Federal learning method, enabling server, enabling client and federal learning system

Abstract

The invention provides a federated learning method, an enabling server, an enabling client and a federated learning system. The federated learning method executed by the enabling server comprises the following steps: receiving a training request for training a federated learning model; sending an evaluation request for evaluating the training ability to a plurality of participants participating in training of the federated learning model, so as to obtain the training ability of each of the plurality of participants; determining whether a preset training condition is satisfied according to the training capability of each of the plurality of participants; in response to the situation that the preset training condition is not met, generating a corresponding weight parameter for each participant; and sending the corresponding weight parameter to each participant, so that each participant performs weighting processing on a local training sample based on the corresponding weight parameter and performs training based on the weighted training sample.

Description

Technical Field

[0001] The present disclosure relates to the field of communication technology, and more specifically, to a federated learning method, an enabling server, an enabling client, and a federated learning system. Background Art

[0002] Machine Learning (ML) refers to analyzing and learning input data through specific algorithms to make decisions or predictions about things. It is mainly divided into three types, including supervised learning, unsupervised learning, and reinforcement learning. Federation Learning (FL) is a distributed machine learning method that aims to protect data privacy while using data distributed in different places to train models.

[0003] With the vigorous development of information technology, the demand for federated learning services is increasing. In some networks, the support capacity for federated learning services is limited, and even the training of federated learning models cannot be completed. Summary of the invention

[0004] A brief overview of the disclosure is given below in order to provide a basic understanding of some aspects of the disclosure. However, it should be understood that this overview is not an exhaustive overview of the disclosure. It is not intended to identify the key or important parts of the disclosure, nor is it intended to limit the scope of the disclosure. Its purpose is simply to give some concepts of the disclosure in a simplified form as a prelude to a more detailed description given later.

[0005] One of the objectives of the present disclosure is to provide a federated learning method, an enabling server, an enabling client, and a federated learning system.

[0006] According to a first aspect of the present disclosure, there is provided a federated learning method performed by an enabling server, comprising: receiving a training request for training a federated learning model; sending an evaluation request for evaluating training capabilities to multiple participants participating in training the federated learning model to obtain the training capabilities of each of the multiple participants; determining whether a preset training condition is met based on the training capabilities of each of the multiple participants; in response to not meeting the preset training conditions, generating corresponding weight parameters for each participant; and sending the corresponding weight parameters to each participant, so that each participant performs weighted processing on local training samples based on the corresponding weight parameters and performs training based on the weighted training samples.

[0007] In some embodiments, determining whether a preset training condition is met based on the training capability of each of the multiple participants includes: for each participant, determining whether the training capability meets the model training requirements; counting the number of participants that meet the model training requirements; and in response to the number being less than or equal to a preset number, determining that the preset training condition is not met.

[0008] In some embodiments, generating corresponding weight parameters for each participant includes: for each participant, generating weight parameters according to the degree of matching between training capabilities and model training requirements.

[0009] In some embodiments, the federated learning method also includes: receiving training results of the federated learning model from each of the multiple participants; and sending the training results of each participant to the training server that initiated the training request, so that the training server summarizes the training results of each participant.

[0010] In some embodiments, the federated learning method further includes: selecting a user terminal in an available state from user terminals registered with the enabling server as a participant participating in training the federated learning model.

[0011] In some embodiments, each participant includes a training client and an enabling client, wherein sending an evaluation request for evaluating training capabilities to multiple participants participating in training the federated learning model includes: sending the evaluation request to the enabling client of each participant, so that the enabling client forwards the evaluation request to the training client, so that the training client sends the training capability to the enabling server via the enabling client when receiving the evaluation request.

[0012] In some embodiments, the federated learning method also includes: determining whether the training request includes a target parameter, the target parameter indicating that the request supports incomplete federated learning; in response to the training request including the target parameter, performing the operations of generating corresponding weight parameters for each participant and sending the corresponding weight parameters to each participant, so that each participant performs weighted processing on local training samples based on the corresponding weight parameters and performs training based on the weighted training samples; and in response to the training request not including the target parameter, forwarding the training request to each of the multiple participants, so that each participant performs training based on local training samples.

[0013] According to a second aspect of the present disclosure, there is provided a federated learning method performed by an enabling client, comprising: receiving weight parameters sent by an enabling server; sending the weight parameters to a training client, so that upon receiving the weight parameters, the training client performs weighted processing on local training samples based on the weight parameters and performs training based on the weighted training samples; and sending the training results of the training client to the enabling server.

[0014] In some embodiments, the federated learning method further includes: sending a registration request to the enabling server to register the training client, wherein the registration request includes configuration information of the training client.

[0015] In some embodiments, the configuration information includes an available time for the training client.

[0016] In some embodiments, the federated learning method further includes: receiving an evaluation request sent by the enabling server to evaluate the training capability of the training client; and sending the training capability of the training client to the enabling server.

