Federal learning method, server, client and system

Abstract

The invention relates to a federal learning method, a server, a client and a system, and relates to the technical field of communication. The federated learning method is executed by a federated learning server, and comprises the following steps: sending a reasoning request to a plurality of clients of federated learning to instruct the plurality of clients to perform reasoning, the reasoning request comprising identifiers of a plurality of models used by the plurality of clients; receiving intermediate reasoning results sent by a plurality of clients; and generating a reasoning result based on the intermediate reasoning result.

Description

Technical Field

[0001] The present disclosure relates to the field of communication technology, and in particular to a method, server, client, and system for federated learning. Background Art

[0002] With the widespread application of artificial intelligence technology in the field of communications, network elements for artificial intelligence model reasoning and training have been introduced into the core network, such as the Network Data Analytics Function (NWDAF). Distributed collaborative training between NWDAFs can fully utilize the computing power of the network. The distributed federated learning method also has the advantage of being able to fully protect data privacy, so it has attracted much attention in artificial intelligence research. Summary of the invention

[0003] A technical problem to be solved by the present disclosure is: how to improve the accuracy of the analysis service provided by the network.

[0004] According to a first aspect of some embodiments of the present disclosure, a method for federated learning is provided, which is executed by a federated learning server, including: sending an inference request to multiple clients of the federated learning to instruct the multiple clients to perform inference, wherein the inference request includes identifiers of multiple models used by the multiple clients; receiving intermediate inference results sent by the multiple clients; and generating an inference result based on the intermediate inference result.

[0005] In some embodiments, the federated learning method also includes: receiving a sample alignment request sent by each of a plurality of clients, wherein the sample alignment request includes an internal identifier of the analysis target in each client; generating a sample alignment result for each client based on the internal identifier and an external identifier of the analysis target; for each client, sending the sample alignment result corresponding to the client to the client, so that the client generates an intermediate inference result based on the sample alignment result.

[0006] In some embodiments, generating an inference result based on the intermediate inference results includes: determining a weight of each client based on an identification of a model used by each of the multiple clients; and aggregating the intermediate inference results based on the weight to generate the inference result.

[0007] In some embodiments, receiving intermediate inference results sent by multiple clients includes: receiving inference responses sent by the multiple clients through a target client among the multiple clients, wherein the inference response includes the intermediate inference result of each client among the multiple clients and an identifier of the model used.

[0008] In some embodiments, the method of federated learning also includes: receiving an analysis request, wherein the analysis request includes an analysis identifier and analysis filter information of the federated learning; determining multiple models used in the federated learning based on the analysis identifier and the analysis filter information, wherein the multiple models are trained through multiple training clients; and determining multiple clients for reasoning from multiple training clients based on the multiple models used in the federated learning, the analysis filter information, and the capability information of the client.

[0009] In some embodiments, the method of federated learning further includes: sending an analysis response to the sender of the analysis request, wherein the analysis response includes an inference result.

[0010] In some embodiments, the sender of the analysis request is a NWDAF including an analysis logic function AnLF.

[0011] In some embodiments, the server is an application function (AF) or an NWDAF including a model training logic function MTLF, and the client is an AF or a NWDAF.

[0012] In some embodiments, the federated learning is vertical federated learning, and the multiple clients have the same sample space, but the feature spaces of the sample spaces of the multiple clients are different.

[0013] According to a second aspect of some embodiments of the present disclosure, a method for federated learning is provided, which is executed by a federated learning client, including: receiving an inference request sent by a federated learning server, wherein the inference request includes an identifier of a model used by the client; generating an intermediate inference result based on the model used by the client, and sending the intermediate inference result to the server.

[0014] In some embodiments, generating an intermediate reasoning result based on a model used by a client includes: sending a sample alignment request to a server, wherein the sample alignment request includes an internal identifier of an analysis target; receiving a sample alignment result sent by the server; and generating an intermediate reasoning result based on the sample alignment result and the model used by the client.

