Federal learning method, system, device, equipment, medium and product

By using inference identifiers and model inference information for model retraining in longitudinal federated learning, the problem of insufficient model performance improvement in existing schemes is solved, and more efficient model performance improvement is achieved.

CN118917386APending Publication Date: 2024-11-08CHINA MOBILE COMM LTD RES INST +1
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202411092992.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-09
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

Existing longitudinal federated learning schemes lack methods to further improve model performance after deep learning models have been trained, resulting in insufficient flexibility and accuracy in improving model performance.

Method used

During model inference, the coordinating node sends inference identifiers to participating nodes, saves model inference information and/or inference identifiers, and initiates retraining when needed by obtaining model inference information and identifiers from participating nodes and the coordinating node.

Benefits of technology

This improves the flexibility and accuracy of retraining the model after inference, thereby enhancing model performance.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118917386A_ABST
    Figure CN118917386A_ABST
Patent Text Reader

Abstract

The invention relates to the field of artificial intelligence, and particularly provides a federal learning method, system, device and equipment, a medium and a product. According to the method, in a model reasoning process, a coordination node sends a distributed reasoning identifier to a participating node and stores model reasoning information and / or the reasoning identifier of the coordination node, so that under the condition that it is determined that a model needs to be retrained, the model is retrained by sending the reasoning identifier to the participating node; according to the method, the model reasoning information and / or reasoning identification of the coordinating node can be obtained, and then the model is retrained based on the model reasoning information and / or reasoning identification of the participating node and / or the model reasoning information and / or reasoning identification of the coordinating node. Compared with the prior art, the method and the device have the advantages that the model is retrained after model reasoning is finished, so that the performance of the model is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence technology, and in particular to a federated learning method, system, device, equipment, medium and product. Background Art

[0002] Deep learning technology has been widely used in various industries, such as recommendation systems, autonomous driving, and healthcare. However, as people's awareness of data privacy and security increases, the traditional data-centralized deep learning model training method may no longer be feasible. In order to address the above privacy challenges, federated learning has emerged as a new machine learning paradigm that can perform distributed training models without leaving the local data.

[0003] Vertical Federated Learning (VFL) is a type of federated learning. According to the current research progress of the 3GPP AIML project, a deep learning model training scheme based on vertical federated learning has emerged. However, there is currently no solution to further improve the model performance based on vertical federated learning after the model training is completed based on the deep learning model training scheme. Summary of the invention

[0004] In view of the above problems, the present disclosure is proposed. The present disclosure provides a federated learning method, system, device, equipment, medium and product.

[0005] According to a first aspect of the present disclosure, a federated learning method is provided, which is applied to a coordination node, and the method includes:

[0006] During the model inference process, the inference identifier is sent to the participating nodes;

[0007] Saving the model reasoning information of the coordination node and / or the reasoning identifier;

[0008] In the case of judging or determining that the model needs to be retrained, sending the reasoning identifier to the participating node;

[0009] Receiving the model reasoning information and / or the reasoning identifier returned by the participating node; and / or obtaining the model reasoning information and / or the reasoning identifier of the coordinating node;

[0010] Initiate retraining of the model.

[0011] In addition, according to the federated learning method of the first aspect of the present disclosure, saving the model reasoning information of the coordination node and / or the reasoning identifier includes:

[0012] Sending the model reasoning information of the coordination node and / or the reasoning identifier to the storage node;

[0013] or,

[0014] The model reasoning information of the coordination node and / or the reasoning identifier are stored locally.

[0015] In addition, according to the federated learning method of the first aspect of the present disclosure, obtaining the model reasoning information and / or reasoning identification of the coordination node includes:

[0016] Sending a request message to the storage node, carrying the reasoning identifier, and receiving the model reasoning information and / or the reasoning identifier returned by the storage node;

[0017] or,

[0018] The model reasoning information and / or reasoning identifier of the coordination node is obtained locally.

[0019] In addition, according to the federated learning method of the first aspect of the present disclosure, initiating retraining of the model includes:

[0020] Based on the inference identifier, the model inference information returned by the participating nodes and / or the model inference information of the coordinating node are associated to obtain retrained data;

[0021] The model is retrained using the data.

[0022] In addition, according to the federated learning method of the first aspect of the present disclosure, the model reasoning information of the coordination node includes at least one of the following:

[0023] Inference identification;

[0024] Model identification;

[0025] Label;

[0026] True value;

[0027] Local data;

[0028] Federated learning related logo.

[0029] In addition, according to the federated learning method of the first aspect of the present disclosure, judging or determining that the model needs to be retrained includes:

[0030] Based on the number of inferences in the model inference process, the deviation value of the inference result, and the standard deviation of the deviation 、 One or more of the deviation ratios judge or determine that the model needs to be retrained.

[0031] According to a second aspect of the present disclosure, a federated learning method is provided, which is applied to a participating node, and the method includes:

[0032] receiving an inference identifier from a coordinating node;

[0033] Saving the model reasoning information of the participating nodes and / or the reasoning identifier;

[0034] Receiving the inference identifier sent by the coordination node;

[0035] Participate in retraining of the model based on the inference identifier.

[0036] In addition, according to the federated learning method of the second aspect of the present disclosure, participating in the retraining of the model based on the inference identifier includes:

[0037] The model reasoning information of the participating nodes is obtained, and the model reasoning information of the participating nodes and / or the reasoning identifier are returned to the coordinating node.

[0038] In addition, according to the federated learning method of the second aspect of the present disclosure, the model reasoning information of the participating nodes and / or the reasoning identifier are saved, including:

[0039] Sending the model reasoning information of the participating node and / or the reasoning identifier to the storage node;

[0040] or,

[0041] The model reasoning information of the participating nodes and / or the reasoning identifier are stored locally.

[0042] In addition, according to the federated learning method of the second aspect of the present disclosure, obtaining the model reasoning information of the participating nodes includes:

[0043] Sending a request message to the storage node, carrying the reasoning identifier, and receiving the model reasoning information of the participating node returned by the storage node;

[0044] or,

[0045] The model reasoning information of the participating nodes is obtained locally.

[0046] In addition, according to the federated learning method of the second aspect of the present disclosure, the model reasoning information of the participating nodes includes at least one of the following:

[0047] Inference identification;

[0048] Model identification;

[0049] Local data;

[0050] Federated learning related logo.