[0017] According to a third aspect of the present disclosure, an enabling server is provided, comprising: a receiving module, the receiving module being configured to receive a training request for training a federated learning model; a sending module, the sending module being configured to send an evaluation request for evaluating training capabilities to multiple participants participating in the training of the federated learning model, so as to obtain the training capabilities of each of the multiple participants; a determining module, the determining module being configured to determine whether a preset training condition is met based on the training capabilities of each of the multiple participants; a generating module, the generating module being configured to generate corresponding weight parameters for each participant in response to not meeting the preset training conditions, wherein the receiving module is further configured to send the corresponding weight parameters to each participant, so that each participant performs weighted processing on local training samples based on the corresponding weight parameters and performs training based on the weighted training samples.

[0018] According to a fourth aspect of the present disclosure, an enabling client is provided, wherein the enabling server comprises: a receiving module, the receiving module being configured to receive weight parameters sent by the enabling server; and a sending module, the sending module being configured to: send the weight parameters to a training client, so that upon receiving the weight parameters, the training client performs weighted processing on local training samples based on the weight parameters and performs training based on the weighted training samples, and sends the training results of the training client to the enabling server.

[0019] According to a fifth aspect of the present disclosure, a federated learning system is provided, comprising: the enabling server as described above; and the enabling client as described above.

[0020] In some embodiments, the federated learning system further includes: a training server, which is configured to send a training request for training a federated learning model to the enabling server; and a training client, which is configured to weight local training samples based on weight parameters and perform training based on the weighted training samples to generate training results.

[0021] According to a sixth aspect of the present disclosure, a device for federated learning is provided, wherein the device includes a processor and a memory, and instructions are stored on the memory. When the instructions are executed by the processor, the steps of the federated learning method as described above are implemented.

[0022] According to a seventh aspect of the present disclosure, a non-transitory computer-readable storage medium is provided, on which instructions are stored. When the instructions are executed by a processor, the steps of the federated learning method as described above are implemented.

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

[0024] Other features and advantages of the present disclosure will become more apparent from the following detailed description of exemplary embodiments of the present disclosure with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The accompanying drawings, which constitute a part of the specification, illustrate embodiments of the present disclosure and, together with the description, serve to explain the principles of the present disclosure.

[0026] The present disclosure may be more clearly understood from the following detailed description with reference to the accompanying drawings, in which:

[0027] Figure 1 A schematic diagram of a process of a federated learning method according to an exemplary embodiment of the present disclosure is shown;

[0028] Figure 2 A schematic diagram of a process of determining whether a preset training condition is met in a federated learning method according to an exemplary embodiment of the present disclosure is shown;

[0029] Figure 3 A schematic diagram of the structure of an enabling server according to an exemplary embodiment of the present disclosure is shown;

[0030] Figure 4A schematic diagram of a process flow of a federated learning method according to another exemplary embodiment of the present disclosure is shown;

[0031] Figure 5 A schematic diagram of the structure of an enabling client according to an exemplary embodiment of the present disclosure is shown;

[0032] Figure 6 A structural schematic diagram of a federated learning system according to an exemplary embodiment of the present disclosure is shown;

[0033] Figure 7 A schematic diagram of a flow chart showing the interaction between a training client, an enabling client, an enabling server and a training server in a federated learning system according to an exemplary embodiment of the present disclosure is shown;

[0034] Figure 8 A schematic diagram of the structure of a device for federated learning according to an exemplary embodiment of the present disclosure is shown.

[0035] Note that in the embodiments described below, the same reference numerals are sometimes used in common between different drawings to represent the same parts or parts with the same functions, and their repeated descriptions are omitted. In this specification, similar reference numerals and letters are used to represent similar items, so once an item is defined in one drawing, it does not need to be further discussed in subsequent drawings.

[0036] For ease of understanding, the positions, sizes, and ranges of various structures shown in the drawings and the like sometimes do not represent actual positions, sizes, and ranges, etc. Therefore, the disclosed invention is not limited to the positions, sizes, and ranges, etc. disclosed in the drawings and the like. In addition, the drawings are not necessarily drawn to scale, and some features may be exaggerated to show details of specific components. DETAILED DESCRIPTION

[0037] Various exemplary embodiments of the present disclosure will be described in detail below with reference to the accompanying drawings. It should be noted that unless otherwise specifically stated, the relative arrangement of components and steps, numerical expressions and numerical values ​​set forth in these embodiments do not limit the scope of the present disclosure.

[0038] The following description of at least one exemplary embodiment is in fact merely illustrative and is in no way intended to limit the present disclosure and its application or use. That is, the structures and methods herein are shown in an exemplary manner to illustrate different embodiments of the structures and methods in the present disclosure. However, those skilled in the art will appreciate that they merely illustrate exemplary ways of the present disclosure that can be implemented, rather than exhaustive ways. In addition, the drawings need not be drawn to scale, and some features may be enlarged to illustrate the details of specific components.

