Communication method, device and system

Abstract

A communication method, device and system are applied to the field of federal learning. The method comprises the steps that a federated learning server receives a first request message from a first federated learning member, the first request message carries a first training requirement or an intermediate training result, and the intermediate training result is used for determining the first training requirement; then, the federated learning server determines a second federated learning member from the candidate federated learning members according to the ability information of the candidate federated learning members and the first training requirement, and the second federated learning member supports continuous execution of the training task of the first model; according to the technical scheme, when the appropriate second federal learning member is selected to continue to execute the training task, the current training condition or training requirement is considered, the model convergence time can be prevented from being increased, and the model performance is ensured.

Description

Technical Field

[0001] The present application relates to the field of communication technology, and more specifically, to a communication method, device and system. Background Art

[0002] Federated learning (FL) is a machine learning technology in the core network. It supports multiple participants (for example, the server network data analytics function (NWDAF) and the client NWDAF) to train locally without exchanging training data and exchange model parameters through a security mechanism, thereby achieving the effect of collaborative training. Federated learning can effectively help multiple institutions use data and conduct learning modeling while meeting the requirements of user privacy protection, data security and government regulations.

[0003] However, in the process of executing federated learning tasks, if the client NWDAF cannot complete the model training task, it may increase the time for model convergence and cause the model performance to deteriorate. Summary of the invention

[0004] The present application provides a communication method, device and system, which can avoid the increase of model convergence time and ensure model performance.

[0005] In the first aspect, a communication method is provided. The method can be executed by a federated learning server. Unless otherwise specified, the "federated learning server" in this application can refer to the communication device itself, or a component in the communication device (for example, a communication module, a processor, a circuit, a chip, or a chip system, etc.), or a logic module or software that can implement all or part of the functions of the communication device.

[0006] The method includes: receiving a first request message from a first federated learning member, the first request message including a first training requirement corresponding to a first model; determining a second federated learning member from candidate federated learning members according to capability information of candidate federated learning members and the first training requirement, the capability information indicating whether the candidate federated learning member supports continuing to execute the training task of the first model, at least one of the candidate federated learning members supports continuing to execute the training task of the first model, and at least one federated learning member includes a second federated learning member.

[0007] In the second aspect, a communication method is provided. The method can be executed by a federated learning server. Unless otherwise specified, the "federated learning server" in this application can refer to the communication device itself, or a component in the communication device (for example, a communication module, a processor, a circuit, a chip, or a chip system, etc.), or a logic module or software that can implement all or part of the functions of the communication device.

[0008] The method includes: receiving a first request message from a first federated learning member, the first request message including an intermediate training result of a first model, the intermediate training result being obtained by the first federated learning member executing a training task of the first model; determining a first training requirement corresponding to the first model according to the intermediate training result; determining a second federated learning member from candidate federated learning members according to capability information of the candidate federated learning members and the first training requirement, the capability information indicating whether the candidate federated learning member supports continuing to execute the training task of the first model, at least one of the candidate federated learning members supports continuing to execute the training task of the first model, and at least one federated learning member includes the second federated learning member.

[0009] Optionally, the number of the first federated learning member or the second federated learning member may be one or more, which is not limited in this application.

[0010] It can be understood that the first training requirement in this application refers to the current training requirement for the training task of the first model, and the second training requirement can be understood as the overall training requirement for the training task of the first model. The first training requirement is determined based on the second training requirement and the current training situation (for example, the intermediate training result of the first model).

[0011] Based on the above scheme, the federated learning server can select at least one second federated learning member from the candidate federated learning members based on the first training requirement or the intermediate training result of the first model carried in the received first request message to continue the training task of the first model and ensure the completion of the training of the first model. That is, when the federated learning server selects a suitable second federated learning member to continue the training task, it can take into account the current training situation or training requirements, thereby avoiding an increase in the model convergence time and a decrease in the model performance.

[0012] In one possible design, the first request message also includes indication information, where the indication information is used to indicate that the first federated learning member cannot continue to perform the training task of the first model, and / or the indication information is used to indicate that at least one federated learning member is found.

[0013] Based on the above scheme, by carrying indication information in the first request message, it is possible to explicitly or implicitly indicate that the first federated learning member cannot continue to perform the training task of the first model, and / or to discover at least one federated learning member so that the federated learning server can perform the process of discovering the second federated learning member, that is, select a suitable second federated learning member according to the first training requirement carried in the first request message to continue the training task of the first model.

[0014] In one possible design, the method also includes: sending a first response message to the first federated learning member, the first response message including information of the second federated learning member.

[0015] Exemplarily, the information of the second federated learning member includes: an identification (identity, ID) and / or address information of the second federated learning member.

[0016] In one possible design, the method also includes: sending a second request message to the second federated learning member, the second request message is used to request to continue the training task of the first model, and the second request message includes the first training requirement and the intermediate training result of the first model.

[0017] Based on the above scheme, when the federated learning server finds a second federated learning member for continuing to execute the training task of the first model, it can feed back the information of the second federated learning member to the first federated learning member, so that the first federated learning member can subsequently request the second federated learning member to execute the training task of the first model; alternatively, the federated learning server may not feed back the information of the second federated learning member to the first federated learning member, and then the federated learning server can request the second federated learning member to execute the training task of the first model, thereby providing multiple implementation methods and being flexible.

[0018] In one possible design, the intermediate training result of the first model includes at least one of the following: a temporary model obtained by executing the training task of the first model; accuracy information of the temporary model; accuracy information required by the training task of the first model; training time of the temporary model; training time required to continue to execute the training task of the first model; resource usage of the temporary model; training resources required to continue to execute the training task of the first model; or, the number of training rounds of the temporary model; the number of training rounds required to continue to execute the training task of the first model.

[0019] In one possible design, the method also includes: receiving a first model obtained through training from a second federated learning member and training information of the first model; wherein the training information of the first model includes at least one of the following: an identifier of the first model; address information of the first model; an analysis identifier corresponding to the first model; a service identifier corresponding to the first model; an application identifier corresponding to the first model; a time of acquiring the first model; a training duration of the first model; or accuracy information of the first model.

[0020] Based on the above scheme, the federated learning server can obtain the trained first model and the training information of the first model, that is, continue to execute the training task of the first model through the second federated learning member, thereby avoiding the increase in model convergence time caused by the inability of the first federated learning member to continue to execute the training task of the first model, and ensuring the model performance.

[0021] In one possible design, before determining the second federated learning member from the candidate federated learning members based on the capability information of the candidate federated learning members and the first training requirement, the method further includes: acquiring the capability information of the candidate federated learning members.

[0022] Exemplarily, the capability information is also used to indicate at least one of the following: the training time supported by the candidate federated learning member; the available training time corresponding to the candidate federated learning member; the training resources supported by the candidate federated learning member; the available training resources corresponding to the candidate federated learning member; or the number of model training rounds supported by the candidate federated learning member.

[0023] Based on the above scheme, by obtaining the capability information of candidate federated learning members, when a first request message from a first federated learning member is received, a suitable second federated learning member can be selected according to the capability information of the candidate federated learning members and the current training situation to continue the training task of the first model, thereby avoiding an increase in the model convergence time and a decrease in model performance.

[0024] In one possible design, the first training requirement includes at least one of the following: accuracy information required to continue the training task of the first model; training time required to continue the training task of the first model; training resources required to continue the training task of the first model; or the number of training rounds required to continue the training task of the first model.

[0025] Exemplarily, the training time required to continue executing the training task of the first model is determined based on the training time required by the training task of the first model and the training time of the temporary model obtained by executing the training task of the first model.

[0026] Exemplarily, the number of training rounds required to continue executing the training task of the first model is determined according to the number of training rounds required by the training task of the first model and the number of training rounds of the temporary model obtained by executing the training task of the first model.

[0027] In one possible design, before receiving a first request message from a first federated learning member, the method also includes: receiving a third request message from a model requester, the third request message being used to request execution of a training task for the first model, the third request message including a second training requirement corresponding to the first model; determining a first federated learning member from candidate federated learning members based on capability information and the second training requirements of the candidate federated learning members; sending a fourth request message to the first federated learning member, the fourth request message being used to request execution of a training task for the first model, the fourth request message including the second training requirement; wherein the first training requirement is determined based on the second training requirement and an intermediate training result of the first model.

[0028] In one possible design, the second training requirement includes at least one item: an analysis identifier corresponding to the first model; a service identifier corresponding to the first model; an application identifier corresponding to the first model; the time for acquiring the first model and the training duration of the first model; the accuracy information required by the training task of the first model; the training time required by the training task of the first model; the training resources required by the training task of the first model; or the number of training rounds required by the training task of the first model.

[0029] In one possible design, the first federated learning member or the second federated learning member is an enabled client or an enabled server.