[0015] In some embodiments, sending the intermediate reasoning results to the server includes: receiving the intermediate reasoning results and the identification of the model used sent by each client other than the client that performs reasoning; sending an inference response to the server, wherein the inference response includes the intermediate reasoning results generated by the client and the identification of the model used, and the intermediate reasoning results and the identification of the model used by each client among the other clients.

[0016] In some embodiments, the method of federated learning further includes: determining whether there is local data for performing reasoning based on the reasoning request; and if there is no local data for performing reasoning, performing data collection to obtain data for performing reasoning.

[0017] In some embodiments, the server is an application function AF or a NWDAF including a model training logic function MTLF, and the client is an AF or a NWDAF.

[0018] According to a third aspect of some embodiments of the present disclosure, a server for federated learning is provided, comprising: a sending module, configured to send inference requests to multiple clients of federated learning to instruct the multiple clients to perform inference, wherein the inference requests include identifiers of multiple models used by the multiple clients; a receiving module, configured to receive intermediate inference results sent by the multiple clients; and a generating module, configured to generate inference results based on the intermediate inference results.

[0019] According to a fourth aspect of some embodiments of the present disclosure, a federated learning client is provided, comprising: a receiving module configured to receive an inference request sent by a federated learning server, wherein the inference request includes an identifier of a model used by the client; a sending module configured to generate an intermediate inference result based on the model used by the client, and send the intermediate inference result to the server.

[0020] According to a fifth aspect of some embodiments of the present disclosure, a federated learning system is provided, comprising: a federated learning server as described above; and a client as described above.

[0021] According to a sixth aspect of some embodiments of the present disclosure, there is provided an electronic device, comprising: a processor; and a memory coupled to the processor, for storing instructions, which, when executed by the processor, causes the processor to execute the method of federated learning as described above.

[0022] According to a seventh aspect of some embodiments of the present disclosure, a computer-readable storage medium is provided, on which computer instructions are stored, wherein the instructions, when executed by a processor, implement the method of federated learning as described above.

[0023] According to an eighth aspect of some embodiments of the present disclosure, a computer program product is provided, comprising instructions, which, when executed by a processor, cause the processor to perform the method of federated learning as described above.

[0024] The present disclosure allows multiple clients of federated learning to participate in reasoning, thereby being able to utilize the characteristics of local models and data of multiple clients, which can improve the accuracy of network reasoning, that is, improve the accuracy of analysis services provided by the network.

[0025] Other features and advantages of the present disclosure will become apparent from the following detailed description of exemplary embodiments of the present disclosure with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the embodiments of the present disclosure or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0027] Figure 1 A flowchart of a method for federated learning according to some embodiments of the present disclosure is shown.

[0028] Figure 2 A flowchart of a method for federated learning according to some other embodiments of the present disclosure is shown.

[0029] Figure 3 A flowchart of a method for federated learning according to some further embodiments of the present disclosure is shown.

[0030] Figure 4 A server for federated learning according to some embodiments of the present disclosure is shown.

[0031] Figure 5 A client of federated learning according to some embodiments of the present disclosure is shown.

[0032] Figure 6 A schematic diagram of the structure of a federated learning system according to some embodiments of the present disclosure is shown.

[0033] Figure 7 A schematic structural diagram of an electronic device according to some embodiments of the present disclosure is shown.

[0034] Figure 8 A schematic structural diagram of an electronic device according to some other embodiments of the present disclosure is shown. DETAILED DESCRIPTION

[0035] The technical solutions 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 following description of at least one exemplary embodiment is actually only illustrative and is by no means intended to limit the present disclosure and its application or use. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present disclosure.

[0036] Federated learning includes two methods: Horizontal Federated Learning (HFL) and Vertical Federated Learning (VFL). HFL is a method where multiple clients share similar data sets and jointly train a model while maintaining data privacy. In VFL, clients have data sets with different feature spaces but the same sample space, and they jointly train their own models while protecting the data from leaving the local environment.

[0037] 3GPP introduced HFL from R18, while VFL is still under research and the relevant processes have not been standardized. Since VFL allows different network entities, i.e. clients, to achieve collaborative training by sharing the intermediate training results of the model rather than directly sharing the model while maintaining data localization, it is of great significance to solve the data privacy protection and data island problems in 5G networks and improve the level of network intelligence.