[0039] Technologies, methods, and apparatus known to ordinary technicians in the relevant field may not be discussed in detail, but where appropriate, such technologies, methods, and apparatus should be considered part of the authorization specification.

[0040] In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limiting. Therefore, other examples of the exemplary embodiments may have different values.

[0041] In the core network based on the Third Generation Partnership Project, the federated learning service can be implemented through the Network Data Analytics Function (NWDAF) network element. Specifically, the federated learning process can be initialized through the NWDAF network element, and the training process of the federated learning model can be initialized through the Access and Mobility Management Function network element, the Session Management Function (SMF) network element and other network elements, and the model training requirements are sent to the network elements participating in the training so that each network element participating in the training completes the corresponding training, and then the corresponding network element sends the training results of each participant to the training server that initiates the training request for the federated learning model.

[0042] However, the network's support for federated learning services may be insufficient. For example, some participants may have insufficient training capabilities, resulting in the inability to train the model, or the training results may be of low accuracy, which results in low accuracy of the aggregated federated learning model. The services provided by the core network to the training server are limited and cannot meet the needs of the training server.

[0043] In order to solve the above problems, the present disclosure proposes a federated learning method, which provides support for federated learning based on an enabling server. When the training capabilities of some participants are insufficient and cannot meet the training conditions of the model, corresponding weight parameters are generated for each participant so that the participant can perform training based on the corresponding weight parameters, thereby ensuring that the model training can continue, improving the accuracy of the federated learning model, and improving the network's support capabilities for vertical federated learning, so that the network can provide more flexible federated learning services.

[0044] According to one aspect of the present disclosure, a federated learning method 100 that can be performed by an enabling server is provided, such as Figure 1 As shown, a federated learning method 100 according to an exemplary embodiment of the present disclosure may include:

[0045] Step S110, receiving a training request for training a federated learning model.

[0046] The training request may be initiated by a training server, for example. In some embodiments, the training request may include at least one of the model configuration information of the federated learning model to be trained and the model training requirements. In a specific example, the model training requirements may include at least one of the number of training samples, the quality of the training samples, and the accuracy of the training results, and the model configuration information may include the feature parameters of the federated learning model to be trained.

[0047] In some embodiments, the training request may include parameters for indicating whether to perform incomplete federated learning, so that the enabling server determines whether to start support for incomplete federated learning. In this article, incomplete federated learning refers to training a federated learning model when the training capacity of a participant is insufficient (for example, the number of training samples of a participant is small). In a specific example, the enabling server may determine whether the training request includes a target parameter indicating a request to support incomplete federated learning upon receiving a training request, and start support for incomplete federated learning when it is determined that the training request includes the target parameter, thereby generating corresponding weight parameters for each participant, as will be described in detail later. In this way, the training server may send a training request including the target parameter to the enabling server when it is determined based on prior information (for example, the historical training situation of each participant) that the current federated learning service requires the enabling server to provide support for incomplete federated learning.

[0048] like Figure 1 As shown, the federated learning method 100 according to an exemplary embodiment of the present disclosure may also include:

[0049] Step S120, sending an evaluation request for evaluating training capabilities to multiple participants participating in training the federated learning model, so as to obtain the training capability of each of the multiple participants.

[0050] In some embodiments, the training capability may include at least one of the number and quality of training samples. The quality of training samples may be characterized by the distribution of training samples (e.g., whether they satisfy the normal distribution) and statistical indicators (e.g., mean and variance). For example, if the distribution of training samples satisfies the normal distribution, the quality of the training samples may be considered good; for another example, the larger the variance of the training samples, the more dispersed the sample distribution, and accordingly, the worse the quality of the training samples.

[0051] In some embodiments, the enabling server may select a UE in an available state from the user equipment (UE) registered with the enabling server as a participant in training the federated learning model. The UE may provide an expected available time when registering so that the enabling server can determine whether it is in an available state.

[0052] The UE may include an enabling client and a training client, or in other words, the participants may include an enabling client and a training client. The training client is used to provide computing power and local training samples to complete the corresponding training, and the enabling client is used to provide support to the training client, for example, the agent training client registers with the enabling server, and for example, transmits messages from other devices on the network to the training client to reduce the working pressure of the training client, and in the case of insufficient transmission capacity of the training client, completing the transmission of messages via the enabling client can better transmit messages to other devices on the network.

[0053] In some embodiments, the enabling server sending an evaluation request for evaluating training capabilities to multiple participants participating in training a federated learning model may include: the enabling server sending an evaluation request to the enabling client of each participant to obtain the training capabilities of the training client of each participant. In a specific example, the training client may evaluate its own training capabilities in advance and store them on the corresponding enabling client, so that when the enabling client receives the evaluation request from the enabling server, the stored training capabilities of the training client can be sent to the enabling server via the enabling client, thereby simplifying the signaling interaction process and improving the efficiency of federated learning. Considering that the training capabilities of the training client may change over time (for example, the training samples increase or decrease), in another specific example, the enabling client may forward the evaluation request from the enabling server to the training client, so that the training client sends the training capabilities to the enabling server via the enabling client when receiving the evaluation request, so that the enabling server can obtain the current training capabilities of each participant.