[0030] In a third aspect, a communication method is provided. The method can be performed by a first federated learning member. Unless otherwise specified, the "first federated learning member" in this application can refer to the communication device itself (for example, a terminal or access network), or a component in the communication device (for example, a communication module, a processor, a circuit, a chip, or a chip system, etc.), or a logic module or software that can implement all or part of the functions of the communication device.

[0031] The method includes: when it is determined that the first federated learning member cannot continue to perform the training task of the first model, and / or it is determined that the intermediate training result of the first model does not meet the second training requirement corresponding to the first model, determine the first training requirement corresponding to the first model according to the intermediate training result and the second training requirement, wherein the intermediate training result is obtained by the first federated learning member performing the training task of the first model; send a first request message to a federated learning server, the first request message includes the first training requirement; receive a first response message from the federated learning server, the first response message includes information of the second federated learning member, and the second federated learning member supports continuing to perform the training task of the first model.

[0032] In a fourth aspect, a communication method is provided. The method can be performed by a first federated learning member. Unless otherwise specified, the "first federated learning member" in this application can refer to the communication device itself (for example, a terminal or access network), or a component in the communication device (for example, a communication module, a processor, a circuit, a chip, or a chip system, etc.), or a logic module or software that can implement all or part of the functions of the communication device.

[0033] The method includes: when it is determined that the first federated learning member cannot continue to perform the training task of the first model, and / or when it is determined that the intermediate training result of the first model does not meet the second training requirement corresponding to the first model, sending a first request message to a federated learning server, wherein the first request message includes the intermediate training result of the first model, the intermediate training result is used to determine the first training requirement corresponding to the first model, and the intermediate training result is obtained by the first federated learning member performing the training task of the first model; receiving a first response message from the federated learning server, the first response message includes information of the second federated learning member, and the second federated learning member supports continuing to perform the training task of the first model.

[0034] Based on the above scheme, when the first federated learning member determines that it cannot continue to perform the training task of the first model, and / or the intermediate training result of the first model does not meet the second training requirement corresponding to the first model, it can request the federated learning server to find at least one federated learning member, such as a second federated learning member, for continuing to perform the training task of the first model. Based on the first training requirement or the intermediate training result of the first model carried in the received first request message, the federated learning server can determine the second federated learning member from the candidate federated learning members to continue to perform the training task of the first model and ensure that the training of the first model is completed. That is, when the federated learning server selects a suitable second federated learning member to continue to perform the training task, it can take into account the current training situation or training requirements, thereby avoiding an increase in model convergence time and a decrease in model performance.

[0035] In one possible design, the method also includes: sending a second request message to the second federated learning member, the second request message is used to request to continue the training task of the first model, and the second request message includes the first training requirement and the intermediate training result.

[0036] In one possible design, the first request message also includes indication information, where the indication information is used to indicate that the first federated learning member cannot continue to perform the training task of the first model, and / or the indication information is used to indicate that at least one federated learning member is found.

[0037] In one possible design, before sending the first request message to the federated learning server, the method also includes: sending capability information of the first federated learning member to the federated learning server, the capability information indicating that the first federated learning member supports executing the training task of the first model.

[0038] In one possible design, the capability information is also used to indicate at least one of the following: the training time supported by the first federated learning member; the available training time corresponding to the first federated learning member; the training resources supported by the first federated learning member; the available training resources corresponding to the first federated learning member; or the number of model training rounds supported by the first federated learning member.

[0039] In one possible design, before determining the first training requirement corresponding to the first model based on the intermediate training result and the second training requirement, or before sending a first request message to the federated learning server, the method also includes: receiving a fourth request message from the federated learning server, the fourth request message being used to request execution of the training task of the first model, the fourth request message including the second training requirement; executing the training task of the first model according to the second training requirement to obtain the intermediate training result.

[0040] In one possible design, the first training requirement includes at least one of the following: accuracy information required to continue the training task of the first model; training time required to continue the training task of the first model; training resources required to continue the training task of the first model; or the number of training rounds required to continue the training task of the first model.

[0041] In one possible design, the second training requirement includes at least one item: an analysis identifier corresponding to the first model; a service identifier corresponding to the first model; an application identifier corresponding to the first model; the time for acquiring the first model; the training duration of the first model; or, the accuracy information required for the training task of the first model; the training time required for the training task of the first model; the training resources required for the training task of the first model; or, the number of training rounds required for the training task of the first model.

[0042] In one possible design, the intermediate training result includes at least one of the following: a temporary model obtained by executing the training task of the first model; accuracy information of the temporary model; accuracy information required by the training task of the first model; training time of the temporary model; training time required to continue to execute the training task of the first model; resource usage of the temporary model; training resources required to continue to execute the training task of the first model; the number of training rounds of the temporary model; or, the number of training rounds required to continue to execute the training task of the first model.

[0043] The beneficial effects corresponding to the third or fourth aspect mentioned above can refer to the relevant description of the beneficial effects corresponding to the first or second aspect, which will not be repeated here.

[0044] In a fifth aspect, a communication method is provided. The method can be performed by a second federated learning member. Unless otherwise specified, the "second federated learning member" in this application can refer to the communication device itself (for example, a terminal or access network), or a component in the communication device (for example, a communication module, a processor, a circuit, a chip, or a chip system, etc.), or a logic module or software that can implement all or part of the functions of the communication device.

[0045] The method includes: receiving a second request message from a first federated learning member or a federated learning server, the second request message is used to request to continue to execute the training task of the first model, the second request message includes an intermediate training result of the first model and a first training requirement corresponding to the first model; continuing to execute the training task according to the intermediate training result and the first training requirement, and obtaining the first model and the training information of the first model.

[0046] Furthermore, the first model and training information of the first model are sent to the federated learning server.

[0047] Based on the above scheme, the second federated learning member can continue to execute the training task in a targeted manner based on the intermediate training results and the first training requirements carried in the received second request message, obtain the first model and the training information of the first model, ensure the completion of the training of the first model, avoid the increase of the model convergence time, and avoid the degradation of model performance.

[0048] In one possible design, before continuing to perform the training task according to the intermediate training result and the first training requirement, the method also includes: determining to continue to perform the training task according to the capability information of the second federated learning member and the first training requirement.

[0049] In one possible design, before receiving a second request message from the first federated learning member or the federated learning server, the method also includes: sending capability information of the second federated learning member to the federated learning server, the capability information indicating that the second federated learning member supports continuing to execute the training task of the first model.

[0050] In one possible design, the capability information is also used to indicate at least one of the following: training time supported by the second federated learning member; available training time corresponding to the second federated learning member; training resources supported by the second federated learning member; available training resources corresponding to the second federated learning member; or the number of model training rounds supported by the second federated learning member.

[0051] In one possible design, the intermediate training results include at least one of the following: a temporary model obtained by executing the training task of the first model; accuracy information of the temporary model; accuracy information required for the training task of the first model; training time of the temporary model; training time required to continue to execute the training task of the first model; resource usage of the temporary model; training resources required to continue to execute the training task of the first model; the number of training rounds of the temporary model; or, the number of training rounds required to continue to execute the training task of the first model.

[0052] In a possible design, the training information of the first model includes at least one of the following: an identifier of the first model; address information of the first model; an analysis identifier corresponding to the first model; a service identifier corresponding to the first model; an application identifier corresponding to the first model; a time of acquiring the first model; a training duration of the first model; or accuracy information of the first model;

[0053] In one possible design, the first training requirement includes at least one of the following: accuracy information required to continue the training task of the first model; training time required to continue the training task of the first model; training resources required to continue the training task of the first model; or the number of training rounds required to continue the training task of the first model.

[0054] In one possible design, the first federated learning member or the second federated learning member is an enabled client or an enabled server.

[0055] The corresponding beneficial effects of the fifth aspect can refer to the relevant description of the first aspect or the second aspect, which will not be repeated here.

[0056] In the sixth aspect, a communication method is provided. The method can be executed by a federated learning server. Unless otherwise specified, the "federated learning server" in this application can refer to the communication device itself, or a component in the communication device (for example, a communication module, a processor, a circuit, a chip, or a chip system, etc.), or a logic module or software that can implement all or part of the functions of the communication device.

[0057] The method includes: receiving a third request message from a model requester, the third request message is used to request execution of a training task for a first model, and the third request message includes a second training requirement corresponding to the first model; determining a first federated learning member and a second federated learning member from candidate federated learning members according to capability information and the second training requirement of the candidate federated learning members, the capability information indicating whether the candidate federated learning members support continuing to execute the training task for the first model, at least one of the candidate federated learning members supports continuing to execute the training task for the first model, at least one federated learning member includes a second federated learning member, and the second federated learning member is an alternative federated learning member for executing the training task for the first model; sending a fourth request message to the first federated learning member, the fourth request message is used to request execution of the training task for the first model, and the fourth request message includes the second training requirement corresponding to the first model and information of the second federated learning member.