[0038] Based on this, the present disclosure proposes a method of federated learning.

[0039] Figure 1 FIG. 2 shows a flow chart of a method for federated learning according to some embodiments of the present disclosure. Figure 1 As shown, the method of this embodiment is executed by a federated learning server, and includes steps S102 to S106.

[0040] In step S102, an inference request is sent to multiple clients of federated learning to instruct the multiple clients to perform inference, wherein the inference request includes identifiers of multiple models used by the multiple clients.

[0041] Federated learning is vertical federated learning, that is, multiple clients have the same sample space, and the feature spaces of the sample spaces of multiple clients are different. That is, reasoning is performed through multiple clients, so that the local models and data of multiple clients can be utilized. In other words, through the collaborative reasoning of the VFL client and the VFL server, the reasoning results can maximize the reflection of the characteristics of different models and client data in the distributed training process, so that the intelligent collaboration benefits between network elements can be fully utilized, thereby improving the accuracy of analysis services, that is, improving the artificial intelligence performance of the network.

[0042] Currently, NWDAF is decomposed into the Analytics Logical Function (AnLF) and the Model Training Logical Function (MTLF). AnLF can provide analysis services based on consumer requests, and MTLF can provide model training services based on consumer requests. NWDAF can include both AnLF and MTLF.

[0043] In some embodiments, the server is an application function AF or a NWDAF including a model training logical function (Model Training Logical Function, abbreviated as: MTLF), and the client is an AF or a NWDAF.

[0044] In other words, the server participates in the training process of vertical federated learning. That is, the VFL server can perform the inference process after the training process is completed. Since the VFL server participates in the training process, the VFL server can obtain information about the VFL training client participating in the training process, such as the region where the VFL training client is located and the information about the model used.

[0045] In some embodiments, after the training process is completed, the VFL server, in response to receiving an analysis request, determines to start a distributed VFL reasoning program according to the analysis request, and generates a more accurate analysis result together with the VFL client.

[0046] In previous standards, NWDAF including AnLF is generally used for model reasoning, but AnLF does not participate in model training and cannot obtain different local models of different clients, so the reasoning accuracy is low. The collaborative reasoning method proposed in the present disclosure by the VFL server (MTLF or AF) and the VFL client can utilize different local models of different clients and improve the reasoning accuracy.

[0047] In some embodiments, the method of federated learning also includes: receiving an analysis request, wherein the analysis request includes an analysis identifier and analysis filter information of the federated learning; determining multiple models used in the federated learning based on the analysis identifier and the analysis filter information, wherein the multiple models are trained through multiple training clients; and determining multiple clients for reasoning from multiple training clients based on the multiple models used in the federated learning, the analysis filter information, and the capability information of the client.

[0048] In some embodiments, the sender of the analysis request is a NWDAF that includes an analysis logic function (Analytics Logical Function, referred to as AnLF). That is, the user of the analysis service, such as a network user, sends an analysis demand or subscription to the NWDAF that includes AnLF, and the NWDAF that includes AnLF sends an analysis request to the VFL server, and the VFL server generates an inference result through multiple clients participating in the inference. Finally, the NWDAF that includes AnLF receives the inference result sent by the VFL server, and then forwards the inference result to the user of the analysis service.

[0049] In some embodiments, the user of the analysis service sends an analysis demand or subscription to the NWDAF including the AnLF through the Nnwdaf_AnalyticsSubs cription_Subscribe service. The analysis demand or subscription includes an analysis identifier, analysis filter information, an analysis period, and an external identifier of the analysis target (e.g., an external identifier of a user device). The analysis period indicates the duration of the analysis demand or subscription.

[0050] If the NWDAF containing MTLF is a VFL server, the NWDAF containing AnLF uses the Nnwdaf_AnalyticsSubscription_Subscribe service to send an analysis request to the VFL server NWDAF, where the analysis request includes an analysis identifier, analysis filter information, and an external identifier of an analysis target.