[0054] like Figure 1 As shown, the federated learning method 100 according to an exemplary embodiment of the present disclosure may also include:

[0055] Step S130, determining whether a preset training condition is met according to the training ability of each of the multiple participants.

[0056] In some embodiments, the training capability of each participant may be summarized to determine whether the preset training condition is met based on whether the summarized training capability meets the model training requirement, wherein the model training requirement may be a summary requirement for the training capabilities of all participants. In a specific example, the training samples of each participant may be summarized to determine whether the total number of summarized training samples meets the model training requirement, so that if the total number of summarized training samples meets the model training requirement, it is determined that the preset training condition is met, and if the total number of summarized training samples does not meet the model training requirement, it is determined that the preset training condition is not met.

[0057] In other embodiments, it is possible to determine whether the preset training conditions are met by judging whether the training capability of each participant meets the model training requirements, wherein the model training requirements may be the requirements for the training capability of each participant. Figure 2 As shown, determining whether a preset training condition is met according to the training capability of each of the multiple participants may include:

[0058] Step S131, for each participant, determining whether the training capability meets the model training requirements;

[0059] Step S132, counting the number of participants that meet the model training requirements;

[0060] Step S133, in response to the number of participants that meet the model training requirements being less than or equal to a preset number, determining that the preset training condition is not met;

[0061] Step S134, in response to the number of participants that meet the model training requirements being greater than a preset number, determining that a preset training condition is met.

[0062] In this way, the enabling server can determine whether to start support for incomplete federated learning by judging the size between the number of participants who meet the model training requirements and the preset number. If the number of participants who meet the model training requirements is greater than the preset number, it can be considered that even if the training capacity of some participants is insufficient, the federated learning model obtained after aggregating the training results of each participant can still have a higher accuracy, that is, the insufficient training capacity of some participants has no effect or basically no effect on the accuracy of the federated learning model. In this way, each participant can be trained normally according to the original local training sample. The number of participants who meet the model training requirements is less than or equal to the preset number, and the number of participants who meet the model training requirements is small. If each participant is still trained according to the original local training sample, the accuracy of the federated learning model finally aggregated will be poor. In this way, the enabling server can start support for incomplete federated learning, so that the participants process the training samples accordingly to improve the accuracy of the federated learning model.

[0063] return Figure 1 According to an exemplary embodiment of the present disclosure, the federated learning method 100 may further include:

[0064] Step S140, in response to not meeting the preset training condition, generating corresponding weight parameters for each participant; and

[0065] Step S150, sending corresponding weight parameters to each participant, so that each participant performs weighted processing on local training samples based on the corresponding weight parameters and performs training based on the weighted training samples.

[0066] In this way, the server is enabled to start supporting incomplete federated learning when it is determined that the preset training conditions are not met based on the training capabilities of each participant, so as to generate and send corresponding weight parameters for each participant.

[0067] In addition, when preset training conditions are met, the training request can be forwarded to each of the multiple participants, so that each participant can perform training based on local training samples.

[0068] In some embodiments, the enabling server may determine whether to start support for incomplete federated learning based on whether the training request includes a target parameter. Specifically, in response to the training request including the target parameter, the enabling server may generate a corresponding weight parameter for each participant and send the corresponding weight parameter to each participant, so that each participant performs weighted processing on the local training sample based on the corresponding weight parameter and performs training based on the weighted training sample; in response to the training request not including the target parameter, the enabling server may forward the training request to each of the multiple participants, so that each participant performs training based on the local training sample.

[0069] In some embodiments, generating a corresponding weight parameter for each participant may include: for each participant, generating a weight parameter according to the degree of match between the training capability and the model training requirement. In a specific example, the higher the degree of match between the training capability and the model training requirement, the higher the weight parameter can be assigned, wherein the maximum weight parameter can be assigned when the training capability and the model training requirement are completely matched (the maximum weight parameter can be 1, for example). In a specific example, the weight parameter can be determined according to the ratio between the number of training samples of the participant and the number of required samples in the model training requirement. Wherein, the maximum weight parameter can be assigned when the number of training samples is greater than or equal to the required number of samples. In another specific example, the weight parameter can be determined according to the degree of match between the training sample quality of the participant and the required sample quality in the model training requirement. Alternatively, the weight parameter can also be generated by comprehensively combining the degree of match between the number of training samples and the required number of samples and the degree of match between the quality of the training samples and the required sample quality. In some embodiments, after generating the corresponding weight parameter for each participant, the weight parameter of each participant can also be normalized.