[0051] According to a third aspect of the present disclosure, a federated learning method is provided, which is applied to a storage node, and the method includes:

[0052] Receiving a reasoning identifier from a coordinating node and / or model reasoning information of the coordinating node; and / or receiving a reasoning identifier from a participating node and / or model reasoning information of the participating node;

[0053] Receiving a request message sent by the coordinating node, carrying the reasoning identifier; and / or a request message sent by the participating node, carrying the reasoning identifier;

[0054] Returning the model reasoning information of the coordinating node to the coordinating node; and / or returning the model reasoning information of the participating node to the participating nodes.

[0055] According to a fourth aspect of the present disclosure, a federated learning system is provided, including:

[0056] Coordinating nodes, participating nodes and storage nodes;

[0057] The coordination node is used to execute the federated learning method described in the first aspect;

[0058] The participating nodes are used to execute the federated learning method described in the second aspect;

[0059] The storage node is used to execute the federated learning method described in the third aspect.

[0060] According to a fifth aspect of the present disclosure, a federated learning device is provided, which is applied to a coordination node, including:

[0061] A first sending module is used to send a reasoning identifier to a participating node during a model reasoning process;

[0062] A first saving module, used for saving the model reasoning information of the coordination node and / or the reasoning identifier;

[0063] A second sending module is used to send the reasoning identifier to the participating node when judging or determining that the model needs to be retrained;

[0064] A processing module, used for receiving the model reasoning information and / or the reasoning identification returned by the participating node; and / or obtaining the model reasoning information and / or the reasoning identification of the coordinating node;

[0065] The first training module is used to initiate retraining of the model.

[0066] According to a sixth aspect of the present disclosure, a federated learning device is provided, which is applied to a participating node, including:

[0067] A first receiving module, configured to receive an inference identifier from a coordination node;

[0068] A second saving module, used for saving the model reasoning information of the participating nodes and / or the reasoning identifier;

[0069] A second receiving module, configured to receive the inference identifier sent by the coordination node;

[0070] The second training module is used to participate in retraining of the model based on the inference identifier.

[0071] According to a seventh aspect of the present disclosure, a federated learning device is provided, which is applied to a storage node, including:

[0072] A third receiving module is used to receive the reasoning identification from the coordinating node and / or the model reasoning information of the coordinating node; and / or receive the reasoning identification from the participating node and / or the model reasoning information of the participating node;

[0073] A fourth receiving module, configured to receive a request message sent by the coordinating node, carrying the reasoning identifier; and / or a request message sent by the participating node, carrying the reasoning identifier;

[0074] The third sending module is used to return the model reasoning information of the coordinating node to the coordinating node; and / or return the model reasoning information of the participating node to the participating node.

[0075] According to an eighth aspect of the present disclosure, an electronic device is provided, comprising a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method described in the first aspect, the second aspect, or the third aspect.

[0076] According to a ninth aspect of the present disclosure, a computer-readable storage medium is provided, on which a computer program / instruction is stored. When the computer program / instruction is executed by a processor, the steps of the method described in the first aspect, the second aspect or the third aspect are implemented.

[0077] According to a tenth aspect of the present disclosure, a computer program product is provided, comprising a computer program / instruction, which implements the steps of the method described in the first aspect, the second aspect or the third aspect when executed by a processor.

[0078] As will be described in detail below, according to the federated learning method of the embodiment of the present disclosure, during the model reasoning process, the coordinating node sends the assigned reasoning identifier to the participating nodes, and saves the model reasoning information and / or reasoning identifier of the coordinating node. In this way, when it is determined that the model needs to be retrained, the model reasoning information and / or reasoning identifier of the coordinating node can be obtained by sending the reasoning identifier to the participating nodes, and then based on the model reasoning information and / or reasoning identifier of the participating nodes, and / or, the model reasoning information and / or reasoning identifier of the coordinating node, the model is retrained. Compared with the related art, the solution of this embodiment is used to achieve retraining of the model after the model reasoning is completed, thereby improving the performance of the model. Compared with the diagnostic method based on fixed thresholds in the related art, both flexibility and accuracy are improved.

[0079] It is to be understood that both the foregoing general description and the following detailed description are exemplary, and are intended to provide further explanation of the technology as claimed. BRIEF DESCRIPTION OF THE DRAWINGS

[0080] The above and other purposes, features and advantages of the present disclosure will become more apparent by describing the embodiments of the present disclosure in more detail in conjunction with the accompanying drawings. The accompanying drawings are used to provide a further understanding of the embodiments of the present disclosure and constitute a part of the specification. Together with the embodiments of the present disclosure, they are used to explain the present disclosure and do not constitute a limitation of the present disclosure. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0081] Figure 1 It is a structural diagram of the federated learning system in related technologies.

[0082] Figure 2 is a structural diagram illustrating a federated learning system according to an embodiment of the present disclosure.

[0083] Figure 3 is a flowchart illustrating the application of the federated learning method according to an embodiment of the present disclosure.

[0084] Figure 4 is a flowchart illustrating the application of the federated learning method according to an embodiment of the present disclosure.

[0085] Figure 5 is a flowchart illustrating the application of the federated learning method according to an embodiment of the present disclosure.

[0086] Figure 6 is a flowchart illustrating the application of the federated learning method according to an embodiment of the present disclosure.

[0087] Figure 7 is a structural diagram illustrating a federated learning apparatus according to an embodiment of the present disclosure.

[0088] Figure 8is a structural diagram illustrating a federated learning apparatus according to an embodiment of the present disclosure.

[0089] Fig. 9 is a structural diagram illustrating a federated learning apparatus according to an embodiment of the present disclosure.

[0090] Fig.10 is a hardware block diagram illustrating an electronic device according to an embodiment of the present disclosure.

[0091] Fig.11 is a schematic diagram illustrating a computer-readable storage medium according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0092] In order to make the purpose, technical scheme and advantages of the embodiments of the present disclosure clearer, the technical scheme in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the drawings in the embodiments of the present disclosure. Obviously, the described embodiments are only part of the embodiments of the present disclosure, rather than all of the embodiments. The components of the embodiments of the present disclosure generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present disclosure provided in the drawings is not intended to limit the scope of the present disclosure for protection, but merely represents the selected embodiments of the present disclosure. Based on the embodiments of the present disclosure, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present disclosure.

[0093] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0094] The term "and / or" herein only describes an association relationship, indicating that three relationships may exist. For example, A and / or B may represent the following three situations: A exists alone, A and B exist at the same time, and B exists alone. In addition, the term "at least one" herein represents any combination of at least two of any one or more of a plurality of. For example, including at least one of A, B, and C may represent including any one or more elements selected from the set consisting of A, B, and C.

