device. The predetermined high parameter α is, for example, α = 2, but is not limited to this. Since the appropriate α differs for each federated system, it is a parameter that can be set appropriately for each federated learning system. The cost function or a function f that can be replaced with the cost function - ,Model parameters u, constraint parameters A i|j For example, the function f - As the function q i (w i ) can be used.

number

[0016] Each model learning device 1001, ..., 100 N The dual variable calculation and transmission unit 150 of the model learning device 100 i and other model learning devices 100 connected to it. j The model parameter update unit 140 updates the model parameters w i r,k+1 and the coefficient γ, the dual variable y i|j and other model learning device 100 jSpecifically, the dual variable calculation and transmission unit 150 calculates the dual variable y i|j In addition, other model learning devices 200 j There must be at least one, but there may be two or more. Also, the arrow in the formula indicates substitution.

number

[0017] Each model learning device 1001, ..., 100 N The dual variable receiving unit 160 of the model learning device 100 i and other model learning devices 100 connected to it. j From the dual variable y j|i The dual variable y received by the dual variable receiving unit 160 is received (S160). j|i The model learning device 100 j Since the dual variable calculation and transmission unit 150 calculates and transmits the dual variable y i|j The positions of "j" and "i" are reversed.

[0018] Each model learning device 1001, ..., 100 N The dual variable setting unit 170 sets the received dual variable y j|i The dual variable z used in the next learning i|j r,k+1 The dual variable setting unit 170 sets the received dual variable y as follows (S170): j|i The dual variable z used in the next learning i|j r,k+1 This can be done as follows.

number

[0019] Each model learning device 1001, ..., 100 NIf the inner loop process has not been completed (if No), the process continues, and if the inner loop process has been completed (if Yes), the process proceeds to check the outer loop process (S180). If the result of step S180 is Yes, each model learning device 1001, ..., 100 N checks whether the processing of the outer loop step has ended, and if it has not ended (if No), continues the repeated processing, and if it has ended (if Yes), ends the processing (S190).

[0020] According to the federated learning system 10, model learning devices 1001, . . . , 100 N The model parameter update unit 140 updates the model parameters w using a coefficient γ that uses a predetermined optimal value η and a predetermined high parameter α. i r,k+1 In addition, the dual variable calculation and transmission unit 150 updates the model learning device 100. i and other model learning devices 100 connected to it. j For each iteration, the dual variables are calculated and transmitted using the updated model parameters and coefficient γ. The degree of updating can be adjusted using coefficient γ, so stable federated learning can be performed even when the variance of the distributed data is large or when there is a large difference in the computational and communication performance of users. [Example]

[0021] In the first embodiment, the first problem, "performing stable federated learning even when the variance of distributed data is large or when there is a large difference in the computational and communication performance of users," was solved. By adding limitations to the means for solving the first problem, it is also possible to solve the problem of "achieving both a reduction in the risk of data leakage and an improvement in accuracy." Therefore, in the second embodiment, "achieving both a reduction in the risk of data leakage and an improvement in accuracy" is set as the second problem.

[0022] FIG. 7 shows an example of the configuration of a federated learning system according to the second embodiment. FIG. 4 shows an example of model learning devices of the present invention connected in a ring configuration, and FIG. 5 shows an example of model learning devices of the present invention connected randomly. FIG. 8 is a diagram showing an example of the processing flow of the federated learning system according to the second embodiment. The federated learning system 20 according to the second embodiment comprises N model learning devices 2001, ..., 200 N Each model learning device 200 i is another model learning device 200 j Each model learning device 200 is connected to one of the following via a network. i includes an initial setting unit 110, a noise generation unit 220, a mini-batch extraction unit 130, a model parameter update unit 140, a dual variable calculation and transmission unit 250, a dual variable reception unit 160, and a dual variable setting unit 170.

[0023] Each model learning device 2001,...,200 N The processing of the initial setting unit 110 and the mini-batch extraction unit 130 (S130) is the same as in the first embodiment. N The noise generating unit 220 generates noise n i Specifically, a predetermined variance σ is generated (S220). 2 Based on Gaussian noise n i can be generated.

[0024] Each model learning device 2001,...,200 N The model parameter update unit 140 performs learning using the dual variables, step size, mini-batch of model learning data, constraint parameters, and coefficient γ using a predetermined optimal value η and a predetermined high parameter α, and updates the model parameters (S140). Specifically, the model parameter update unit 140 updates the model parameters as shown in the following equation, and updates the model parameters w to be used in the next processing. i r,k+1 Ask for.

number

[0025] Each model learning device 2001,...,200 N The dual variable calculation and transmission unit 250 is a model learning device 200 i and another model learning device 200 connected to j The model parameter update unit 140 updates the model parameters w i r,k+1 and coefficient γ, we add noise to the dual variable y i|j and other model learning device 200 j In addition, the other model learning device 200 j Specifically, the dual variable calculation and transmission unit 250 calculates the dual variable y i|j Ask for.

number

[0026] The processing by the dual variable receiving unit 160 (S160) and the processing by the dual variable setting unit 170 (S170) are the same as in embodiment 1. Also, as in embodiment 1, the inner loop process is confirmed (S180) and the outer loop process is confirmed (S190) to execute the iterative processing.