[0058] It can be understood that the second federated learning member is an alternative (or standby) federated learning member for executing the training task of the first model. Specifically, the second federated learning member can be regarded as an alternative federated learning member of the first federated learning member. Among them, the alternative (or standby) federated learning member can be understood as: the second federated learning member can continue to execute the training task of the first model when the first federated learning member cannot continue to execute the training task of the first model to ensure that the first model can be trained to completion, that is, the second federated learning member supports the continued execution of the training task of the first model, or the second federated learning member has the ability to continue to execute the training task of the first model.

[0059] Based on the above scheme, the federated learning server can select the first federated learning member and the second federated learning member from the candidate federated learning members according to the capability information of the candidate federated learning members and the second training requirement, wherein the second federated learning member supports the continuation of the training task of the first model, especially when the first federated learning member cannot continue to perform the training task of the first model, and / or the intermediate training result of the first model does not meet the second training requirement, it can ensure the completion of the training of the first model. That is, when the federated learning server selects the federated learning member to perform the training task, it can simultaneously determine the first federated learning member and the second federated learning member, so that when the first federated learning member cannot continue to perform the training task of the first model later, it can request the second federated learning member to continue to perform the training task of the first model according to the information of the second federated learning member, that is, the first federated learning member and the federated learning server do not need to perform signaling interaction later (for example, the first federated learning member requests the federated learning server to discover at least one federated learning member), which can avoid the increase of model convergence time, avoid the degradation of model performance, and improve the efficiency of model training.

[0060] In one possible design, before determining the first federated learning member and the second federated learning member from the candidate federated learning members based on the capability information and the second training requirement of the candidate federated learning members, the method further includes: acquiring the capability information of the candidate federated learning members.

[0061] In one possible design, the capability information is also used to indicate at least one of the following: the training time supported by the candidate federated learning member; the available training time corresponding to the candidate federated learning member; the training resources supported by the candidate federated learning member; the available training resources corresponding to the candidate federated learning member; or the number of model training rounds supported by the candidate federated learning member.

[0062] In one possible design, the second training requirement includes at least one item: an analysis identifier corresponding to the first model; a service identifier corresponding to the first model; an application identifier corresponding to the first model; the time for acquiring the first model; the training duration of the first model; the accuracy information required by the training task of the first model; the training time required by the training task of the first model; the training resources required by the training task of the first model; or the number of training rounds required by the training task of the first model.

[0063] In one possible design, the first federated learning member or the second federated learning member is an enabled client or an enabled server.

[0064] The corresponding beneficial effects of the sixth aspect mentioned above can refer to the relevant description of the first aspect or the second aspect, which will not be repeated here.

[0065] In the seventh aspect, a communication method is provided. The method can be performed by a first federated learning member. Unless otherwise specified, the "first federated learning member" in this application can refer to the communication device itself (for example, a terminal or access network), or a component in the communication device (for example, a communication module, a processor, a circuit, a chip, or a chip system, etc.), or a logic module or software that can implement all or part of the functions of the communication device.

[0066] The method includes: obtaining information of a second federated learning member, the second federated learning member supports continuing to execute the training task of the first model; when it is determined that the first federated learning member cannot continue to execute the training task of the first model, and / or it is determined that the intermediate training result of the first model does not meet the second training requirement corresponding to the first model, determining the first training requirement corresponding to the first model according to the intermediate training result and the second training requirement; sending a second request message to the second federated learning member, the second request message is used to request to continue to execute the training task, and the second request message includes the first training requirement and the intermediate training result.

[0067] Based on the above scheme, when the first federated learning member cannot continue to perform the training task of the first model, and / or the intermediate training result of the first model does not meet the second training requirement, the first federated learning member can request the second federated learning member to continue to perform the training task of the first model based on the information obtained from the second federated learning member to ensure that the training of the first model is completed. That is to say, at this time, the first federated learning member and the federated learning server do not need to perform subsequent signaling interaction (for example, the first federated learning member requests the federated learning server to discover at least one federated learning member), which can improve the efficiency of model training and ensure model performance.

[0068] In one possible design, obtaining information about a second federated learning member includes: receiving a fourth request message from a federated learning server, the fourth request message being used to request execution of a training task for the first model, the fourth request message including a second training requirement and information about the second federated learning member; the method also includes: executing the training task according to the second training requirement to obtain an intermediate training result.

[0069] In one possible design, before obtaining information of the second federated learning member, the method also includes: sending capability information of the first federated learning member to the federated learning server, where the capability information indicates that the first federated learning member supports executing the training task of the first model.

[0070] In one possible design, the capability information is also used to indicate at least one of the following: the training time supported by the first federated learning member; the available training time corresponding to the first federated learning member; the training resources supported by the first federated learning member; the available training resources corresponding to the first federated learning member; or the number of model training rounds supported by the first federated learning member.

[0071] In one possible design, the first training requirement includes at least one of the following: accuracy information required to continue the training task of the first model; training time required to continue the training task of the first model; training resources required to continue the training task of the first model; or the number of training rounds required to continue the training task of the first model.

[0072] In one possible design, the second training requirement includes at least one item: an analysis identifier corresponding to the first model; a service identifier corresponding to the first model; an application identifier corresponding to the first model; the time for acquiring the first model; the training duration of the first model; the accuracy information required by the training task of the first model; the training time required by the training task of the first model; the training resources required by the training task of the first model; or the number of training rounds required by the training task of the first model.

[0073] In one possible design, the intermediate training result includes at least one of the following: a temporary model obtained by executing the training task of the first model; accuracy information of the temporary model; accuracy information required by the training task of the first model; training time of the temporary model; training time required to continue to execute the training task of the first model; resource usage of the temporary model; training resources required to continue to execute the training task of the first model; the number of training rounds of the temporary model; or, the number of training rounds required to continue to execute the training task of the first model.

[0074] In one possible design, the first training requirement corresponding to the first model is determined based on the intermediate training results and the second training requirements, including: determining the training time required to continue executing the training task of the first model based on the training time required for the training task of the first model and the training time of the temporary model obtained by executing the training task of the first model; and / or determining the number of training rounds required to continue executing the training task of the first model based on the number of training rounds required for the training task of the first model and the number of training rounds of the temporary model obtained by executing the training task of the first model.

[0075] In one possible design, the first federated learning member or the second federated learning member is an enabled client or an enabled server.

[0076] The corresponding beneficial effects of the seventh aspect mentioned above can be referred to the relevant description of the sixth aspect and will not be elaborated here.

[0077] In an eighth aspect, a communication device is provided, which has the function of implementing the first aspect mentioned above. For example, the communication device includes a module or unit or means corresponding to executing the operations involved in the first aspect mentioned above. The module or unit or means can be implemented by software, or by hardware, or by a combination of software and hardware.

[0078] Exemplarily, the communication device may be the above-mentioned federated learning server, for example, it may be a module or unit (such as a chip, or a chip system, or a circuit) corresponding one to one to executing the method, operation, step or action described in the first aspect above.

[0079] In a possible implementation, the communication device includes: a transceiver unit (or a communication module), and a processing unit (or a processing module) connected to the transceiver unit.

[0080] Exemplarily, a transceiver unit is used to receive a first request message from a first federated learning member, the first request message including a first training requirement corresponding to a first model; a processing unit is used to determine a second federated learning member from candidate federated learning members based on capability information of the candidate federated learning members and the first training requirement, the capability information indicating whether the candidate federated learning member supports continuing to execute the training task of the first model, at least one of the candidate federated learning members supports continuing to execute the training task of the first model, and at least one federated learning member includes a second federated learning member.

[0081] In the ninth aspect, a communication device is provided, which has the function of implementing the second aspect. For example, the communication device includes a module or unit or means corresponding to the operation involved in the second aspect. The module or unit or means can be implemented by software, or by hardware, or by a combination of software and hardware.

[0082] Exemplarily, the communication device may be the above-mentioned federated learning server, for example, it may be a module or unit (such as a chip, or a chip system, or a circuit) corresponding one to one to executing the method, operation, step or action described in the above-mentioned second aspect.

[0083] In a possible implementation, the communication device includes: a transceiver unit (or a communication module), and a processing unit (or a processing module) connected to the transceiver unit.

[0084] Exemplarily, a transceiver unit is used to receive a first request message from a first federated learning member, the first request message includes an intermediate training result of a first model, and the intermediate training result is obtained by the first federated learning member performing a training task of the first model; a processing unit is used to determine a first training requirement corresponding to the first model according to the intermediate training result; the processing unit is also used to determine a second federated learning member from candidate federated learning members according to capability information of the candidate federated learning members and the first training requirement, the capability information indicates whether the candidate federated learning member supports continuing to perform the training task of the first model, at least one of the candidate federated learning members supports continuing to perform the training task of the first model, and at least one federated learning member includes a second federated learning member

[0085] In the tenth aspect, a communication device is provided, which has the function of implementing the third aspect mentioned above. For example, the communication device includes a module or unit or means corresponding to executing the operations involved in the third aspect mentioned above. The module or unit or means can be implemented by software, or by hardware, or by a combination of software and hardware.