[0095] In this specification, the term "model" refers to a machine learning model that is applied to a business scenario to provide business prediction services, such as a machine learning model for classification prediction, business risk prediction, etc. Examples of machine learning models may include, but are not limited to: linear regression model, logistic regression model, neural network model, decision tree model, support vector machine, etc. Examples of neural network models may include, but are not limited to, deep neural network (DNN) model, convolutional neural network (CNN) model, BP neural network, etc.

[0096] The specific implementation form of the model depends on the business scenario in which it is applied. For example, in an application scenario where the model is applied to classify users, the model is implemented as a user classification model. Accordingly, the user feature data of the classified users can be used to make user classification predictions. In an application scenario where the model is applied to predict business risks of business transactions occurring on a business system, the model is implemented as a business risk prediction model. Accordingly, the business transaction feature data of the business transaction can be used to make business risk predictions.

[0097] With the development of artificial intelligence technology, machine learning technology has been widely used as a model in various business application scenarios to provide various business prediction services, such as classification prediction, business risk prediction, etc. For example, the model has a wide range of applications in financial fraud, recommendation systems, image recognition and other fields. In order to achieve better model performance, more training data is needed to train the model. In application fields such as medicine and finance, different companies or institutions have different data samples. Once these data are jointly trained, the model accuracy of the model will be greatly improved, thereby bringing huge economic benefits to the company.

[0098] In view of the above, a federated learning scheme is proposed. In the federated learning scheme, multiple data owners jointly train the model with the assistance of the server. Figure 1 An example schematic diagram of a federated learning system is shown.

[0099] like Figure 1As shown, the federated learning system includes a server 110 and multiple clients (clients) 120, for example, client A 120-1, client B 120-2 and client C 120-3. The server 110 has a model 10. When performing federated learning, the server 110 sends the model 10 to each client, and each client uses it to perform model calculation based on the local data at the client to obtain the model prediction value at the client. Then, each client determines the gradient information of its own model based on the calculated model prediction value and label value, and provides the gradient information to the server 110. The server 110 uses the obtained gradient information to update the model.

[0100] Judging from the current research progress of the 3GPP AIML project, a deep learning model training scheme based on vertical federated learning has emerged. However, there is currently no solution to further improve model performance based on vertical federated learning after the model training is completed based on this deep learning model training scheme.

[0101] In order to alleviate the technical problems existing in the related art, the embodiments of the present disclosure provide a federated learning method, system, device, equipment, medium and product. In the process of model reasoning, the coordinating node sends the assigned reasoning identifier to the participating nodes, and saves the model reasoning information and / or reasoning identifier of the coordinating node. In this way, when it is determined that the model needs to be retrained, the model reasoning information and / or reasoning identifier of the coordinating node can be obtained by sending the reasoning identifier to the participating nodes, and then the model reasoning information and / or reasoning identifier of the coordinating node can be retrained based on the model reasoning information and / or reasoning identifier of the participating nodes, and / or the model reasoning information and / or reasoning identifier of the coordinating node. Compared with the related art, the scheme of this embodiment is adopted to realize the retraining of the model after the model reasoning is completed, thereby improving the performance of the model. Compared with the diagnostic method based on fixed thresholds in the related art, flexibility and accuracy are improved.

[0102] The federated learning method, system, apparatus, device, medium, and product according to the embodiments of this specification will be described in detail below with reference to the accompanying drawings.

[0103] Figure 2 A schematic diagram of the architecture of a federated learning system according to an embodiment of the present specification is shown.

[0104] like Figure 2 As shown, the federated learning system includes a coordination node 210 and at least one participating node 220. Figure 23 participating nodes 220-1, 220-2 and 220-3 are shown in FIG. In other embodiments of the present specification, more or fewer participating nodes 220 may be included. The coordination node 210 and at least one participating node 220 may communicate with each other via a network 230 such as, but not limited to, the Internet or a local area network.

[0105] In the embodiments of this specification, the participating node 220 may be a device or device party for locally collecting data samples, such as an intelligent terminal device, a server device, an edge device, etc. In this specification, the term "participating node" and the term "client" may be used interchangeably. In some optional embodiments, the participating node 220 may be a client (VFL Client) of vertical federated learning. In some other optional embodiments, the participating node 220 may also be a passive participant (VFL Passive Participant) of vertical federated learning.

[0106] The coordination node 210 maintains the model to be trained. For example, the coordination node 210 may be a terminal device or a server device at a model provider. The model maintained on the coordination node 210 may also be referred to as a global business model. In this specification, the term "coordination node" and the term "server" may be used interchangeably. In some optional embodiments, the coordination node 210 may be a server (VFL Server) of vertical federated learning. In some other optional embodiments, the coordination node 210 may also be an active participant (VFL Active Participant) of vertical federated learning.

[0107] In this specification, each participating node 220 and the coordination node 210 may be any suitable electronic device with computing capabilities, including but not limited to: personal computers, server computers, workstations, desktop computers, laptop computers, notebook computers, mobile electronic devices, smart phones, tablet computers, cellular phones, personal digital assistants (PDAs), handheld devices, messaging devices, wearable electronic devices, consumer electronic devices, etc.

[0108] In addition, the participating nodes 220-1, 220-2, 220-3 and the coordination node 210 each have a federated learning device. The federated learning devices at the participating nodes 220-1, 220-2, 220-3 and the coordination node 210 can perform network communication via the network 230 to perform data interaction, thereby collaboratively processing to perform a model training process for the business model. The operation and structure of the federated learning device will be described in detail below with reference to the accompanying drawings.

[0109] In some embodiments, the network 230 may be any one or more of a wired network or a wireless network. Examples of the network 230 may include, but are not limited to, a cable network, an optical fiber network, a telecommunications network, an enterprise intranet, the Internet, a local area network (LAN), a wide area network (WAN), a wireless local area network (WLAN), a metropolitan area network (MAN), a public switched telephone network (PSTN), a Bluetooth network, a ZigZee network (ZigZee), near field communication (NFC), a bus within a device, a line within a device, etc., or any combination thereof.

[0110] In this manual, if Figure 2 As shown, the federated learning system may further include a storage node 240 , which communicates with the participating nodes 220 - 1 , 220 - 2 , 220 - 3 and the coordination node 210 through the network 230 .