[0070] In some embodiments, the enabling server can generate a parameter correspondence table based on the weight parameters of each participant, so that the corresponding weight parameters can be sent to each participant in an orderly manner according to the parameter correspondence table. In addition, the enabling server can also send the generated parameter correspondence table to the training server so that the training server can know the corresponding weight parameters of each participant in a timely manner.

[0071] In some embodiments, the enabling server may also send the accuracy requirements for the training results and the corresponding model configuration to each participant so that each participant completes the corresponding training. The enabling server may receive the training results of each participant on the federated learning model, and send the training results of each participant to the training server that initiates the training request, so that the training server aggregates the training results of each participant to generate a federated learning model. Among them, federated learning may include horizontal federated learning, vertical federated learning, and migration federated learning. In vertical federated learning, each participant's data set may have the same sample space and different feature spaces, that is, each participant may share the same data object but have different data object features. In some embodiments, data alignment can be achieved between the participants through a preset encryption algorithm to find common data objects between the participants while protecting data privacy. In a specific example, for each participant, the model configuration sent by the enabling server may include the corresponding training parameters of the vertical federated learning model so that each participant completes the corresponding training, so that the training results of each participant are subsequently aggregated to generate a vertical federated learning model.

[0072] According to another aspect of the present disclosure, an enabling server 300 is further provided. The enabling server 300 may be configured to execute the federated learning method 100 described in any of the foregoing embodiments.

[0073] like Figure 3 As shown, the enabling server 300 according to an exemplary embodiment of the present disclosure may include a receiving module 310, a sending module 320, a determining module 330 and a generating module 340. The receiving module 310 may be configured to receive a training request for training a federated learning model. The sending module 320 may be configured to send an evaluation request for evaluating training capabilities to multiple participants participating in the training of the federated learning model to obtain the training capabilities of each of the multiple participants. The determining module 330 may be configured to determine whether a preset training condition is met based on the training capabilities of each of the multiple participants. The generating module 340 may be configured to generate a corresponding weight parameter for each participant in response to not meeting the preset training condition, and the sending module 320 may also be configured to send a corresponding weight parameter to each participant, so that each participant performs weighted processing on a local training sample based on the corresponding weight parameter and performs training based on the weighted training sample.

[0074] According to another aspect of the present disclosure, a federated learning method 400 that can be executed by an enabling client is provided, such as Figure 4 As shown, a federated learning method 400 according to an exemplary embodiment of the present disclosure may include:

[0075] Step S410, receiving a weight parameter sent by an enabling server;

[0076] Step S420, sending the weight parameter to the training client, so that the training client performs weighted processing on the local training samples based on the weight parameter and performs training based on the weighted training samples when receiving the weight parameter; and

[0077] Step S430: sending the training result of the training client to the enabling server.

[0078] In this way, when the training capacity of some participants is insufficient and the training conditions of the model cannot be met, the training client of each participant can complete the training of the model based on the corresponding weight parameters, thereby ensuring that the model training can continue and improving the accuracy of the federated learning model finally generated. In addition, the enabling client can act as a proxy for the training client to implement message interaction with other federated learning devices such as the enabling server, reducing the transmission capacity requirements for the training client and alleviating the work pressure of the training client, thereby better ensuring that the training client can complete the corresponding training normally.

[0079] In some embodiments, the enabling client sends a registration request to the enabling server to register the training client, wherein the registration request includes configuration information of the training client so that the enabling server registers the configuration information of the training client. The configuration information may include the available time of the training client so that the enabling server can subsequently select a training client in an available state from the registered training clients as a participant in this federated learning service. In a specific example, the configuration information may also include data labels, types, and number information of the training client.

[0080] In some scenarios, the enabling server needs to obtain the training capability of each participant to determine whether the preset training conditions of this federated learning model are met. In some embodiments, the enabling client can receive an evaluation request sent by the enabling server to evaluate the training capability of the training client, and forward the evaluation request to the training client, so that the training client sends the training capability to the enabling server via the enabling client when receiving the evaluation request.

[0081] According to another aspect of the present disclosure, an enabling client 500 is further provided. The enabling client 500 may be configured to execute the federated learning method 400 described in any of the foregoing embodiments.

[0082] like Figure 5 As shown, the enabling client 500 according to an exemplary embodiment of the present disclosure may include a receiving module 510 and a sending module 520. The receiving module 510 may be configured to receive the weight parameter sent by the enabling server. The sending module 520 may be configured to send the weight parameter to the training client, so that the training client performs weighted processing on the local training samples based on the weight parameter and performs training based on the weighted training samples when receiving the weight parameter. The sending module 520 may also be configured to send the training result of the training client to the enabling server.

[0083] According to another aspect of the present disclosure, a federated learning system is also provided. Figure 6 As shown, a federated learning system 600 according to an exemplary embodiment of the present disclosure may include an enabling server 300 and an enabling client 500. The enabling server 300 and the enabling client 500 may be located in a service enabler architecture layer (SEAL), and may provide support for the implementation of services through corresponding platform services, for example, providing support for incomplete vertical federated learning services.