[0086] Exemplarily, the communication device may be the first federated learning member mentioned above, for example, it may be a module or unit (such as a chip, or a chip system, or a circuit) corresponding one to one to executing the method, operation, step or action described in the third aspect mentioned above.

[0087] In a possible implementation, the communication device includes: a transceiver unit (or a communication module), and a processing unit (or a processing module) connected to the transceiver unit.

[0088] Exemplarily, when it is determined that the first federated learning member cannot continue to perform the training task of the first model, and / or it is determined that the intermediate training result of the first model does not meet the second training requirement corresponding to the first model, the processing unit is used to determine the first training requirement corresponding to the first model based on the intermediate training result and the second training requirement, wherein the intermediate training result is obtained by the first federated learning member performing the training task of the first model; the transceiver unit is used to send a first request message to the federated learning server, the first request message includes the first training requirement; the transceiver unit is also used to receive a first response message from the federated learning server, the first response message includes information of the second federated learning member, and the second federated learning member supports continuing to perform the training task of the first model.

[0089] In the eleventh aspect, a communication device is provided, which has the function of implementing the fourth aspect. For example, the communication device includes a module or unit or means corresponding to the operation involved in the fourth aspect. The module or unit or means can be implemented by software, or by hardware, or by a combination of software and hardware.

[0090] Exemplarily, the communication device may be the first federated learning member mentioned above, for example, it may be a module or unit (such as a chip, or a chip system, or a circuit) corresponding one to each other for executing the method, operation, step or action described in the fourth aspect mentioned above.

[0091] In a possible implementation, the communication device includes: a transceiver unit (or a communication module), and a processing unit (or a processing module) connected to the transceiver unit.

[0092] Exemplarily, when it is determined that the first federated learning member cannot continue to perform the training task of the first model, and / or it is determined that the intermediate training result of the first model does not meet the second training requirement corresponding to the first model, the transceiver unit is used to send a first request message to the federated learning server, wherein the first request message includes the intermediate training result of the first model, the intermediate training result is used to determine the first training requirement corresponding to the first model, and the intermediate training result is obtained by the first federated learning member performing the training task of the first model; the transceiver unit is also used to receive a first response message from the federated learning server, the first response message includes information of the second federated learning member, and the second federated learning member supports continuing to perform the training task of the first model.

[0093] In the twelfth aspect, a communication device is provided, which has the function of implementing the fifth aspect. For example, the communication device includes modules or units or means corresponding to executing the operations involved in the fifth aspect. The modules or units or means can be implemented through software, or through hardware, or through a combination of software and hardware.

[0094] Exemplarily, the communication device may be the second federated learning member mentioned above, for example, it may be a module or unit (such as a chip, or a chip system, or a circuit) corresponding one to each other for executing the method, operation, step or action described in the fifth aspect mentioned above.

[0095] In a possible implementation, the communication device includes: a transceiver unit (or a communication module), and a processing unit (or a processing module) connected to the transceiver unit.

[0096] Exemplarily, the transceiver unit is used to receive a second request message from a first federated learning member or a federated learning server, the second request message is used to request to continue the training task of the first model, and the second request message includes the intermediate training result of the first model and the first training requirement corresponding to the first model; the processing unit is used to continue the training task according to the intermediate training result and the first training requirement to obtain the first model and the training information of the first model.

[0097] The transceiver unit is also used to send the first model and training information of the first model to the federated learning server.

[0098] In the thirteenth aspect, a communication device is provided, which has the function of implementing the sixth aspect. For example, the communication device includes a module or unit or means corresponding to the operation involved in the sixth aspect. The module or unit or means can be implemented by software, or by hardware, or by a combination of software and hardware.

[0099] Exemplarily, the communication device may be the above-mentioned federated learning server, for example, it may be a module or unit (such as a chip, or a chip system, or a circuit) corresponding one to one to executing the method, operation, step or action described in the sixth aspect above.

[0100] In a possible implementation, the communication device includes: a transceiver unit (or a communication module), and a processing unit (or a processing module) connected to the transceiver unit.

[0101] Exemplarily, a transceiver unit is used to receive a third request message from a model requester, the third request message is used to request execution of a training task for a first model, and the third request message includes a second training requirement corresponding to the first model; a processing unit is used to determine a first federated learning member and a second federated learning member from candidate federated learning members according to capability information and the second training requirement of the candidate federated learning members, the capability information indicating whether the candidate federated learning members support continuing to execute the training task for the first model, at least one of the candidate federated learning members supports continuing to execute the training task for the first model, at least one federated learning member includes a second federated learning member, and the second federated learning member is an alternative federated learning member for executing the training task for the first model; the transceiver unit is also used to send a fourth request message to the first federated learning member, the fourth request message is used to request execution of the training task for the first model, and the fourth request message includes the second training requirement corresponding to the first model and information of the second federated learning member.

[0102] In a fourteenth aspect, a communication device is provided, which has the function of implementing the seventh aspect. For example, the communication device includes a module or unit or means corresponding to the operation involved in the seventh aspect. The module or unit or means can be implemented by software, or by hardware, or by a combination of software and hardware.

[0103] Exemplarily, the communication device may be the first federated learning member mentioned above, for example, it may be a module or unit (such as a chip, or a chip system, or a circuit) corresponding one to each other for executing the method, operation, step or action described in the seventh aspect mentioned above.

[0104] In a possible implementation, the communication device includes: a transceiver unit (or a communication module), and a processing unit (or a processing module) connected to the transceiver unit.

[0105] Exemplarily, a processing unit is used to obtain information of a second federated learning member, and the second federated learning member supports continuing the training task of the first model; when it is determined that the first federated learning member cannot continue the training task of the first model, and / or it is determined that the intermediate training result of the first model does not meet the second training requirement corresponding to the first model, the processing unit is used to determine the first training requirement corresponding to the first model based on the intermediate training result and the second training requirement; the transceiver unit is used to send a second request message to the second federated learning member, the second request message is used to request to continue the training task, and the second request message includes the first training requirement and the intermediate training result.

[0106] In a fifteenth aspect, a communication device is provided. The communication device may be the above-mentioned federated learning server or the first federated learning member or the second federated learning member. The communication device includes a transceiver, a processor and a memory, the processor is used to control the transceiver to send and receive signals, the memory is used to store a computer program, and the processor is used to call and run the computer program from the memory, so that the communication device performs the method in any possible implementation of the first to seventh aspects above.

[0107] Optionally, there are one or more processors and one or more memories.

[0108] Optionally, the memory may be integrated with the processor, or the memory may be provided separately from the processor.

[0109] Optionally, the transceiver includes a transmitter (transmitter) and a receiver (receiver).

[0110] In the sixteenth aspect, a communication device is provided, wherein the communication device includes one or more processors, and the one or more processors are used to execute computer programs or instructions. When the computer program or instructions are executed, the communication device implements the method in any possible design or implementation method in the first to seventh aspects above.

[0111] Optionally, the communication device also includes a memory, which is used to store part or all of the computer programs or instructions for implementing the functions involved in the first to third aspects above.

[0112] Optionally, the communication device further includes an interface circuit, and the processor is used to communicate with other devices or components through the interface circuit.

[0113] In a seventeenth aspect, a communication system is provided. The communication system includes a federated learning server and / or a first federated learning member, wherein the federated learning server is used to execute the method in any possible implementation of the first aspect, the second aspect, or the sixth aspect, and the first federated learning member is used to execute the method in any possible implementation of the third aspect, the fourth aspect, or the seventh aspect.

[0114] Optionally, the communication system further includes a second federated learning member, wherein the second federated learning member is used to execute the method in any possible implementation manner of the fifth aspect above.

[0115] In an eighteenth aspect, a computer-readable storage medium is provided. The computer-readable storage medium stores computer program codes or instructions so that the method in any possible implementation of the first to seventh aspects is executed, for example, when a computer reads and executes the computer program codes or instructions, the method in any possible implementation of the first to seventh aspects is implemented.

[0116] In a nineteenth aspect, a computer program product is provided. The computer program product includes: computer program code or instructions, so that the method in any possible implementation of the first to seventh aspects is implemented. For example, when a computer reads and executes the computer program product, the method in any possible implementation of the first to third aspects is implemented.

[0117] In a twentieth aspect, a computer program is provided. When the computer program is executed, the method in any possible implementation manner of the first to seventh aspects is implemented.