[0111] Figure 3 A flowchart of a federated learning method according to an embodiment of the present specification is shown, which can be applied to Figure 2 In the coordination node 210 shown, Figure 3 As shown, the method may include the following steps:

[0112] Step 301: During the model reasoning process, a reasoning identifier is sent to participating nodes.

[0113] In each model inference process, the coordinating node assigns an inference ID to each model inference process, and the inference ID can uniquely identify the model inference process, and the model inference information in the model inference process can be obtained through the inference ID. It should be understood that for participating nodes, the participating nodes can obtain the model inference information related to the participating nodes in the model inference process through the inference ID, and for the coordinating node, the coordinating node can obtain the model inference information related to the coordinating node in the model inference process through the inference ID.

[0114] Step 302: Save the model reasoning information and / or reasoning identifier of the coordination node.

[0115] In this embodiment, the model reasoning information may also be referred to as model information, model-related information, model reasoning data, model data or model-related data.

[0116] The model reasoning information of the coordination node includes at least one of the following:

[0117] Model identification;

[0118] Label;

[0119] True value;

[0120] Local data;

[0121] Federated learning related logo.

[0122] The model identifier is used to uniquely identify the model based on federated learning reasoning and retraining in this embodiment. In the application, the model identifier includes but is not limited to the name and number of the model.

[0123] Among them, the label is the prediction result of the model obtained at the coordination node during the model reasoning process. The ground truth is used to represent the expected output result of the model reasoning process. In the application, the ground truth can be directly obtained based on the business scenario adapted by the model reasoning process, or it can be manually input.

[0124] As an example, in a scheme for reasoning about a member's creditworthiness based on a federated learning model, the label may be the prediction result of a model obtained at a coordination node that characterizes the member's poor creditworthiness, and the true value may be the member's actual creditworthiness obtained from a bank or other financial institution. For example, the true value obtained from a bank may indicate that the member has a good creditworthiness.

[0125] In practical applications, the coordinating node may also take on the role of a participating node at the same time. In this case, the local data of the coordinating node includes data such as the input data required for model reasoning when it takes on the role of a participating node.

[0126] In some optional embodiments, the federated learning correlation identifier may also be referred to as a vertical federated learning correlation identifier (VFL correlation ID) or a federated learning correlation identifier, as well as other identifiers.

[0127] The federated learning related identifier is used to uniquely identify a federated learning system. It should be understood that the federated learning system identified by the federated learning related identifier includes the coordinating nodes and participating nodes participating in the federated learning.

[0128] In one or more embodiments, the following two implementation schemes for saving model reasoning information of a coordination node and / or a reasoning identifier are provided:

[0129] Sending the model reasoning information and / or reasoning identifier of the coordination node to the storage node;

[0130] or,

[0131] The model inference information and / or inference identifier of the coordination node is stored locally.

[0132] The storage node is a node device that is independent of the coordination node and the participating node and has a storage function. In the application, the storage node includes but is not limited to ADRF (Analytic Data Repository Function) network elements. It should be understood that storing model reasoning identifiers and / or reasoning identifiers by the storage node can alleviate the pressure on the end side. In actual applications, after the coordination node and the participating node complete the model reasoning process each time, the model reasoning information and / or reasoning identifier of the coordination node can be saved to the storage node.

[0133] In this embodiment, when the model reasoning information and the reasoning identification of the coordination node are simultaneously sent to the storage node, an association relationship between the model reasoning information and the reasoning identification can be established, and the model reasoning information and the reasoning identification can be stored according to the association relationship.

[0134] It should be understood that when the model reasoning information (or reasoning identifier) ​​of the coordination node is sent to the storage node, the storage node stores the received model reasoning information (or reasoning identifier). The reasoning identifier (or model reasoning information) not stored by the storage node can be stored locally in the coordination node. It should be noted here that although the model reasoning information and reasoning identifier of the coordination node are stored locally and stored in the storage node respectively, when the coordination node assigns the reasoning identifier, the coordination node has established an association relationship between the reasoning identifier and the reasoning process. Through this association relationship, the model reasoning information of the coordination node involved in the reasoning process can still be obtained based on the reasoning identifier.

[0135] Step 303: When it is judged or determined that the model needs to be retrained, a reasoning identifier is sent to the participating nodes.

[0136] When judging or determining that the model needs to be retrained, the reasoning identifier sent by the coordination node to the participating nodes may be one reasoning identifier, multiple reasoning identifiers, or a list of reasoning identifiers.

[0137] The judgment or determination that the model needs to be retrained indicates that the performance of the model during the inference process is poor, and therefore the model needs to be trained again. In one or more embodiments, the number of inferences during the model inference process, the deviation value of the inference result, the standard deviation of the deviation, and the like can be used to determine the need for retraining the model. 、 One or more of the deviation ratios are used to judge or determine that the model needs to be retrained.

[0138] The deviation value of the inference result is calculated based on the true value and the label, and is used to characterize the difference between the inference result of the model and the true result.

[0139] Deviation standard deviation is the square root of the sum of the squared deviations from the mean, denoted by σ. Standard deviation is also called standard deviation or experimental standard deviation, and is the most commonly used measure of the degree of statistical distribution in probability statistics. Standard deviation is an important indicator to measure the degree of dispersion of data distribution, which represents the average distance between data points and the mean. The smaller the standard deviation, the more concentrated the data; the larger the standard deviation, the more dispersed the data.

[0140] Deviation ratio is also called deviation rate, which is |actual value - expected value| / expected value × 100%. This formula is used to measure the difference between the actual value and the expected value, usually expressed as a percentage.

[0141] As examples, the following four specific implementation processes are given to judge or determine whether the model needs to be retrained based on one or more of the number of inferences, the deviation value of the inference result, the standard deviation of the deviation, and the deviation ratio:

[0142] In one implementation, when the number of inferences is greater than or equal to a number threshold, it is determined or judged that the model needs to be retrained.

[0143] In yet another implementation, when the deviation value of the inference result is greater than a deviation threshold, it is determined or judged that the model needs to be retrained.

[0144] In yet another implementation, when the number of inference results satisfying the deviation value of the inference result greater than the deviation threshold is greater than the number threshold, and the deviation standard deviation is greater than the standard deviation threshold, it is determined or judged that the model needs to be retrained.

[0145] In yet another implementation, when the deviation value of the inference result is greater than the deviation threshold and the deviation ratio is greater than the ratio threshold, it is determined or judged that the model needs to be retrained.