[0051] If the trusted AF is a VFL server, the NWDAF including the AnLF sends a request to the Network Exposure Function (NEF) through the Nnef_AnalyticsExposure_Subscribe service to send an analysis request to the VFL server AF through the Naf_EventExposure_Subscribe service of the NEF, and the analysis request includes an analysis identifier, analysis filter information, and an external identifier of the analysis target. In some embodiments, when the AF is a VFL server, the analysis identifier can be replaced by an event identifier corresponding to the analysis identifier.

[0052] The analysis identifier is associated with the training process of federated learning. In some embodiments, the analysis identifier is the identifier used in the training process of federated learning. That is, the server can determine multiple models used in federated learning based on the analysis identifier, and further determine multiple training clients based on the multiple models.

[0053] The analysis filter information is used to provide information for this analysis request, so that the server can collect data and select models. The analysis filter information includes, for example, network slice information, target area information, etc.

[0054] Since the VFL server and VFL client have registered their corresponding NF configuration files, analysis identifiers, address information, service areas, and capability information with the Network Repository Function (NRF), where the capability information indicates whether it supports becoming a VFL server or a VFL client, the VFL server can determine the information of the model required for this analysis request based on the analysis identifier and analysis filter information, that is, multiple models used for vertical federated learning, thus preliminarily determining multiple training clients that can be used for reasoning. Further, combined with the analysis filter information and the client's capability information, multiple clients for reasoning are determined from these multiple training clients.

[0055] In step S104, intermediate inference results sent by multiple clients are received.

[0056] In some embodiments, after sending inference requests to multiple clients, a sample alignment request sent by each of the multiple clients is received, wherein the sample alignment request includes an internal identifier of the analysis target in each client; a sample alignment result for each client is generated based on the internal identifier and an external identifier of the analysis target; for each client, the sample alignment result corresponding to the client is sent to the client, so that the client generates an intermediate inference result based on the sample alignment result.

[0057] Since the final analysis result of the analysis target is obtained through reasoning through multiple clients, and the internal identification of the analysis target in multiple clients may be different from the external identification, sample alignment enables the client to accurately use the data related to the analysis target for reasoning. The external identification of the analysis target is obtained by the server through an analysis request, for example, the user of the analysis service informs the NWDAF containing AnLF, the NWDAF containing AnLF informs the server, and the server informs the client.

[0058] Through sample alignment, it is possible to perform reasoning by combining shared data (i.e., data related to the analysis target) without exposing the client's original data. This allows the reasoning process to adopt local models of multiple clients to improve the accuracy of reasoning, while also maintaining the privacy of the client and improving the security of reasoning.

[0059] In some embodiments, receiving intermediate inference results sent by multiple clients includes: receiving inference responses sent by the multiple clients through a target client among the multiple clients, wherein the inference response includes the intermediate inference result of each client among the multiple clients and an identifier of the model used.

[0060] That is to say, each of the multiple clients can send the intermediate inference results and the identifier of the model used to the server, or select a target client from the multiple clients, and other clients send the intermediate inference results and the identifier of the model used to the target client, and the target client sends the intermediate inference results and the identifier of the model used of all clients to the server. In this way, after receiving the inference response sent by the target client, the server can directly start the process of generating the inference result, reducing the waiting time between receiving the intermediate inference result sent by the first client and receiving the intermediate inference result sent by the last client, which helps to reduce the process and improve efficiency. That is, VFL training and inference do not involve model transmission, and can be compatible with network elements from different manufacturers.

[0061] In step S106 , an inference result is generated based on the intermediate inference result.

[0062] In some embodiments, generating an inference result based on the intermediate inference results includes: determining a weight of each client based on an identification of a model used by each of the multiple clients; and aggregating the intermediate inference results based on the weight to generate the inference result.

[0063] The weight of the client used when the server generates the inference result is determined according to the training process. That is, the server determines the multiple models used in federated learning based on the analysis identifier, and can also determine the weights of these multiple models during training. When the server generates the inference result, it can directly use the weights during training, or it can perform secondary weighting on the weights during training based on the analysis filter information. For example, if the analysis filter information indicates the target area information, then the weight of the client in the target area of ​​the service area can be increased.