[0084] In some embodiments, the federated learning system 600 may further include a training client 610 and a training server 620, wherein the training server 620 may be configured to send a training request for training a federated learning model to the enabling server 300, and the training client 610 may be configured to perform weighted processing on local training samples based on weight parameters and perform training based on the weighted training samples to generate training results. The training client 610 and the training server 620 may be located in a vertical application layer (Vertical Application Layer, VAL) for executing corresponding vertical application services, such as federated learning services.

[0085] The enabling client 500 and the training client 610 may be deployed in the same UE 630, so that the UE 630 may register on the enabling server 300 through the enabling client 500 and complete the corresponding federated learning training task through the training client 610. The training capability of the training client 610 deployed in the UE 630 may be used as the training capability of the UE 630. In the case where multiple training clients 610 are deployed in the UE 630, the training capabilities of all the training clients 610 deployed in the UE 630 may be aggregated as the training capability of the UE 630. In some deployment modes, one enabling client 500 may manage multiple training clients 610 and may provide services to multiple training clients 610. Similarly, in some deployment modes, one enabling server 300 may manage multiple training servers 620 and may provide services to multiple training servers 620.

[0086] like Figure 6 As shown, the training client 610 and the enabling client 500 can interact with the training server 620 and the enabling server 300 via the 3GPP network system 640. For example, the training client 610 and the enabling client 500 can be connected to the training server 620 and the enabling server 300 via the AMF and SMF network elements in the 3GPP network system 640, thereby realizing message interaction between the training client 610, the enabling client 500, the training server 620 and the enabling server 300.

[0087] The following combination Figure 7 Some embodiments are described in which the enabling client 500, the training client 610, the enabling server 300 and the training server 620 interact to implement incomplete longitudinal federated learning model training.

[0088] At step S710, the training client 610 registers in the enabling server 300 through the enabling client 500 to register its own configuration information. The configuration information may include its own expected available time, that is, the time period during which the training client 610 is in an available state.

[0089] At step S720, the training server 620 sends a training request to the enabling server 300 to request training of the vertical federated learning model. The training request may include a parameter for indicating whether to perform incomplete vertical federated learning.

[0090] At step S730, the enabling server 300 broadcasts an evaluation request for evaluating the training capability of the training client 610 to the enabling client 500 corresponding to each of the registered training clients 610. The registered training clients 610 may be stored in the enabling server 300 in the form of a list.

[0091] In step S740 , the training client 610 reports the training capability to the enabling server 300 through the corresponding enabling client 500 , and the enabling server 300 performs data integration and data analysis.

[0092] At step S750, the enabling server 300 compares the reported training capabilities and the requirements of the training server 620, and selects the training clients 610 that fully meet the requirements. When the number of training clients 610 that fully meet the requirements is insufficient (for example, less than a preset number), support for incomplete vertical federated learning is triggered, and corresponding weight parameters are generated for the training clients 610 of each participant. The maximum weight parameter (the maximum weight parameter is, for example, 1) can be generated for the training clients 610 that fully meet the requirements, and the weight parameters of other training clients 610 that do not fully meet the requirements can be assigned according to the degree of matching with the requirements. The assignment method can be provided by the training server 620, for example, the enabling server 300 can generate corresponding weight parameters for other training clients 610 that do not fully meet the requirements according to the parameters indicating the assignment method in the training request.

[0093] At step S760, the enabling server 300 sends a notification to the enabling client 500 corresponding to the selected training client 610, so as to inform the training client 610 of the service configuration of incomplete vertical federated learning through the enabling client 500. The service configuration may include corresponding weight parameters and expected training effects.

[0094] At step S770, in response to receiving a reply regarding the notification received by the training client 610, the enabling server 300 informs the training server 620 of the service configuration of the incomplete vertical federated learning and the selected training client 610. The selected training client 610 may be sent to the training server 620 in the form of a list.

[0095] At step S780 , the enabling server 300 sends training parameters for training the vertical federated learning model to each participant, so that the training client 610 of each participant completes the corresponding training, and sends the training results of the training client 610 of each participant to the training server 620 .

[0096] According to another aspect of the present disclosure, a device for federated learning is also provided. Figure 8 As shown, in an exemplary embodiment of the present disclosure, a device 800 for federated learning may include a processor 810 and a memory 820. The memory 820 stores instructions, which, when executed by the processor 810, implement the steps of the federated learning method described in any of the foregoing embodiments of the present disclosure.

[0097] The processor 810 may be an integrated circuit chip having the ability to process signals. The processor may be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), an off-the-shelf programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, to implement or execute the various methods, steps and logic block diagrams disclosed in the embodiments of the present disclosure. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc., and may be an X86 architecture or an ARM architecture, etc.