[0146] It should be understood that the above four specific implementation processes are only some of the solutions given in the present disclosure for judging or determining that the model needs to be retrained. Any other solutions that determine that the model needs to be retrained by referring to the number of inferences, the deviation value of the inference results, the standard deviation of the deviation and / or the deviation ratio given in this embodiment should be included in the protection scope of the present disclosure.

[0147] Step 304: Receive the model reasoning information and / or reasoning identification returned by the participating nodes; and / or obtain the model reasoning information and / or reasoning identification of the coordinating node.

[0148] It should be understood that step 304 may specifically include the following implementation schemes:

[0149] In one implementation, the model reasoning information and / or reasoning identification returned by the participating nodes is received, and at the same time, the model reasoning information and / or reasoning identification of the coordinating node is obtained.

[0150] In this solution, the coordination node can initiate retraining of the model locally based on the model reasoning information returned by the participating nodes and the model reasoning information of the coordination node. In specific implementation, the coordination node associates the model reasoning information returned by the participating nodes and / or the model reasoning information of the coordination node based on the reasoning identifier, obtains the retraining data, and uses the obtained retraining data to initiate retraining of the model.

[0151] In yet another implementation, the model reasoning information and / or reasoning identifier of the coordination node is obtained.

[0152] In this implementation, the coordinating node also receives the retraining result data returned by the participating nodes, and initiates retraining of the model based on the model reasoning information and retraining result data of the coordinating node. The retraining result data returned by the participating nodes includes but is not limited to gradients, which are the data obtained by the participating nodes when they initiate retraining of the model locally based on the model reasoning information of the participating nodes.

[0153] Based on the two storage methods of the model reasoning information and / or reasoning identification of the coordination node in the model reasoning process, the following two implementation methods of obtaining the model reasoning information and / or reasoning identification of the coordination node are given:

[0154] If during the model reasoning process, the model reasoning information and / or reasoning identifier of the coordinating node is stored in the storage node, then when it is determined that the model needs to be retrained, the model reasoning information and / or reasoning identifier of the coordinating node is obtained from the storage node.

[0155] In specific implementation, the coordination node sends a request message to the storage node, carrying an inference identifier, and receives the model inference information and / or inference identifier returned by the storage node.

[0156] If during the model inference process, the model inference information and / or inference identifier of the coordinating node is stored locally on the coordinating node, then when it is determined that the model needs to be retrained, the model inference information and / or inference identifier of the coordinating node is obtained locally from the coordinating node.

[0157] The reasoning identifier carried in the request message sent by the coordinating node to the storage node may also be one reasoning identifier, multiple reasoning identifiers, or a reasoning identifier list.

[0158] Step 305: Initiate retraining of the model.

[0159] In the solution provided by this embodiment, during the model reasoning process, the coordinating node sends the assigned reasoning identifier to the participating nodes, and saves the model reasoning information and / or reasoning identifier of the coordinating node. In this way, when it is determined that the model needs to be retrained, the model reasoning information and / or reasoning identifier of the coordinating node can be obtained by sending the reasoning identifier to the participating nodes, and then the model is retrained based on the model reasoning information and / or reasoning identifier of the participating nodes and / or the model reasoning information and / or reasoning identifier of the coordinating node. Compared with the related art, the solution of this embodiment realizes the retraining of the model after the model reasoning is completed, thereby improving the performance of the model.

[0160] Figure 4 A flowchart of a federated learning method according to an embodiment of the present specification is shown, which can be applied to Figure 2 In the participating nodes 220 shown, Figure 4 As shown, the method may include the following steps:

[0161] Step 401: Receive an inference identifier from a coordination node.

[0162] During the model inference process, the coordination node assigns an inference ID to each model inference process. The inference ID can uniquely identify each model inference process. The coordination node will send the inference ID to the participating nodes that jointly complete the model inference with the coordination node.

[0163] Step 402: Save the model reasoning information of the participating nodes and / or the reasoning identification.

[0164] In this embodiment, the model reasoning information of the participating nodes includes at least one of the following:

[0165] Inference identification;

[0166] Model identification;

[0167] Local data;

[0168] Federated learning related logo.

[0169] Among them, the inference ID, model ID and federated learning related ID are all described in the above embodiments, so as to avoid redundancy, they are not described here. The local data of the participating node includes the input data and other data required for the coordination node to perform model inference.

[0170] In one or more embodiments, the following two implementation schemes for saving model reasoning information of participating nodes and / or reasoning identifiers are provided:

[0171] Sending model reasoning information of participating nodes and / or reasoning identifiers to storage nodes;

[0172] or,

[0173] The model reasoning information and / or reasoning identification of the participating nodes are stored locally.

[0174] The storage node is a node device independent of the coordination node and the participating node. In the application, the storage node includes but is not limited to the ADRF (Analytic Data Repository Function) network element. It should be understood that the storage node stores the model reasoning identifier and / or reasoning identifier, which can alleviate the pressure on the end side. In actual applications, after the participating node cooperates with the coordination node to complete the model reasoning process each time, the model reasoning information and / or reasoning identifier of the participating node can be saved to the storage node.

[0175] In this embodiment, when the model reasoning information and reasoning identification of the participating nodes are sent to the storage node at the same time, an association relationship between the model reasoning information and the reasoning identification can be established, and the model reasoning information and the reasoning identification can be stored according to the association relationship.

[0176] It should be understood that when the model reasoning information (or reasoning identification) of the participating node is sent to the storage node, the storage node stores the received model reasoning information (or reasoning identification). The reasoning identification (or model reasoning information) not stored by the storage node can be stored locally in the participating node. It should be noted here that although the model reasoning information and reasoning identification of the participating node are stored locally and in the storage node respectively, after the participating node receives the reasoning identification of the coordinating node during the model reasoning process, the participating node can establish an association relationship between the reasoning identification and the reasoning process. Through this association relationship, the model reasoning information of the participating node involved in the reasoning process can still be obtained based on the reasoning identification.

[0177] Step 403: Receive the inference identifier sent by the coordination node.

[0178] It should be understood that the coordination node will send a reasoning identifier to the participating nodes when judging or determining that the model needs to be retrained. The reasoning identifier here can be one reasoning identifier, or multiple reasoning identifiers, or a reasoning identifier list, which is not specifically limited in this embodiment.

[0179] Step 404: Retrain the model based on the inference identifier.