[0118] It should be understood that the beneficial effects of the eighth to twentieth aspects mentioned above can be referred to the first to seventh aspects mentioned above and any possible implementation methods thereof, and will not be elaborated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0119] Figure 1 to Figure 2 is a schematic diagram of a communication system applicable to an embodiment of the present application;

[0120] Figure 3 It is a flowchart of a method for executing federated learning;

[0121] Figure 4 It is a flow chart of a method for transmitting an intermediate model;

[0122] Figure 5 It is a flow chart of a communication method provided in an embodiment of the present application;

[0123] Figure 6 It is a flowchart of another communication method provided in an embodiment of the present application;

[0124] Figure 7 It is a flowchart of another communication method provided in an embodiment of the present application;

[0125] Figure 8 It is a flowchart of another communication method provided in an embodiment of the present application;

[0126] Fig. 9 It is a flowchart of another communication method provided in an embodiment of the present application;

[0127] Fig.10 is a schematic block diagram of a communication device provided in an embodiment of the present application;

[0128] Fig.11 It is a schematic block diagram of another communication device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0129] The technical solution in this application will be described below in conjunction with the accompanying drawings.

[0130] To facilitate understanding of the embodiments of the present application, the following explanations are given.

[0131] (1) In this application, unless otherwise specified or there is a logical conflict, the terms and / or descriptions between different embodiments are consistent and can be referenced to each other. The technical features in different embodiments can be combined to form new embodiments based on their internal logical relationships.

[0132] (2) In this application, "at least one" means one or more, and "more than one" means two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone, where A and B can be singular or plural. In the text description of this application, the character " / " generally indicates that the associated objects before and after are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b and c can mean: a, or b, or c, or a and b, or a and c, or b and c, or a, b and c. Where a, b and c can be single or plural, respectively.

[0133] (3) In the present application, "first", "second" and various numerical references are used to distinguish for the convenience of description and are not used to limit the scope of the embodiments of the present application. For example, they are used to distinguish different messages, etc., rather than to describe a specific order or sequence. It should be understood that the objects described in this way can be interchanged where appropriate so as to be able to describe solutions other than the embodiments of the present application.

[0134] (4) In this application, "indicating" or "used to indicate" may include being used to indicate directly or indirectly. When describing that a certain indication information is used to indicate A, it may include that the indication information directly indicates A or indirectly indicates A, but it does not mean that the indication information must contain A.

[0135] The indication method involved in the embodiments of the present application should be understood to include various methods that can enable the party to be indicated to know the information to be indicated. The information to be indicated can be sent as a whole or divided into multiple sub-information and sent separately, and the sending period and / or sending time of these sub-information can be the same or different. This application does not limit the sending method, for example.

[0136] The "indication information" in the embodiments of the present application may be an explicit indication, i.e., directly indicated by signaling, or obtained by combining other rules or other parameters or by deduction according to the parameters indicated by the signaling. It may also be an implicit indication, i.e., obtained by combining other rules or other parameters or by deduction according to a rule or relationship. The present application does not make any specific limitation on this.

[0137] (5) In this application, "protocol" may refer to a standard protocol in the field of communications, for example, including (5 thThe present application does not limit the following: "pre-defined" may include pre-defined. For example, protocol definition. "Pre-configured" may be implemented by pre-saving corresponding codes, tables or other methods that can be used to indicate relevant information in the device, and the present application does not limit the implementation method thereof.

[0138] (6) In the present application, the terms “message”, “information”, “signal” or “information element (IE)” can be used interchangeably. There is no limitation on the name of the message or information, as long as the corresponding function can be realized.

[0139] "Sending information to XX (device)" can be understood as the destination of the information is the device. It can include sending information to the device directly or indirectly. "Receiving information from XX (device), or receiving information from XX (device)" can be understood as the source of the information is the device, which can include receiving information from the device directly or indirectly. The information may be processed as necessary between the source and destination of the information, such as format changes, but the destination can understand the valid information from the source. Similar expressions in this application can be understood similarly and are not explained here.

[0140] "Communication" can also be described as "communication", "information transmission", "data processing", etc. "Transmission" includes "sending" and "receiving". "Transmission" can be described as "output". "Sending" can also be understood as the "output" of the chip interface, and "receiving" can also be understood as the "input" of the chip interface. In other words, "sending" or "receiving" can be carried out between devices, for example, a network device and a terminal device send or receive respectively through an air interface, and "sending" or "receiving" can also be carried out within a device, for example, sending or receiving between components, modules, chips, software modules or hardware modules within the device through a bus, a line or an interface.

[0141] (7) In this application, words such as "exemplarily" and "such as" are used to indicate examples, illustrations or descriptions. Any embodiment or design described as an "example" in this application should not be interpreted as being preferred or advantageous over other embodiments or designs. Specifically, the use of the word "example" is intended to present concepts in a concrete way. In the embodiments of this application, "of", "corresponding, relevant", "correspondingly" and "associate" can sometimes be used interchangeably. It should be noted that when the distinction is not emphasized, the meanings to be expressed are consistent.

[0142] (8) In the present application, configuration may be signaling configuration, and may also be described as configuration signaling. For example, signaling configuration includes configuration by signaling sent by a base station, and these signaling may be radio resource control (RRC) messages, downlink control information (DCI), or system information block (SIB). Optionally, the signaling configuration may also be configured to the terminal device by pre-configured signaling, or configured to the terminal device by pre-configuration. The pre-configuration here means defining or configuring the values ​​of corresponding parameters in advance in a protocol manner, and storing them in the terminal device when communicating with the terminal device. The pre-configured message can be modified or updated when the terminal device is connected to the network.

[0143] The technical solution provided in this application can be applied to various communication systems, such as: new radio (NR) system, long term evolution (LTE) system, LTE frequency division duplex (FDD) system, LTE time division duplex (TDD) system, etc. The technical solution provided in this application can also be applied to device to device (D2D) communication, vehicle to everything (V2X) communication, machine to machine (M2M) communication, machine type communication (MTC), and Internet of things (IoT) communication system or other communication systems.

[0144] In the communication system, the part operated by the operator can be called a public land mobile network (PLMN), or an operator network, etc. PLMN is a network established and operated by the government or an operator approved by it for the purpose of providing land mobile communication services to the public. It is mainly a public network in which mobile network operators (MNO) provide mobile broadband access services to users. The PLMN described in the embodiment of the present application may specifically be a network that meets the requirements of the third generation partnership project (3GPP) standards, referred to as a 3GPP network. 3GPP networks generally include but are not limited to fifth-generation mobile communication (5th-generation, 5G) networks, fourth-generation mobile communication (4th-generation, 4G) networks, and other future communication systems, such as sixth-generation mobile communication (6th-generation, 6G) networks, etc.

[0145] For ease of description, the embodiments of the present application will be described using PLMN or 5G network as an example.

[0146] Figure 1 A schematic diagram of a network architecture 100 is shown, taking the 5G network architecture based on service-based architecture (SBA) in a non-roaming scenario defined in the 3GPP standardization process as an example. Figure 1 As shown, the network architecture may include a terminal device part, a data network (DN) part and an operator network PLMN part. The operator network PLMN part may include but is not limited to a (radio) access network ((R)AN) 120 and a core network (CN) part.

[0147] The functions of each network element are briefly described below.

[0148] The terminal equipment part may include a terminal equipment 110, which is a device that provides voice and / or data connectivity to a user. The terminal equipment 110 may also be referred to as user equipment (UE). The terminal equipment 110 in the present application is a device with wireless transceiver function, which can communicate with one or more core network (CN) devices via an access network device (or also referred to as an access device) in a (radio) access network (R) AN 120. The terminal equipment 110 may also be referred to as an access terminal, a terminal, a user unit, a user station, a mobile station, a mobile station, a remote station, a remote terminal, a mobile device, a user terminal, a user agent or a user device, etc. The terminal equipment 110 may be a cellular phone, a cordless phone, a session initiation protocol (SIP) phone, a smart phone, a mobile phone, a wireless local loop (WLL) station, a personal digital assistant (PDA), etc. Alternatively, the terminal device 110 may also be a handheld device with wireless communication function, a computing device or other device connected to a wireless modem, a vehicle-mounted device, a wearable device, a drone device or an Internet of Things, a terminal in the Internet of Vehicles, a 5G network and any form of terminal in the future network, a relay user device or a terminal in the future evolved 6G network, etc. Among them, the relay user device may be, for example, a 5G residential gateway (RG). For example, the terminal device 110 may be a virtual reality (VR) terminal, an augmented reality (AR) terminal, a wireless terminal in industrial control, a wireless terminal in unmanned driving, a wireless terminal in telemedicine, a wireless terminal in a smart grid, a wireless terminal in transportation safety, a wireless terminal in a smart city, a wireless terminal in a smart home, etc. The terminal device here refers to a 3GPP terminal. The embodiment of the present application does not limit the type or type of the terminal device. For ease of explanation, the present application will be described later using UE as an example to refer to the terminal device.