[0098] The memory 820 may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memory. Among them, the non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may be a random access memory (RAM), which is used as an external cache. By way of example but not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous connection dynamic random access memory (SLDRAM), and direct memory bus random access memory (DRRAM). It should be noted that the memory described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0099] According to another aspect of the present disclosure, a non-transitory computer-readable storage medium is provided, on which instructions are stored. When the instructions are executed, the steps of the federated learning method described in any of the foregoing embodiments of the present disclosure can be implemented.

[0100] Similarly, the non-transitory computer-readable storage medium in the embodiments of the present disclosure may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memory. It should be noted that the non-transitory computer-readable storage medium described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0101] According to another aspect of the present disclosure, a computer program product is further provided. The computer program product includes instructions. When the instructions are executed by a processor, the steps of the federated learning method described in any of the foregoing embodiments of the present disclosure can be implemented.

[0102] Instructions may be any set of instructions to be executed directly by one or more processors, such as machine code, or any set of instructions to be executed indirectly, such as a script. The terms "instructions", "applications", "processes", "steps" and "programs" (including "computer programs") herein may be used interchangeably herein. Instructions may be stored in an object code format for direct processing by one or more processors, or in any other computer language, including scripts or collections of independent source code modules that are interpreted on demand or compiled in advance. Other sections of this document explain the functions, methods and routines of instructions in more detail.

[0103] In the technical solution disclosed in the present invention, by enabling the server to provide support for incomplete federated learning, when the training capacity of some participants is insufficient and the training conditions of the model cannot be met, corresponding weight parameters are generated for each participant so that the participants can perform training based on the corresponding weight parameters, ensuring that the model training can continue, thereby improving the network's support capacity for federated learning services, and being able to provide more flexible federated learning services and expand the application scenarios of federated learning services. In addition, according to the solution disclosed in the present invention, the utilization of sample data is improved, and the influence of insufficient sample quantity on the poor effect of vertical federated learning can be reduced, so as to improve the accuracy of the federated learning model.

[0104] As used herein, the word "exemplary" means "serving as an example, instance, or illustration" rather than as a "model" to be exactly copied. Any implementation described as an example herein is not necessarily to be construed as preferred or advantageous over other implementations. Furthermore, the present disclosure is not to be limited by any expressed or implied theory given in the technical field, background, summary, or detailed description.

[0105] As used herein, the term "substantially" is intended to include any minor variations due to design or manufacturing imperfections, device or component tolerances, environmental influences, and / or other factors. The term "substantially" also allows for deviations from a perfect or ideal situation due to parasitic effects, noise, and other practical considerations that may exist in actual implementations.

[0106] In addition, the terms "first", "second" and the like may also be used herein for reference purposes only and are not intended to be limiting. For example, the terms "first", "second" and other such numerical terms referring to structures or elements do not imply a sequence or order unless the context clearly indicates otherwise.

[0107] It should also be understood that when the term “include / comprises” is used in this document, it indicates the presence of the specified features, integers, steps, operations, units and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, units and / or components and / or their combinations.

[0108] In this disclosure, the term "provide" is used in a broad sense to cover all ways of obtaining an object, and thus "providing an object" includes but is not limited to "purchasing", "preparing / manufacturing", "arranging / setting up", "installing / assembling", and / or "ordering" an object, etc.

[0109] As used herein, the term "and / or" includes any and all combinations of one or more of the associated listed items. The terms used herein are for the purpose of describing specific embodiments only and are not intended to limit the present disclosure. As used herein, the singular forms "a", "an", and "the" are also intended to include the plural forms, unless the context clearly indicates otherwise.

[0110] Those skilled in the art will appreciate that the boundaries between the above operations are merely illustrative. Multiple operations can be combined into a single operation, a single operation can be distributed in additional operations, and operations can be performed at least partially overlapping in time. Moreover, alternative embodiments may include multiple instances of a specific operation, and the order of operations may be changed in various other embodiments. However, other modifications, variations, and replacements are equally possible. Aspects and elements of all embodiments disclosed above may be combined in any manner and / or in combination with aspects or elements of other embodiments to provide multiple additional embodiments. Therefore, this specification and accompanying drawings should be considered illustrative, not restrictive.

[0111] Although some specific embodiments of the present disclosure have been described in detail by way of example, it should be understood by those skilled in the art that the above examples are for illustration only and are not intended to limit the scope of the present disclosure. The various embodiments disclosed herein may be combined in any manner without departing from the spirit and scope of the present disclosure. It should also be understood by those skilled in the art that various modifications may be made to the embodiments without departing from the scope and spirit of the present disclosure. The scope of the present disclosure is defined by the appended claims.

Claims

1. A federated learning method, characterized in that: The federated learning method is executed by an enabling server, and the federated learning method includes: Receive training requests for training federated learning models; Sending an evaluation request for evaluating training capabilities to multiple participants participating in training the federated learning model to obtain the training capability of each of the multiple participants; Determining whether a preset training condition is met according to the training ability of each of the plurality of participants; In response to not meeting the preset training condition, generating corresponding weight parameters for each participant; and The corresponding weight parameters are sent to each participant, so that each participant performs weighted processing on the local training samples based on the corresponding weight parameters and performs training based on the weighted training samples.