[0064] In some embodiments, the method of federated learning further includes sending an analysis response to the sender of the analysis request, wherein the analysis response includes an inference result. For example, the server sends the inference result to the NWDAF including An LF through Nnwdaf_AnalyticsSubscription_Notify (when the server is NWDAF including MTLF) or Naf_EventExposure_Notify (when the server is AF), and the NWDAF including AnLF sends the inference result to the user of the analysis service through Nnwdaf_AnalyticsInfo_Request or Nnwdaf_AnalyticsSubscription_Notify.

[0065] The federated learning method disclosed in the present invention involves multiple clients in the federated learning inference, thereby being able to utilize the local models and data features of multiple clients, thereby improving the accuracy of the network's reasoning, that is, improving the accuracy of the analysis service provided by the network.

[0066] This paper proposes a new distributed reasoning process that can adapt to the characteristics of VFL, effectively promote 5GC's comprehensive support for VFL, and promote the implementation of federated learning technology in a wider range of application scenarios.

[0067] The following describes the method of federated learning disclosed in the present invention based on the client side. Figure 2 FIG. 2 shows a flow chart of a method for federated learning according to some other embodiments of the present disclosure. Figure 2 As shown, the method of this embodiment is executed by a client of federated learning, and includes steps S202 to S204.

[0068] In step S202, an inference request sent by a federated learning server is received, wherein the inference request includes an identifier of a model used by the client.

[0069] In some embodiments, the server is an application function AF or a NWDAF including a model training logic function MTLF, and the client is an AF or a NWDAF.

[0070] In some embodiments, the method of federated learning further includes: determining whether there is local data for participating in reasoning based on the reasoning request; and if there is no local data for participating in reasoning, performing data collection to obtain data for participating in reasoning.

[0071] In step S204, an intermediate inference result is generated according to the model used by the client, and the intermediate inference result is sent to the server.

[0072] After obtaining the data for inference, a sample alignment request is sent to the server to determine the data (i.e., feature space) relevant to the analysis target.

[0073] In some embodiments, the model used by the client generates intermediate inference results, including: sending a sample alignment request to the server, wherein the sample alignment request includes an internal identifier of the analysis target; receiving the sample alignment result sent by the server; and generating an intermediate inference result based on the sample alignment result and the model used by the client.

[0074] In some embodiments, sending the intermediate reasoning results to the server includes: receiving the intermediate reasoning results and the identification of the model used sent by each client other than the client that performs reasoning; sending an inference response to the server, wherein the inference response includes the intermediate reasoning results generated by the client and the identification of the model used, and the intermediate reasoning results and the identification of the model used by each client among the other clients.

[0075] The federated learning method disclosed in the present invention can realize collaborative training between different network elements in the core network through VFL without data and model sharing. Through VFL distributed reasoning, the accuracy of reasoning results is improved. It can be used to analyze user behavior and network performance data in real time, intelligently adjust resource allocation, optimize network traffic management, and improve network efficiency and user experience.

[0076] Figure 3 FIG. 2 shows a flow chart of a method for federated learning according to some other embodiments of the present disclosure. Figure 3 As shown, the method of this embodiment includes steps S301 to S309.

[0077] In step S301, the user of the analysis service sends an analysis request or subscription to the NWDAF including the AnLF, where the analysis request or subscription includes an analysis identifier, analysis filter information, an analysis cycle, and an external identifier of an analysis target (eg, an external identifier of a UE).

[0078] In step S302a, when the VFL server is the NWDAF including the MTFL, the NWDAF including the AnLF sends an analysis request to the VFL server, where the analysis request includes an analysis identifier, analysis filter information, and an external identifier of an analysis target.

[0079] In step S302b-1, when the VFL server is the AF, the NWDAF including the AnLF sends an analysis request to the NEF.

[0080] In step S302b-2, the NEF sends an analysis request to the VFL server.