[0149] (R)AN 120 may include one or more access network elements or access network devices. The interface between the access network device and the terminal device may be a Uu interface (or air interface, i.e., the messages exchanged between the access network device and the terminal device may be called air interface messages). In future communications, the interface name may remain unchanged or may be replaced by other names, which is not limited in this application. (R)AN 120 is a device that provides wireless communication functions for the terminal device 110, which can access the terminal device to a node or device of a wireless network, and may also be called a network device. The above-mentioned RAN may be a 3GPP-related cellular system, such as a 5G mobile communication system, or a future-oriented evolution system (such as a 6G mobile communication system). RAN may also be an open radio access network (open RAN, O-RAN or ORAN), a cloud radio access network (cloud radio access network, CRAN), or a wireless fidelity (wireless fidelity, WiFi) system. (R)AN 120 may be regarded as a subnetwork of the operator network, and is an implementation system between a service node and a terminal device 110 in the operator network. For example, the terminal device 110 can be connected to the service node of the operator network through the (R)AN 120, so as to obtain the service provided by the service node. (R)AN 120 includes but is not limited to: the next generation node base station (gNB) in the 5G system, the evolved node B (eNB) in LTE, the radio network controller (RNC), the node B (NB), the base station controller (BSC), the base transceiver station (BTS), the home base station (for example, home evolved node B, or homenode B, HNB), the base band unit (BBU), the transmission point (TRP), the transmission point (TP), the small base station equipment, the mobile switching center, or the network equipment in the future network. The access network equipment can also be a module or unit that completes the base station function, for example, including a centralized unit (CU) and a distributed unit (DU). The embodiments of the present application do not limit the specific technology and specific equipment form adopted by the access network equipment. In systems using different wireless access technologies, the names of devices with access network device functions may be different.For the convenience of description, in all embodiments of the present application, the above-mentioned apparatuses providing wireless communication functions for the terminal device 110 are collectively referred to as access network equipment or RAN for short. It should be understood that the specific type of access network equipment is not limited herein.

[0150] In different systems, CU (including open CU-CP (open CU-CP, O-CU-CP) and open CU-UP (openCU-UP, O-CU-UP), DU or RU may also have different names, but those skilled in the art can understand their meanings. For example, in the ORAN system, CU may also be called an open central unit (open central unit, O-CU), DU may also be called an open distributed unit (open distributed unit, O-DU), CU-CP may also be called O-CU-CP, CU-UP may also be called O-CU-UP, and RU may also be called an open radio unit (open radio unit, O-RU). For the convenience of description, CU, CU-CP, CU-UP, DU and RU are described as examples in this application. Any of the CU (or CU-CP, CU-UP), DU and RU in this application may be implemented by a software module, a hardware module, or a combination of a software module and a hardware module.

[0151] The terminal device 110 or (R)AN 120 can be deployed on land, including indoors or outdoors, handheld or vehicle-mounted; it can also be deployed on the water surface (such as a ship, etc.); it can also be deployed in the air (such as an airplane, a balloon, and a satellite, etc.).

[0152] The CN part may include but is not limited to the following network functions (NF): user plane function (UPF) 130, network exposure function (NEF) 131, network function repository function (NRF) 132, policy control function (PCF) 133, unified data management function (UDM) 134, unified data repository function (UDR) 135, network data analytics function (NWDAF) 136, application function (AF) 141, authentication server function (AUSF) 137, access and mobility management function (AMF) 138, session management function (SMF) 139.

[0153] The data network DN 140, also referred to as a packet data network (PDN), is usually a network located outside the operator's network, such as a third-party network.

[0154] The following is a brief description of the NF functions included in CN.

[0155] 1. UPF 130 is a gateway provided by the operator and is the gateway for the operator network to communicate with DN 140. UPF 130 network functions include data packet routing and transmission, data packet detection, service usage reporting, quality of service (QoS) processing, legal monitoring, uplink data packet detection, downlink data packet storage and other user-related functions.

[0156] 2. NEF 131 is a control plane function provided by the operator, which mainly enables third parties to use the services provided by the network, supports the network to open its capabilities, event and data analysis, provide PLMN security configuration information from external applications, and convert interactive information inside and outside the PLMN.

[0157] 3. NRF 132 is a control plane function provided by the operator and can be used to maintain real-time information of network functions and services in the network.

[0158] 4. PCF 133 is a control plane function provided by the operator. It mainly supports providing a unified policy framework to control network behavior, provides policy rules to the control layer network function, and is responsible for obtaining user contract information related to policy decisions.

[0159] 5. UDM 134 is a control plane function provided by the operator, responsible for storing information such as the subscriber permanent identifier (SUPI) of the subscriber in the operator's network, the generic public subscription identifier (GPSI) of the subscriber, and credentials.

[0160] 6. UDR 135 is a control plane function provided by the operator, which provides the UDM with the function of saving and obtaining subscription data, the PCF with the function of saving and obtaining policy data, and the user's NF group ID information.

[0161] 7. NWDAF 136 is a control plane function provided by the operator, which has functions such as data collection, model training, data analysis, and model reasoning. Among them, the NWDAF network element containing the analysis logical function (AnLF) can be used to infer and export analysis information and expose analysis services. Analysis can refer to statistical information and / or predictions generated or provided according to the request of the analysis consumer (consumer). The NWDAF network element containing the model training logical function (MTLF) can be used to train machine learning (ML) models or artificial intelligence (AI) models, and expose new training services, such as providing trained AI models or ML models to AnLF. For the process of obtaining model-related data, AnLF can act as a data producer (producer) network element; MTLF can act as a data consumer network element.

[0162] Currently, AnLF can request a model from MTLF through a model subscription (MLModelProvision_Subscribe) service or message, and the model can be obtained by MTLF training based on relevant data of the model (such as samples). In addition, AnLF can use a data set tag to specify a data set, and the data set can contain data related to a model. Therefore, a data set tag can be used to mark a data set, that is, to mark data related to the model. The data set tag can also be understood as an index of a data set or an index of data. MTLF can act as a consumer network element of model-related data to obtain data stored in a data store by a data producer network element, wherein the data producer network element can be AnLF, and the data storage network element can be a network element used to store data such as an analytics data repository functional (ADRF). In this application, relevant data can be understood as data used for vertical federated learning, such as input data, training data, inference data, model parameters or sample data, etc.

[0163] 8. AF 141 is a control plane function provided by the operator. It mainly provides corresponding services by interacting with other NFs in the PLMN, such as providing roaming UE with access network selection information, guiding the routing of data flows, and accessing NEF131.

[0164] 9. AUSF 137 is a control plane function provided by the operator and is usually used for level 1 authentication, i.e. authentication between the terminal device 110 (subscriber) and the operator network.

[0165] 10. AMF 138 is a control plane network function provided by the operator network, responsible for access control and mobility management of the terminal device 110 accessing the operator network, such as mobility status management, allocation of user temporary identity, authentication and authorization of users, etc.

[0166] 11. SMF 139 is a control plane network function provided by the operator network, responsible for managing the protocol data unit (PDU) session of the terminal device 110 (including session establishment, modification and release), and is used for the selection and reselection of user plane function network elements, the allocation of Internet protocol (IP) addresses of terminal devices, and quality of service (QoS) control.

[0167] It is understandable that the above network elements or functions can be physical entities in hardware devices, software instances running on dedicated hardware, or virtualized functions instantiated on a shared platform (e.g., a cloud platform). In short, an NF can be implemented by hardware or software.

[0168] Figure 1 Where Nnef, Nnrf, Npcf, Nudm, Nudr, Nnwdaf, Naf, Nausf, Namf, Nsmf, N1, N2, N3, N4, and N6 are interface serial numbers. For example, the meaning of the above interface serial numbers can be found in the meaning defined in the 3GPP standard protocol, and this application does not limit the meaning of the above interface serial numbers. It should be noted that, Figure 1 The interface name between the various network functions in the system architecture is only an example, and the interface name of the system architecture may also be other names, which are not limited in this application. In addition, the name of the message (or signaling) transmitted between the above-mentioned network elements is only an example and does not constitute any limitation on the function of the message itself.

[0169] It should be noted that in Figure 1 In the architecture shown, the interface between the (R)AN and the CN may also be referred to as the NG interface (not shown in the figure), and the (R)AN and the CN are connected via the NG interface. The NG interface may include an NG-C interface and an NG-U interface, wherein the NG-C interface is a control plane interface, and the two connected parties are the (R)AN and the AMF, and is used to transmit control plane data; the NG-U interface is a user plane interface, and the two connected parties are the (R)AN and the UPF, and is used to transmit user plane data.

[0170] It should be understood that the above network architecture 100 is only a network architecture described from the perspective of service-oriented architecture. In this service-based architecture, PLMN can combine some or all network functions in an orderly manner according to specific scenario requirements to achieve customization of network capabilities and services, thereby deploying dedicated networks for different services, that is, realizing 5G network slicing. Network slicing technology can enable operators to respond to customer needs more flexibly and quickly, and support flexible allocation of network resources.