[0180] This embodiment provides the following two implementation schemes for participating nodes to participate in model retraining based on inference identification:

[0181] In one implementation, a participating node obtains model reasoning information of the participating node based on a reasoning identifier, and returns the model reasoning information and / or reasoning identifier of the participating node to the coordination node.

[0182] In this implementation, after the coordinating node receives the model reasoning information returned by the participating nodes, it associates the model reasoning information returned by the participating nodes with the model reasoning information of the coordinating node through the reasoning identifier, obtains the data for model retraining, and uses the data to retrain the model.

[0183] In another implementation, the participating node obtains the model reasoning information of the participating node based on the reasoning identifier, and initiates retraining of the model locally at the participating node using the model reasoning information of the participating node, obtains retraining result data, and returns the retraining result data to the coordination node. The retraining result data includes but is not limited to gradients.

[0184] Based on the two storage methods of the model reasoning information and / or reasoning identification of the participating nodes in the model reasoning process, the following two implementation methods of obtaining the model reasoning information and / or reasoning identification of the participating nodes are given:

[0185] If, during the model reasoning process, the model reasoning information and / or reasoning identification of the participating nodes are stored in the storage node, then the model reasoning information and / or reasoning identification of the participating nodes are obtained from the storage node.

[0186] In specific implementation, the participating node sends a request message to the storage node, carrying an inference identifier, and receives the model inference information and / or inference identifier returned by the storage node.

[0187] The reasoning identifier carried in the request message sent by the participating node to the storage node may be one reasoning identifier, multiple reasoning identifiers, or a reasoning identifier list.

[0188] If during the model reasoning process, the model reasoning information and / or reasoning identification of the participating node is stored locally at the participating node, then the model reasoning information and / or reasoning identification of the participating node is obtained locally from the participating node.

[0189] Figure 5 A flowchart of a federated learning method according to an embodiment of the present specification is shown, which can be applied to Figure 2 In the storage node 240 shown, Figure 5 As shown, the method may include the following steps:

[0190] Step 501: receiving a reasoning identifier from a coordinating node and / or model reasoning information of the coordinating node; and / or receiving a reasoning identifier from a participating node and / or model reasoning information of the participating node;

[0191] Step 502: receiving a request message sent by a coordinating node, carrying a reasoning identifier; and / or receiving a request message sent by a participating node, carrying a reasoning identifier;

[0192] Step 503: Return the model reasoning information of the coordinating node to the coordinating node; and / or return the model reasoning information of the participating nodes to the participating nodes.

[0193] It should be understood that during the model reasoning process, the coordinating node and the participating nodes respectively send model reasoning information and / or reasoning identification to the storage node, and the storage node stores the received model reasoning information and / or reasoning identification of the participating nodes, and stores the received model reasoning information and / or reasoning identification of the coordinating node.

[0194] When the coordinating node determines or judges that the model needs to be retrained, the coordinating node sends a request message carrying an inference identifier to the storage node, and the participating node sends a request message carrying an inference identifier to the storage node. In response to the request message, the storage node returns the model inference information and / or inference identifier of the coordinating node to the coordinating node, and the storage node returns the model inference information and / or inference identifier of the participating node to the participating node.

[0195] The following example takes the coordination node as VFL Server, the participating node as VFL Client, and the storage node as ADRF as an example. Figure 6 The flowchart of the federated learning method shown in Figure 1 can be applied to Figure 2 In the federated learning system shown in Figure 6 As shown, the method may include the following steps:

[0196] Step 1: During each model inference process, the VFL Server assigns an inference ID and sends it to each VFL Client.

[0197] Step 2: After VFL Server and VFL Client complete model inference, they obtain the ground truth and save the inference id, model id, label, ground truth, VFL Client local data (other data features), VFL correlation ID and other information to ADRF.

[0198] Step 3: After completing model inference, the VFL Client needs to save the inference ID, model ID, VFL Client local data (other data features), VFL correlation ID and other information to ADRF.

[0199] Step 4: VFL Server determines whether to initiate the retraining process based on information such as the number of inferences, the deviation value of the inference result, the deviation, and the deviation ratio.

[0200] If retraining is initiated, the following steps are performed:

[0201] Step 5: VFL Server sends the Inference ID or Inference ID list to ADRF and obtains relevant data from ADRF, including: model ID, label and inference ID, ground truth, VFL correlation ID, VFL Server local data and other information.

[0202] Step 6: VFL Server sends Inference ID or Inference ID list to each VFL Client, VFL Client sends Inference ID or Inference ID list to ADRF, and obtains relevant data from ADRF, including: model ID, inference ID, VFL Client local data, VFL correlation ID, etc. VFL Client can feed back relevant information to VFL Server.

[0203] Step 7:

[0204] Method 1: VFL Server jointly initiates the retraining process with VFL Client. During the training process, VFL Server needs to carry the inference ID or the inference ID list. Based on the inference ID or the inference ID list, the gradient obtained by the VFL Client in the local retraining is associated to retrain the model.

[0205] Method 2: If the VFL Server receives the data returned by the VFL Client, it is associated through the inference ID, and the data returned by the VFL Server from ADRF and the data returned by the VFL Client from ADRF are used as the input data for retraining, and retraining is performed locally.

[0206] The present disclosure also provides a federated learning device, which can be applied to a coordination node. Figure 7 As shown, the device comprises:

[0207] A first sending module 71 is used to send a reasoning identifier to a participating node during a model reasoning process;

[0208] A first saving module 72, used for saving the model reasoning information of the coordination node and / or the reasoning identifier;

[0209] A second sending module 73 is used to send the reasoning identifier to the participating node when judging or determining that the model needs to be retrained;

[0210] The processing module 74 is used to receive the model reasoning information and / or the reasoning identifier returned by the participating node; and / or obtain the model reasoning information and / or the reasoning identifier of the coordinating node;

[0211] The first training module 75 is used to initiate retraining of the model.

[0212] In one or more embodiments, the first saving module 72 is used to:

[0213] Sending the model reasoning information of the coordination node and / or the reasoning identifier to the storage node;

[0214] or,

[0215] The model reasoning information of the coordination node and / or the reasoning identifier are stored locally.

[0216] In one or more embodiments, the processing module 74 is used to:

[0217] Sending a request message to the storage node, carrying the reasoning identifier, and receiving the model reasoning information and / or the reasoning identifier returned by the storage node;

[0218] or,

[0219] The model reasoning information and / or reasoning identifier of the coordination node is obtained locally.