[0081] In step S303 , inference requests are sent to multiple VFL clients (N VFL clients are shown in the figure) through steps S301 - 1 to S301 -N respectively, and the inference requests include identifiers of multiple models used by the multiple clients.

[0082] In step S304, the VFL client collects data through a data provider (eg, a network function).

[0083] In step S305 , the VFL client performs inference based on the local model.

[0084] In step S306 , the N VFL clients send reasoning responses to the VFL server through steps S306 - 1 to S306 -N respectively, where the reasoning responses include intermediate reasoning results and an identifier of the model used.

[0085] In step S307 , the VFL server aggregates the intermediate inference results of the N VFL clients based on the models used by the N VFL clients.

[0086] In step S308a, when the VFL server is the NWDAF including the MTFL, the VFL server sends the inference result to the NWDAF including the MTFL.

[0087] In step S308b-1, when the VFL server is the AF, the VFL server sends the inference result to the NEF.

[0088] In step S308b-2, the NEF sends the inference result to the VFL server.

[0089] In step S309 , the VFL server sends the inference result to the user of the analysis service.

[0090] Figure 4 FIG. 4 shows a server for federated learning according to some embodiments of the present disclosure. Figure 4 As shown, the federated learning server 40 includes:

[0091] A sending module 410 is configured to send an inference request to multiple clients of federated learning to instruct the multiple clients to perform inference, wherein the inference request includes identifiers of multiple models used by the multiple clients;

[0092] The receiving module 420 is configured to receive intermediate inference results sent by multiple clients;

[0093] The generating module 430 is configured to generate a reasoning result based on the intermediate reasoning result.

[0094] In some embodiments, the federated learning server 40 is also configured to receive a sample alignment request sent by each of a plurality of clients, wherein the sample alignment request includes an internal identifier of the analysis target in each client; generate a sample alignment result for each client based on the internal identifier and an external identifier of the analysis target; and for each client, send the sample alignment result corresponding to the client to the client, so that the client generates an intermediate inference result based on the sample alignment result.

[0095] In some embodiments, the generation module 430 is configured to determine a weight of each client based on an identification of a model used by each of the multiple clients; and aggregate the intermediate inference results based on the weight to generate an inference result.

[0096] In some embodiments, the receiving module 420 is configured to receive reasoning responses sent by multiple clients through a target client among the multiple clients, wherein the reasoning response includes an intermediate reasoning result of each client among the multiple clients and an identification of a used model.

[0097] In some embodiments, the federated learning server 40 is also configured to receive an analysis request, wherein the analysis request includes an analysis identifier and analysis filter information of the federated learning; determine multiple models used in the federated learning based on the analysis identifier and the analysis filter information, and the multiple models are trained through multiple training clients; and determine multiple clients for reasoning from multiple training clients based on the multiple models used in the federated learning, the analysis filter information, and the capability information of the client.

[0098] In some embodiments, the federated learning server 40 is further configured to send an analysis response to the sender of the analysis request, wherein the analysis response includes an inference result.

[0099] In some embodiments, the sender of the analysis request is a NWDAF including an analysis logic function AnLF.

[0100] In some embodiments, the server is an application function AF or a NWDAF including a model training logic function MTLF, and the client is an AF or a NWDAF.

[0101] In some embodiments, the federated learning is vertical federated learning, and the multiple clients have the same sample space, but the feature spaces of the sample spaces of the multiple clients are different.

[0102] Figure 5 FIG. 4 shows a client of federated learning according to some embodiments of the present disclosure. Figure 5 As shown, the client 50 of the federated learning includes:

[0103] A receiving module 510 is configured to receive an inference request sent by a federated learning server, wherein the inference request includes an identifier of a model used by the client;

[0104] The sending module 520 is configured to generate an intermediate reasoning result according to the model used by the client, and send the intermediate reasoning result to the server.

[0105] In some embodiments, the sending module 520 is configured to send a sample alignment request to the server, wherein the sample alignment request includes an internal identifier of the analysis target; receive the sample alignment result sent by the server; and generate an intermediate inference result based on the sample alignment result and the model used by the client.