[0171] It should also be understood that the above naming is only defined for the convenience of distinguishing different functions and should not constitute any limitation to this application. This application does not exclude the possibility of using other naming in 5G networks and other future networks. For example, in a 6G network, some or all of the above network elements may use the terminology in 5G, or may use other names.

[0172] Figure 2 is a schematic diagram of a network architecture 200. Figure 2 As shown, in the artificial intelligence / machine learning (AI / ML) architecture, an artificial intelligence / machine learning enabling client AI / ML enabler client (or AI / MLenablement client) is deployed on the UE, and the AI / ML enabler client has AI / ML functions, including federated learning local model training, and / or, model reasoning and other functions. An artificial intelligence / machine learning enabling server AI / MLenabler server (or AI / ML enablement server) is deployed on the network side, and the AI / ML enabler server has AI / ML functions, including AI / ML member selection functions, and / or, allocating federated learning tasks and assigning tasks to the AI / MLenabler client. Optionally, a central server AI / ML central server can also be deployed in the network architecture, and the AI / ML central server can be used to store trained AI / ML models. For example, the AI / ML centralserver is connected to the AI / ML enablement server, so that the AI / ML enablement server reports the trained AI / ML model to the AI / ML central server.

[0173] As an example, an AI / ML enablement server requests at least one AI / ML enablement client to perform model training, at least one AI / ML enablement client performs model training locally to obtain model training results, and reports the obtained model training results to the AI / ML enablement server, and the AI / ML enablement server aggregates the model training results from at least one AI / ML enablement client to obtain a trained AI / ML model, and reports the trained AI / ML model to the AI / ML central server for storage. Optionally, the trained AI / ML model can be used for applications such as network performance analysis, network abnormal behavior detection, or face recognition.

[0174] Optionally, the present application Figure 2 The specific names of the network elements involved in the network architecture are not limited.

[0175] For ease of understanding, the following is an introduction to the relevant terms, concepts, or technologies that may be involved in the embodiments of the present application:

[0176] 1. Federated learning;

[0177] Federated learning (FL) is a distributed machine learning method in which multiple participants exchange model parameters through a security mechanism without interacting or sharing original training data, thereby achieving collaborative training effects. In other words, federated learning is an encrypted distributed machine learning technology. Federated learning can make full use of the data and computing power of the participants, allowing multiple parties to collaboratively build a general and robust machine learning model without sharing data. Therefore, federated learning can effectively help multiple institutions use data and conduct learning modeling while meeting the requirements of user privacy protection, data security, and government regulations. It can be said that federated learning is designed to share knowledge and parameters without exchanging any of its own data.

[0178] Federated learning includes horizontal federated learning and vertical federated learning. In an environment where data regulation is becoming increasingly stringent, federated learning can solve key issues such as data ownership, data privacy, data access rights, and access to heterogeneous data. A horizontal row of a data matrix represents a training sample, and a vertical column represents a data feature. Horizontal federated learning is to conduct federated learning by combining multiple rows of samples with the same features from multiple participants, that is, the training data of each participant is divided horizontally. Horizontal federated learning is also called feature-aligned federated learning, that is, the data features of the participants are aligned, and horizontal federated learning can increase the total amount of training samples. FL, as a machine learning technology, can be used to solve model training and reasoning when each member is unwilling to share the original data. It is suitable for situations where the identification (ID) of the training samples of the participants overlaps a lot, but the data features of the participants overlap less. FL combines different data features of the common samples of multiple participants for federated learning, that is, the training data of each participant is divided vertically, so it is called vertical federated learning. Vertical federated learning is to conduct federated learning by combining different data features of the common samples of multiple participants, that is, the training data of each participant is divided vertically. Vertical federated learning is also called sample-aligned federated learning, that is, the training samples of the participants are aligned. Vertical federated learning can increase the feature dimension of training data.

[0179] 2. Analysis ID or analytics ID;

[0180] The analysis identifier can be used to indicate an analysis business or an analysis service, or simply a service. The service is associated with a model, that is, the model can be used to perform the service; or it can be described as, the analysis identifier is associated with the model, that is, the model is used to perform the service corresponding to the analysis identifier.

[0181] Or it can also be understood that MTLF is related to the analysis identifier, that is, the model support provided by MTLF is used to execute the service corresponding to the analysis identifier. Exemplarily, MTLF can be related to one or more analysis identifiers. It can be understood that the MTLF can provide a model of the service corresponding to each analysis identifier in one or more analysis identifiers. For example, MTLF1 is related to analysis identifier 1 and analysis identifier 2, that is, MTLF1 corresponds to analysis identifier 1 and analysis identifier 2, then MTLF1 can provide a model of the service corresponding to analysis identifier 1, and a model of the service corresponding to analysis identifier 2.

[0182] The above briefly describes the terms involved in this application, which will not be repeated in the following embodiments. In addition, the above description of terms is only for the sake of ease of understanding and does not limit the protection scope of the embodiments of this application.

[0183] Currently, the AI / ML enabler server can start the federated learning process, that is, the AI / ML enabler server and the AI / ML enabler client can perform federated learning training. For the specific process of federated learning training, see the following Figure 3 For ease of description and understanding, Figure 3 In this paper, AI / ML enabler server is used as server NWDAF and AI / ML enabler client is used as client NWDAF as an example. NRF and NF can be regarded as Figure 2 A network element in a 3GPP network system.

[0184] Figure 3 It is a flowchart of a federated learning execution method, such as Figure 3 As shown, the process includes the following steps, and the parts not described in detail can refer to the existing protocols.

[0185] S301, the consumer sends a subscription request message #1 to the server NWDAF (eg, FL server NWDAF).

[0186] Correspondingly, the server NWDAF receives the subscription request message #1 from the consumer.

[0187] The subscription request message #1 is used to subscribe to ML model provisioning or training. Exemplarily, a consumer (e.g., NWDAF including AnLF, or NWDAF including MTLF) uses the Nnwdaf_MLModelProvision service to send a subscription request message #1, such as subscription request for ML model provisioning / training, to the server NWDAF to retrieve the ML model.

[0188] The subscription request message #1 includes one or more of the following: analysis ID, ML model metric (e.g., ML model accuracy), accuracy reporting interval, and predetermined status (ML model accuracy threshold or time when ML model is required). It should be understood that the ML model accuracy threshold can be used to indicate the target ML model accuracy of the training process, and when the ML model accuracy threshold is reached during the training process, the server NWDAF can stop the training process. If the consumer provides the time when the ML model is required, the server NWDAF can consider this information to determine the maximum response time of its client NWDAF.

[0189] S302: The server NWDAF determines the client NWDAF(s).

[0190] Exemplarily, the server NWDAF can perform the discovery process of the client NWDAF with the NRF, and the specific implementation method can refer to the existing relevant description. For example, the client NWDAF (s) registers with the NRF, and the server NWDAF sends a discovery request message (Nnrf_NFDiscovery_Request) to the NRF to request the discovery of at least one client NWDAF that performs the federated learning task, and the NRF authorizes, for example, the NRF decides whether to allow the server NWDAF to discover the desired NF instance based on the type of the desired NF / NF service and the type of the server NWDAF, and the NRF sends a discovery response message (Nnrf_NFDiscovery_Response) to the server NWDAF, and the discovery response message includes the information of the client NWDAF (s) determined by the server NWDAF. Among them, the client NWDAF (s) can be one or more, such as client NWDAF1, ..., client NWDAF N (such as FL client NWDAF), without limitation.

[0191] S303, the server NWDAF sends a subscription request message #2 to the client NWDAF(s).

[0192] Correspondingly, the client NWDAF(s) receives the subscription request message #2 from the server NWDAF.

[0193] Exemplarily, the subscription request message #2 may be Nnwdaf_MLModelTraining_Subscribe, for example, the server NWDAF sends Nnwdaf_MLModelTraining_Subscribe or Nnwdaf_MLModelTrainingInfo_Request to the client NWDAF(s), requesting to perform local model training.

[0194] The subscription request message #2 carries one or more of the following: initial federated learning parameters provisioning, ML model metrics, initial ML model, or maximum response time, wherein the maximum response time refers to the maximum response time for the client NWDAF to report temporary local ML model information to the server NWDAF.

[0195] S304: Optionally, the client NWDAF(s) collects data.

[0196] Illustratively, if the client NWDAF does not have local data available yet, each client NWDAF may collect its local data from the NF (data provider) using the current mechanism in TS23.288.

[0197] S305, the client NWDAF(s) sends a subscription response message #2 to the server NWDAF.

[0198] Correspondingly, the server NWDAF receives the subscription response message #2 from the client NWDAF(s).

[0199] Exemplarily, subscription response message #2 may be Nnwdaf_MLModelTraining_Notify, which is used to report the training information of the local model. For example, during the federated learning training process, each client NWDAF trains the initial ML model provided by the server NWDAF based on its own data, and reports the temporary local ML model information to the server NWDAF in Nnwdaf_MLModelTraing_Notify or Nnwdaf_MLModelTraingInfo_Response.