2. The federated learning method according to claim 1, characterized in that: Determining whether a preset training condition is met according to the training capability of each of the multiple participants includes: For each participant, determine whether the training capacity meets the model training requirements; Count the number of participants that meet the model training requirements; In response to the number being less than or equal to a preset number, it is determined that the preset training condition is not satisfied.

3. The federated learning method according to claim 1, characterized in that: The corresponding weight parameters generated for each participant include: For each participant, weight parameters are generated based on the degree of match between training capabilities and model training requirements.

4. The federated learning method according to claim 1, characterized in that: The federated learning method further includes: receiving a training result of the federated learning model from each of the plurality of participants; and The training result of each participant is sent to the training server that initiated the training request, so that the training server summarizes the training result of each participant.

5. The federated learning method according to claim 1, characterized in that: The federated learning method further includes: A user terminal in an available state is selected from user terminals registered with the enabling server as a participant participating in training the federated learning model.

6. The federated learning method according to claim 1, characterized in that: Each participant includes a training client and an enabling client, wherein sending an evaluation request for evaluating training capabilities to multiple participants participating in training the federated learning model includes: The evaluation request is sent to the enabling client of each participant, so that the enabling client forwards the evaluation request to the training client, so that the training client sends the training capability to the enabling server via the enabling client when receiving the evaluation request.

7. The federated learning method according to claim 1, characterized in that: The federated learning method further includes: determining whether the training request includes a target parameter, the target parameter indicating that the request supports incomplete federated learning; In response to the training request including the target parameter, performing the operations of generating a corresponding weight parameter for each participant and sending the corresponding weight parameter to each participant, so that each participant performs weighted processing on a local training sample based on the corresponding weight parameter and performs training based on the weighted training sample; and In response to the training request not including the target parameter, forwarding the training request to each of the multiple participants, so that each participant performs training based on a local training sample.

8. A federated learning method, characterized in that: The federated learning method is executed by an enabling client, and the federated learning method includes: Receive weight parameters sent by the enabling server; Sending the weight parameter to the training client, so that the training client, upon receiving the weight parameter, performs weighted processing on the local training samples based on the weight parameter and performs training based on the weighted training samples; and The training result of the training client is sent to the enabling server.

9. The federated learning method according to claim 8, characterized in that: The federated learning method further includes: A registration request for registering the training client is sent to the enabling server, wherein the registration request includes configuration information of the training client.

10. The federated learning method according to claim 9, characterized in that: The configuration information includes the available time of the training client.

11. The federated learning method according to claim 8, characterized in that: The federated learning method further includes: receiving an evaluation request sent by the enabling server to evaluate the training capability of the training client; and The training capability of the training client is sent to the enabling server.

12. An enabling server, characterized in that: The enabling server comprises: A receiving module, wherein the receiving module is configured to receive a training request for training a federated learning model; A sending module, wherein the sending module is configured to send an evaluation request for evaluating training capabilities to multiple participants participating in training the federated learning model, so as to obtain the training capability of each of the multiple participants; a determination module, the determination module being configured to determine whether a preset training condition is met according to the training ability of each of the plurality of participants; A generation module, wherein the generation module is configured to generate corresponding weight parameters for each participant in response to not meeting a preset training condition, The receiving module is further configured to send corresponding weight parameters to each participant, so that each participant performs weighted processing on local training samples based on the corresponding weight parameters and performs training based on the weighted training samples.

13. An enabling client, characterized in that: The enabling server comprises: a receiving module, the receiving module being configured to receive a weight parameter sent by an enabling server; and A sending module, wherein the sending module is configured to: sending the weight parameter to the training client, so that the training client, upon receiving the weight parameter, performs weighted processing on the local training samples based on the weight parameter and performs training based on the weighted training samples, and The training result of the training client is sent to the enabling server.

14. A federated learning system, characterized in that: The federated learning system includes: The enabling server according to claim 12; and The enabling client according to claim 13.

15. The federated learning system according to claim 14, characterized in that: The federated learning system further includes: a training server, the training server being configured to send a training request for training a federated learning model to the enabling server; and The training client is configured to perform weighted processing on local training samples based on weight parameters and perform training based on the weighted training samples to generate training results.

16. A device for federated learning, characterized in that: The device includes a processor and a memory, wherein instructions are stored in the memory, and when the instructions are executed by the processor, the steps of the federated learning method according to any one of claims 1 to 11 are implemented.

17. A non-transitory computer-readable storage medium, characterized in that: The non-transitory computer-readable storage medium stores instructions, and when the instructions are executed, the steps of the federated learning method according to any one of claims 1 to 11 are implemented.

18. A computer program product, characterized in that The computer program product comprises instructions, and when the instructions are executed by a processor, the steps of the federated learning method according to any one of claims 1 to 11 are implemented.

Images