[0106] In some embodiments, the sending module 520 is configured to receive intermediate reasoning results and identification of the model used sent by each client other than the client that performs reasoning; and send an inference response to the server, wherein the inference response includes the intermediate reasoning results generated by the client and identification of the model used, and the intermediate reasoning results and identification of the model used by each client among the other clients.

[0107] In some embodiments, the client 50 of federated learning is further configured to determine whether there is local data for participating in reasoning based on the reasoning request; if there is no local data for participating in reasoning, perform data collection to obtain data for participating in reasoning.

[0108] In some embodiments, the server is an application function AF or a NWDAF including a model training logic function MTLF, and the client is an AF or a NWDAF.

[0109] Figure 6 FIG. 2 shows a schematic diagram of the structure of a federated learning system according to some embodiments of the present disclosure. Figure 6 As shown, the federated learning system 60 includes the federated learning server 610 as described above; and the federated learning client 620 as described above.

[0110] The server or client of the federated learning in the embodiments of the present disclosure can be implemented by various computing devices or computer systems. Figure 7 as well as Figure 8 Give a description.

[0111] Figure 7 FIG. 1 shows a schematic diagram of the structure of an electronic device according to some embodiments of the present disclosure. Figure 7 As shown, the electronic device 70 of this embodiment includes: a memory 710 and a processor 720 coupled to the memory 710, and the processor 720 is configured to execute the method of federated learning in any of some embodiments of the present disclosure based on instructions stored in the memory 710.

[0112] The memory 710 may include, for example, a system memory, a fixed non-volatile storage medium, etc. The system memory may store, for example, an operating system, an application program, a boot loader, a database, and other programs.

[0113] Figure 8 FIG. 2 shows a schematic diagram of the structure of an electronic device according to some other embodiments of the present disclosure. Figure 8 As shown, the electronic device 80 of this embodiment includes: a memory 810 and a processor 820, which are similar to the memory 710 and the processor 720 respectively. It can also include an input and output interface 830, a network interface 840, a storage interface 850, etc. These interfaces 830, 840, 850 and the memory 810 and the processor 820 can be connected, for example, via a bus 860. Among them, the input and output interface 830 provides a connection interface for input and output devices such as a display, a mouse, a keyboard, and a touch screen. The network interface 840 provides a connection interface for various networked devices, for example, it can be connected to a database server or a cloud storage server. The storage interface 850 provides a connection interface for external storage devices such as SD cards and USB flash drives.

[0114] An embodiment of the present disclosure further provides a computer-readable storage medium having computer instructions stored thereon, wherein when the program is executed by a processor, any one of the aforementioned federated learning methods is implemented.

[0115] An embodiment of the present disclosure also provides a computer program product, comprising instructions, which, when executed by a processor, cause the processor to perform any of the aforementioned federated learning methods.

[0116] Those skilled in the art will appreciate that the embodiments of the present disclosure may be provided as methods, systems, or computer program products. Therefore, the present disclosure may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present disclosure may take the form of a computer program product implemented on one or more computer-usable non-transient storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0117] The present disclosure is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present disclosure. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0118] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0119] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0120] The above description is only a preferred embodiment of the present disclosure and is not intended to limit the present disclosure. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present disclosure should be included in the protection scope of the present disclosure.

Claims

1. A method for federated learning, performed by a federated learning server, comprising: Sending an inference request to multiple clients of federated learning to instruct the multiple clients to perform inference, wherein the inference request includes identifiers of multiple models used by the multiple clients; Receiving intermediate inference results sent by the multiple clients; An inference result is generated based on the intermediate inference result.

2. The method according to claim 1, further comprising: Receiving a sample alignment request sent by each of the multiple clients, wherein the sample alignment request includes an internal identifier of an analysis target in each of the clients; Generate a sample alignment result for each client based on the internal identifier and the external identifier of the analysis target; For each of the clients, the sample alignment result corresponding to the client is sent to the client, so that the client generates the intermediate inference result based on the sample alignment result.