[0220] In one or more embodiments, the first training module 75 is used to:

[0221] Based on the inference identifier, the model inference information returned by the participating nodes and / or the model inference information of the coordinating node are associated to obtain retrained data;

[0222] The model is retrained using the data.

[0223] In one or more embodiments, the model reasoning information of the coordination node includes at least one of the following:

[0224] Inference identification;

[0225] Model identification;

[0226] Label;

[0227] True value;

[0228] Local data;

[0229] Federated learning related logo.

[0230] In one or more embodiments, the second sending module 73 is used to:

[0231] Based on the number of inferences in the model inference process, the deviation value of the inference result, and the standard deviation of the deviation 、 One or more of the deviation ratios judge or determine that the model needs to be retrained.

[0232] The present disclosure also provides a federated learning device, which can be applied to participating nodes. Figure 8 As shown, the device comprises:

[0233] A first receiving module 81, configured to receive an inference identifier from a coordination node;

[0234] A second saving module 82, used for saving the model reasoning information of the participating nodes and / or the reasoning identifier;

[0235] A second receiving module 83, configured to receive the inference identifier sent by the coordination node;

[0236] The second training module 84 is used to participate in retraining of the model based on the inference identifier.

[0237] In one or more embodiments, the second training module 84 is used to:

[0238] The model reasoning information of the participating nodes is obtained, and the model reasoning information of the participating nodes and / or the reasoning identifier are returned to the coordinating node.

[0239] In one or more embodiments, the second saving module 82 is used to:

[0240] Sending the model reasoning information of the participating node and / or the reasoning identifier to the storage node;

[0241] or,

[0242] The model reasoning information of the participating nodes and / or the reasoning identifier are stored locally.

[0243] In one or more embodiments, the second training module 84 is used to:

[0244] Sending a request message to the storage node, carrying the reasoning identifier, and receiving the model reasoning information of the participating node returned by the storage node;

[0245] or,

[0246] The model reasoning information of the participating nodes is obtained locally.

[0247] In one or more embodiments, the model reasoning information of the participating nodes includes at least one of the following:

[0248] Inference identification;

[0249] Model identification;

[0250] Local data;

[0251] Federated learning related logo.

[0252] The present disclosure also provides a federated learning device, which can be applied to a storage node. Fig. 9 As shown, the device comprises:

[0253] The third receiving module 91 is used to receive the reasoning identification from the coordinating node and / or the model reasoning information of the coordinating node; and / or receive the reasoning identification from the participating node and / or the model reasoning information of the participating node;

[0254] The fourth receiving module 92 is used to receive the request message sent by the coordinating node, carrying the reasoning identifier; and / or the request message sent by the participating node, carrying the reasoning identifier;

[0255] The third sending module 93 is used to return the model reasoning information of the coordinating node to the coordinating node; and / or return the model reasoning information of the participating node to the participating node.

[0256] The present disclosure also provides an electronic device to perform the above-mentioned federated learning method. Fig.10 It shows a schematic diagram of an electronic device provided by some embodiments of the present disclosure. Fig.10As shown, the electronic device 8 includes: a processor 800, a memory 801, a bus 802 and a communication interface 803, wherein the processor 800, the communication interface 803 and the memory 801 are connected via the bus 802; the memory 801 stores a computer program that can be run on the processor 800, and when the processor 800 runs the computer program, the federated learning method provided in any of the aforementioned embodiments of the present disclosure is executed.

[0257] The memory 801 may include a high-speed random access memory (RAM), and may also include a non-volatile memory, such as at least one disk storage. The communication connection between the device network element and at least one other network element is realized through at least one communication interface 803 (which may be wired or wireless), and the Internet, wide area network, local area network, metropolitan area network, etc. may be used.

[0258] The bus 802 may be an ISA bus, a PCI bus, or an EISA bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. The memory 801 is used to store programs, and the processor 800 executes the program after receiving an execution instruction. The federated learning method disclosed in any implementation of the aforementioned embodiment of the present disclosure may be applied to the processor 800, or implemented by the processor 800.

[0259] The processor 800 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the hardware integrated logic circuit in the processor 800 or the instruction in the form of software. The above processor 800 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The disclosed methods, steps and logic block diagrams in the embodiments of the present disclosure can be implemented or executed. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor. The steps of the method disclosed in conjunction with the embodiments of the present disclosure can be directly embodied as a hardware decoding processor to be executed, or the hardware and software modules in the decoding processor can be executed. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory 801, and the processor 800 reads the information in the memory 801 and completes the steps of the above method in combination with its hardware.

[0260] The electronic device provided by the embodiment of the present disclosure and the federated learning method provided by the embodiment of the present disclosure are based on the same inventive concept and have the same beneficial effects as the methods adopted, operated or implemented by them.

[0261] The present disclosure also provides a computer-readable storage medium corresponding to the federated learning method provided in the above embodiment. Fig.11 The computer-readable storage medium shown is a CD 30 on which a computer program (ie, a program product) is stored. When the computer program is run by a processor, the federated learning method provided by any of the aforementioned embodiments is executed.

[0262] It should be noted that examples of the computer-readable storage medium may also include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other optical or magnetic storage media, which are not listed here one by one.

[0263] The computer-readable storage medium provided by the above-mentioned embodiments of the present disclosure and the federated learning method provided by the embodiments of the present disclosure are based on the same inventive concept and have the same beneficial effects as the methods adopted, run or implemented by the application programs stored therein.

[0264] It should be noted that:

[0265] In the description provided herein, a large number of specific details are described. However, it is understood that the embodiments of the present disclosure can be practiced without these specific details. In some instances, well-known structures and technologies are not shown in detail so as not to obscure the understanding of this description.

[0266] Similarly, it should be understood that in order to streamline the present disclosure and aid in understanding one or more of the various inventive aspects, in the above description of the exemplary embodiments of the present disclosure, the various features of the present disclosure are sometimes grouped together into a single embodiment, figure, or description thereof. However, the disclosed method should not be interpreted as reflecting the following schematic diagram: the claimed disclosure requires more features than the features explicitly recited in each claim. More specifically, as reflected in the claims below, the inventive aspects lie in less than all the features of the single embodiment disclosed above. Therefore, the claims that follow the specific embodiment are hereby expressly incorporated into the specific embodiment, with each claim itself serving as a separate embodiment of the present disclosure.