[0027] FIG. 9 is a diagram showing the algorithm of the federated learning system 20 of the second embodiment described above in the same manner as in Non-Patent Document 1. According to the federated learning system 20, ξ i r,k Since noise is not added to the model learning device 200, the influence of noise on learning can be reduced, and the accuracy of model learning can be improved. j The dual variable y to send to i|j Since noise is added to the data, the risk of information leakage can be reduced. Therefore, in addition to solving the first problem of "performing stable federated learning even when the variance of distributed data is large or when there are large differences in the computational and communication performance of users," it also solves the problem of achieving both reduced risk of information leakage and improved accuracy.

[0028] [Programs, recording media] The various processes described above can be implemented by loading a program that executes each step of the above method into the recording unit 2020 of the computer 2000 shown in Figure 10, and operating the control unit 2010, input unit 2030, output unit 2040, display unit 2050, etc.

[0029] The program describing the processing contents can be recorded on a computer-readable recording medium, which may be, for example, a magnetic recording device, an optical disk, a magneto-optical recording medium, a semiconductor memory, or any other suitable recording medium.

[0030] The program may be distributed, for example, by selling, transferring, lending, etc. a portable recording medium such as a DVD or CD-ROM on which the program is recorded. Furthermore, the program may be stored in a storage device of a server computer, and then transferred from the server computer to another computer via a network, thereby distributing the program.

[0031] A computer that executes such a program may first temporarily store the program recorded on a portable recording medium or transferred from a server computer in its own storage device. Then, when executing a process, the computer reads the program stored on its own recording medium and executes the process in accordance with the read program. Alternatively, the computer may read the program directly from a portable recording medium and execute the process in accordance with the program. Furthermore, the computer may execute the process in accordance with the received program each time a program is transferred from a server computer to the computer. Alternatively, the server computer may not transfer the program to the computer, but may execute the process through a so-called ASP (Application Service Provider) service, which realizes the processing function by issuing an execution instruction and obtaining the results. In this embodiment, the program includes information used for processing by a computer that is equivalent to a program (such as data that is not a direct instruction to the computer but has properties that define computer processing).

[0032] Furthermore, in this embodiment, the device is configured by executing a predetermined program on a computer, but at least a part of the processing contents may be realized by hardware. [Explanation of symbols]

[0033] 10,20 Federated Learning System 100,200 model learning devices 110 Initial setting section 130 Mini-batch extraction part 140 Model parameter update unit 150,250 Dual variable calculation and transmission section 160 Dual variable receiver 170 Dual variable setting section 220 Noise Generation Unit

Claims

1. A federated learning system having a plurality of model learning devices, each of which is connected to any of the other model learning devices via a network, The model parameters, dual variables, step size, model training data, and constraint parameters are set to predetermined initial values. Each model learning device is a mini-batch extraction unit that extracts a predetermined amount of data as a mini-batch from the model learning data; a model parameter update unit that performs learning using a coefficient γ using a dual variable, a step size, a mini-batch of model learning data, a constraint parameter, a predetermined optimal value η, and a predetermined high parameter α, and updates the model parameters; a dual variable calculation and transmission unit that calculates and transmits dual variables for each of the other model learning devices connected to the model learning device using the model parameters and coefficient γ updated by the model parameter update unit; a dual variable receiving unit that receives dual variables from another model learning device connected to the model learning device; a dual variable setting unit that sets the received dual variables as dual variables to be used in the next learning; Equipped with A federated learning system characterized by:

2. A federated learning system having a plurality of model learning devices, each of which is connected to any of the other model learning devices via a network, The model parameters, dual variables, step size, model training data, and constraint parameters are set to predetermined initial values. Each model learning device is a noise generating unit that generates noise; a mini-batch extraction unit that extracts a predetermined amount of data as a mini-batch from the model learning data; a model parameter update unit that performs learning using a coefficient γ using a dual variable, a step size, a mini-batch of model learning data, a constraint parameter, a predetermined optimal value η, and a predetermined high parameter α, and updates the model parameters; a dual variable calculation and transmission unit that calculates and transmits the noise-added dual variables for each of the other model learning devices connected to the model learning device, using the model parameters and coefficient γ updated by the model parameter update unit; a dual variable receiving unit that receives dual variables from another model learning device connected to the model learning device; a dual variable setting unit that sets the received dual variables as dual variables to be used in the next learning; Equipped with A federated learning system characterized by:

3. A model learning device that configures a federated learning system having a plurality of model learning devices, each of which is connected to any of the other model learning devices via a network, The model parameters, dual variables, step size, model training data, and constraint parameters are set to predetermined initial values. a mini-batch extraction unit that extracts a predetermined amount of data as a mini-batch from the model learning data; a model parameter update unit that performs learning using a coefficient γ using a dual variable, a step size, a mini-batch of model learning data, a constraint parameter, a predetermined optimal value η, and a predetermined high parameter α, and updates the model parameters; a dual variable calculation and transmission unit that calculates and transmits dual variables for each of the other model learning devices connected to the model learning device using the model parameters and coefficient γ updated by the model parameter update unit; a dual variable receiving unit that receives dual variables from another model learning device connected to the model learning device; a dual variable setting unit that sets the received dual variables as dual variables to be used in the next learning; Equipped with A model learning device characterized by:

4. 4. The model learning device according to claim 3, The model parameter update unit [Equation 7] However, w ｉ ｒ，ｋ＋１ is the model parameter, r is the number of learning iterations in the entire federated learning system, k is the number of learning iterations in the model learning device, i and j are symbols indicating the model learning device, f － is a cost function or a function that can be replaced with the cost function, u is the parameter of the model before updating, ξ ｉ ｒ，ｋ is a mini-batch of data for model learning, the coefficient γ is 1 + αη, N is the number of model learning devices that make up the federated learning system, A ｉ｜ｊ is the constraint parameter, z ｉ｜ｊ ｒ is the dual variable As shown above, the model parameters w ｉ ｒ，ｋ＋１ Update The dual variable calculation and transmission unit [Equation 8] However, y ｉ｜ｊ is the dual variable To find the dual variables, A model learning device characterized by:

5. A model learning device that configures a federated learning system having a plurality of model learning devices, each of which is connected to any of the other model learning devices via a network, The model parameters, dual variables, step size, model training data, and constraint parameters are set to predetermined initial values. a noise generating unit that generates noise; a mini-batch extraction unit that extracts a predetermined amount of data as a mini-batch from the model learning data; a model parameter update unit that performs learning using a coefficient γ using a dual variable, a step size, a mini-batch of model learning data, a constraint parameter, a predetermined optimal value η, and a predetermined high parameter α, and updates the model parameters; a dual variable calculation and transmission unit that calculates and transmits the noise-added dual variables for each of the other model learning devices connected to the model learning device, using the model parameters and coefficient γ updated by the model parameter update unit; a dual variable receiving unit that receives dual variables from another model learning device connected to the model learning device; a dual variable setting unit that sets the received dual variables as dual variables to be used in the next learning; Equipped with A model learning device characterized by:

6. 6. The model learning device according to claim 5, The model parameter update unit [Equation 9] However, w ｉ ｒ，ｋ＋１ is the model parameter, r is the number of learning iterations in the entire federated learning system, k is the number of learning iterations in the model learning device, i and j are symbols indicating the model learning device, f － is a cost function or a function that can be replaced with the cost function, u is the parameter of the model before updating, ξ ｉ ｒ，ｋ is a mini-batch of data for model learning, the coefficient γ is 1 + αη, N is the number of model learning devices that make up the federated learning system, A ｉ｜ｊ is the constraint parameter, z ｉ｜ｊ ｒ is the dual variable As shown above, the model parameters w ｉ ｒ，ｋ＋１ Update The dual variable calculation and transmission unit [Equation 10] However, y ｉ｜ｊ is the noise-added dual variable, n ｉ is the noise generated by the noise generating unit To find the dual variables with added noise, A model learning device characterized by:

7. A federated learning method using a federated learning system having a plurality of model learning devices, each of which is connected to any of the other model learning devices via a network, comprising: The model parameters, dual variables, step size, model training data, and constraint parameters are set to predetermined initial values. Each model learning device, A model parameter update step in which learning is performed using a coefficient γ using a dual variable, a step size, a mini-batch of model learning data, a constraint parameter, a predetermined optimal value η, and a predetermined high parameter α, and the model parameters are updated; a dual variable calculation and transmission step of calculating and transmitting dual variables for each of the other model learning devices connected to the model learning device using the model parameters and coefficient γ updated in the model parameter update step; a dual variable receiving step of receiving dual variables from another model learning device connected to the model learning device; a dual variable setting step of setting the received dual variables as dual variables to be used in the next learning; an inner loop process for executing the above a first predetermined number of times; an outer loop step for executing the inner loop step a second predetermined number of times; Associative learning methods.

8. A model learning program that causes a computer to function as the model learning device according to any one of claims 3 to 6.

Images