3. The method according to claim 1, wherein: Generating the inference result based on the intermediate inference result comprises: determining a weight for each client based on an identification of a model used by each of the plurality of clients; The intermediate inference results are aggregated based on the weights to generate an inference result.

4. The method according to claim 1, wherein: The receiving the intermediate inference results sent by the multiple clients includes: An inference response is received, which is sent by the multiple clients through a target client among the multiple clients, wherein the inference response includes an intermediate inference result of each client among the multiple clients and an identifier of a used model.

5. The method according to any one of claims 1 to 4, further comprising: Receiving an analysis request, wherein the analysis request includes an analysis identifier and analysis filter information of the federated learning; Determining multiple models used in the federated learning based on the analysis identifier and the analysis filter information, wherein the multiple models are trained by multiple training clients; Based on the multiple models used in the federated learning, the analysis filter information, and the capability information of the client, multiple clients for performing inference are determined from the multiple training clients.

6. The method according to claim 5, further comprising: An analysis response is sent to the sender of the analysis request, wherein the analysis response includes the inference result.

7. The method according to claim 5, wherein: The sender of the analysis request is the NWDAF including the analysis logic function AnLF.

8. The method according to claim 1 or 7, wherein: The server is an application function AF or an NWDAF including a model training logic function MTLF, and the client is an AF or an NWDAF.

9. The method according to claim 1, wherein: The federated learning is vertical federated learning, the multiple clients have the same sample space, and the feature spaces of the sample spaces of the multiple clients are different.

10. A method for federated learning, performed by a client of federated learning, comprising: Receiving an inference request sent by a federated learning server, wherein the inference request includes an identifier of a model used by the client; An intermediate inference result is generated according to the model used by the client, and the intermediate inference result is sent to the server.

11. According to the method of claim 10, generating an intermediate inference result according to the model used by the client comprises: Sending a sample alignment request to the server, wherein the sample alignment request includes an internal identifier of an analysis target; Receiving the sample alignment result sent by the server; An intermediate inference result is generated based on the sample alignment result and the model used by the client.

12. The method according to claim 10, wherein: The sending the intermediate reasoning result to the server comprises: receiving an intermediate inference result and an identifier of a used model sent by each of the other clients except the client that performs inference; An inference response is sent to the server, wherein the inference response includes the intermediate inference result generated by the client and the identifier of the model used, and the intermediate inference result and the identifier of the model used by each of the other clients.

13. The method according to any one of claims 10 to 12, further comprising: Determining whether there is local data for performing reasoning based on the reasoning request; In the absence of local data for performing reasoning, data collection is performed to obtain data for performing reasoning.

14. The method according to any one of claims 10 to 12, wherein: The server is an application function AF or an NWDAF including a model training logic function MTLF, and the client is an AF or an NWDAF.

15. A federated learning server, comprising: A sending module, configured to send an inference request to multiple clients of federated learning to instruct the multiple clients to perform inference, wherein the inference request includes identifiers of multiple models used by the multiple clients; A receiving module, configured to receive the intermediate reasoning results sent by the multiple clients; A generating module is configured to generate an inference result based on the intermediate inference result.

16. A federated learning client, comprising: A receiving module, configured to receive an inference request sent by a federated learning server, wherein the inference request includes an identifier of a model used by the client; The sending module is configured to generate an intermediate reasoning result according to the model used by the client and send the intermediate reasoning result to the server.

17. A federated learning system, comprising: The federated learning server as claimed in claim 15; and The client of federated learning as claimed in claim 16.

18. An electronic device, comprising: processor; as well as A memory coupled to the processor, for storing instructions, wherein when the instructions are executed by the processor, the processor executes the method for federated learning as claimed in any one of claims 1 to 14.

19. A computer-readable storage medium having computer instructions stored thereon, wherein: When the instruction is executed by a processor, the method for federated learning as described in any one of claims 1 to 14 is implemented.

20. A computer program product comprising instructions, which, when executed by a processor, cause the processor to perform the method of federated learning according to any one of claims 1 to 14.

Images