[0267] In addition, those skilled in the art will appreciate that, although some embodiments described herein include certain features included in other embodiments but not other features, the combination of features of different embodiments is meant to be within the scope of this disclosure and form different embodiments. For example, in the claims below, any one of the claimed embodiments may be used in any combination.

[0268] The above is only a preferred specific implementation of the present disclosure, but the protection scope of the present disclosure is not limited thereto. Any changes or substitutions that can be easily thought of by any technician familiar with the technical field within the technical scope disclosed in the present disclosure should be included in the protection scope of the present disclosure. Therefore, the protection scope of the present disclosure should be based on the protection scope of the claims.

Claims

1. A federated learning method, characterized in that: Applied to a coordination node, the method comprises: During the model inference process, the inference identifier is sent to the participating nodes; Saving the model reasoning information of the coordination node and / or the reasoning identifier; In the case of judging or determining that the model needs to be retrained, sending the reasoning identifier to the participating node; Receiving the model reasoning information and / or the reasoning identifier returned by the participating node; and / or obtaining the model reasoning information and / or the reasoning identifier of the coordinating node; Initiate retraining of the model.

2. The method according to claim 1, characterized in that Saving the model reasoning information of the coordination node and / or the reasoning identifier includes: Sending the model reasoning information of the coordination node and / or the reasoning identifier to the storage node; or, The model reasoning information of the coordination node and / or the reasoning identifier are stored locally.

3. The method according to claim 2, characterized in that Acquiring the model reasoning information and / or reasoning identification of the coordination node includes: Sending a request message to the storage node, carrying the reasoning identifier, and receiving the model reasoning information and / or the reasoning identifier returned by the storage node; or, The model reasoning information and / or reasoning identifier of the coordination node is obtained locally.

4. The method according to claim 1, characterized in that Initiate retraining of the model, including: Based on the inference identifier, the model inference information returned by the participating nodes and / or the model inference information of the coordinating node are associated to obtain retrained data; The model is retrained using the data.

5. The method according to any one of claims 1 to 4, characterized in that: The model reasoning information of the coordination node includes at least one of the following: Inference identification; Model identification; Label; True value; Local data; Federated learning related logo.

6. The method according to claim 1, characterized in that Judging or determining that the model needs to be retrained includes: Based on the number of inferences in the model inference process, the deviation value of the inference result, and the standard deviation of the deviation 、 One or more of the deviation ratios judge or determine that the model needs to be retrained.

7. A federated learning method, characterized in that: Applied to a participating node, the method comprises: receiving an inference identifier from a coordinating node; Saving the model reasoning information of the participating nodes and / or the reasoning identifier; Receiving the inference identifier sent by the coordination node; Participate in retraining of the model based on the inference identifier.

8. The method according to claim 7, characterized in that Participating in retraining the model based on the inference identifier includes: The model reasoning information of the participating nodes is obtained, and the model reasoning information of the participating nodes and / or the reasoning identifier are returned to the coordination node.

9. The method according to claim 8, characterized in that Saving the model reasoning information of the participating nodes and / or the reasoning identifier includes: Sending the model reasoning information of the participating node and / or the reasoning identifier to the storage node; or, The model reasoning information of the participating nodes and / or the reasoning identifier are stored locally.

10. The method according to claim 9, characterized in that Obtaining model reasoning information of the participating nodes, including: Sending a request message to the storage node, carrying the reasoning identifier, and receiving the model reasoning information of the participating node returned by the storage node; or, The model reasoning information of the participating nodes is obtained locally.

11. The method according to any one of claims 8 to 10, characterized in that: The model reasoning information of the participating nodes includes at least one of the following: Inference identification; Model identification; Local data; Federated learning related logo.

12. A federated learning method, characterized in that: Applied to a storage node, the method comprises: Receiving a reasoning identifier from a coordinating node and / or model reasoning information of the coordinating node; and / or receiving a reasoning identifier from a participating node and / or model reasoning information of the participating node; Receiving a request message sent by the coordinating node, carrying the reasoning identifier; and / or a request message sent by the participating node, carrying the reasoning identifier; Returning the model reasoning information of the coordinating node to the coordinating node; and / or returning the model reasoning information of the participating node to the participating nodes.

13. A federated learning system, characterized in that: include: Coordinating nodes, participating nodes and storage nodes; The coordination node is used to execute the federated learning method according to any one of claims 1 to 6; The participating node is used to execute the federated learning method according to any one of claims 7 to 11; The storage node is used to execute the federated learning method described in claim 12.

14. A federated learning device, characterized in that: Applied to the coordination node, including: A first sending module is used to send a reasoning identifier to a participating node during a model reasoning process; A first saving module, used for saving the model reasoning information of the coordination node and / or the reasoning identifier; A second sending module is used to send the reasoning identifier to the participating node when judging or determining that the model needs to be retrained; A processing module, used for receiving the model reasoning information and / or the reasoning identification returned by the participating node; and / or obtaining the model reasoning information and / or the reasoning identification of the coordinating node; The first training module is used to initiate retraining of the model.

15. A federated learning device, characterized in that: Applicable to participating nodes, including: A first receiving module, configured to receive an inference identifier from a coordination node; A second saving module, used for saving the model reasoning information of the participating nodes and / or the reasoning identifier; A second receiving module, configured to receive the inference identifier sent by the coordination node; The second training module is used to participate in retraining of the model based on the inference identifier.

16. A federated learning device, characterized in that: Applicable to storage nodes, including: A third receiving module is used to receive the reasoning identification from the coordinating node and / or the model reasoning information of the coordinating node; and / or receive the reasoning identification from the participating node and / or the model reasoning information of the participating node; A fourth receiving module, configured to receive a request message sent by the coordinating node, carrying the reasoning identifier; and / or a request message sent by the participating node, carrying the reasoning identifier; The third sending module is used to return the model reasoning information of the coordinating node to the coordinating node; and / or return the model reasoning information of the participating node to the participating node.

17. An electronic device comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the method of any one of claims 1-6, any one of claims 7-11 or any one of claim 12.

18. A computer-readable storage medium having a computer program / instruction stored thereon, characterized in that: When the computer program / instructions are executed by a processor, the steps of the method described in any one of claims 1 to 6, any one of claims 7 to 11 or any one of claim 12 are implemented.

19. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the method described in any one of claims 1 to 6, any one of claims 7 to 11 or any one of claim 12 are implemented.

Citation Information

Cited By

  • Communication method, apparatus and system

    WO2026184354A1