[0200] Optionally, Nnwdaf_MLModelTraining_Notify or Nnwdaf_MLModelTrainingInfo_Response may also include local ML model metrics calculated by the client NWDAF(s) and training input data information (e.g., the area covered by the dataset, sampling ratio, maximum / minimum value of each dimension of the data, etc.).

[0201] Optionally, the ML model is sent from the client NWDAF(s) to the server NWDAF during the federated learning training process and is the information required by the server NWDAF to build an aggregate model based on the locally trained ML model.

[0202] Optionally, if the client NWDAF cannot complete the training of the temporary local ML model within the maximum response time provided by the server NWDAF, the client NWDAF(s) may send a delay event notification, including a delay event indication, an optional cause code (e.g., local ML model training failed, or more time is required for local ML model training), and the expected time available for the client NWDAF to complete the training before the maximum response time elapses, etc. That is, the client NWDAF(s) may report the temporary local model, or the server NWDAF may notify the client NWDAF(s) to skip this round of training.

[0203] S306, optionally, the server NWDAF sends a subscription request message #3 to the client NWDAF(s).

[0204] Correspondingly, the client NWDAF(s) receives the subscription request message #3 from the server NWDAF.

[0205] Exemplarily, the subscription response message #3 may be Nnwdaf_MLModelTraining_Notify, carrying an extended response time and / or current iteration round ID. For example, if the server NWDAF receives a notification / response that the client NWDAF(s) cannot complete the training within the maximum response time, the server NWDAF may send an extended maximum response time_MLModel TrainingInfo_Request to the client NWDAF in Nnwdaf_MLModelTraining_Subscribe or Nnwdaf, before which the client NWDAF needs to report the temporary local ML model information to the server NWDAF. Otherwise, the server NWDAF may instruct the client NWDAF to skip the report of this iteration. The client NWDAF includes the current iteration round ID in the request message to indicate that this request is for modifying the training parameters of the current iteration round.

[0206] Optionally, the server NWDAF may notify the client NWDAF to stop ML model training by sending a termination request and report the current local ML model update.

[0207] S307: The client NWDAF performs model aggregation.

[0208] Exemplarily, the client NWDAF aggregates all local ML model information retrieved in step S305 to update the global ML model. Optionally, the server NWDAF may also calculate global ML model metrics, for example based on local ML model metrics, or by applying the global model on a validation dataset (if available). The server NWDAF may update the global ML model each time the client NWDAF provides updated local ML model information, or the server NWDAF may decide to wait for local ML model information from all client NWDAFs before updating the global ML model.

[0209] If the server NWDAF provides a maximum response time for the client NWDAF to provide temporary local ML model information in step S303, or provides an extended maximum response time in step S306, the server NWDAF decides to wait for the client NWDAF that has not provided its temporary local ML model within the (extended) maximum response time, or to aggregate only the retrieved local ML model information instances to update the global ML model. The server NWDAF makes a decision based on the notification / response of the client NWDAF, or if no notification is received, makes a decision based on the local configuration.

[0210] S308, optionally, the client NWDAF sends an update message to the consumer.

[0211] Correspondingly, the consumer receives update messages from the client NWDAF.

[0212] The update message is used to indicate the current ML training status to the consumer. Exemplarily, in response to step S301, the client NWDAF sends a Nnwdaf_MLModelProvision_Notify message to the consumer to dynamically update the global ML model metrics to the consumer periodically (e.g., a certain number of training rounds or every 10 minutes) or when certain predetermined states are reached (e.g., the ML model accuracy threshold is reached or the training time expires).

[0213] S309, optionally, the consumer sends a subscription request message #4 to the server NWDAF.

[0214] Correspondingly, the server NWDAF receives a subscription request message #4 from the consumer, wherein the subscription request message #4 is used to modify the subscription to update or terminate.

[0215] Exemplarily, the consumer decides whether the current model can meet the requirements, for example, the global ML model metric satisfies the consumer, and decides to stop or continue the training process. The user can re-invoke the Nnwdaf_MLModelProvision_Subscribe service operation used in step S301 to stop or continue the training process.

[0216] S310, optionally, the server NWDAF updates or terminates the federated learning training process.

[0217] Exemplarily, the server NWDAF updates or terminates the current federated learning training process according to the subscription request message #4 sent by the consumer in step S309. Optionally, if the server NWDAF receives a request to stop the federated training process in step S309, the following steps S311 and S312 are skipped.

[0218] S311, optionally, the server NWDAF sends aggregation model information to the client NWDAF(s).

[0219] Correspondingly, the client NWDAF(s) receives the aggregation model information from the server NWDAF.

[0220] In other words, if the federated learning training process continues, the server NWDAF will determine the client NWDAF and send Nnwdaf_MLModelTraingInfo_Request including the aggregated ML model information to the selected client NWDAF(s) for the next round of federated training.

[0221] S312: Optionally, the client NWDAF(s) updates the local model according to the aggregated model information.

[0222] Exemplarily, each client NWDAF updates its local ML model according to the aggregated ML model information distributed by the server NWDAF in step S311 .

[0223] It should be noted that the above steps S304-S312 are repeated until a training termination condition (for example, a maximum number of iterations, or a result of a loss function is lower than a threshold) is reached.

[0224] When the federated training process is completed, the server NWDAF requests the client NWDAF(s) to terminate the federated learning training process. In one implementation, the server NWDAF calls the Nnwdaf_MLModelTraing_Unsubscribe service with the reason code "The federated learning process has been completed" and can choose to use the final aggregated ML model information. Then, the client NWDAF(s) terminates the local model training. If the final aggregated ML model information is received from the server NWDAF, the client NWDAF(s) can store the aggregated ML model information for further use.

[0225] Based on the above Figure 3 In the federated learning training process shown, there may be a situation where the client NWDAF (for example, one or more of the client NWDAF1, ..., client NWDAF N) cannot complete the temporary local ML model training within the maximum response time provided by the server NWDAF. For the convenience of description, the client NWDAFs that cannot complete the temporary local ML model training can be collectively referred to as source AI / ML members (member). In this case, the source AI / ML can request that the intermediate training results or intermediate model information obtained from the local training be passed to other AI / ML members to facilitate other AI / ML members to continue to perform the model training task. For the convenience of description, the federated learning member that obtains the intermediate training results or intermediate model information from the source AI / ML can be referred to as the target AI / ML member (member).

[0226] Figure 4 is a flow chart of a method for transmitting an intermediate model, such as Figure 4 As shown, the process includes the following steps, and the parts not described in detail can refer to the existing protocols.

[0227] S401, the source AI / ML member sends a request message to an AI / ML enabler server (AI / ML enabler server), and correspondingly, the AI / ML enabler server receives the request message from the source AI / ML member.

[0228] The request message carries requirement information, including at least one of the following: operation type (ML model training), intermediate model information (such as ML model information, which may include model status, or operation results), data transmission requirements (such as data volume, or completion time of data transmission, or resources required for data transmission), or indication information for instructing the AI / ML enabler server to perform transmission.

[0229] Exemplarily, when the source AI / ML member needs to transmit intermediate model information, the source AI / ML member sends a request message carrying requirement information to the AI / ML enabler server, requesting the AI / ML enabler server to perform the transmission.

[0230] S402, the AI / ML enabler server discovers the target AI / ML member.

[0231] Exemplarily, the AI / ML enabler server discovers the target AI / ML member according to the requirement information. The specific discovery process can be referred to the existing related description and will not be described here.

[0232] S403, the AI / ML enabler server sends a request message to the target AI / ML member, and correspondingly, the target AI / ML member receives the request message from the AI / ML enabler server.

[0233] S404, target AI / ML member authentication and authorization, for details, please refer to the existing relevant descriptions and will not be explained here.

[0234] S405, the target AI / ML member sends a response message to the AI / ML enabler server, and correspondingly, the AI / ML enabler server receives the response message from the target AI / ML member.

[0235] The response message is used to indicate that the target AI / ML member supports receiving the intermediate model information from the source AI / ML member.

[0236] S406, the AI / ML enabler server determines the transmission mode.

[0237] Exemplarily, the transmission mode includes: transiting through an AI / ML enabler server, that is, the AI / ML enabler server can send the intermediate model information from the source AI / ML member to the target AI / ML member; or, the intermediate model information can be directly transmitted between the source AI / ML member and the target AI / ML member.

[0238] S407, the AI / ML enabler server generates AI / ML auxiliary information, for example, transmission time window information, which is used to indicate the moment or time period for transmitting the intermediate model information.

[0239] S408, the intermediate model information is transmitted between the source AI / ML member and the target AI / ML member.

[0240] Method 1: The AI / ML enabler server sends the target AI / ML member’s information to the source AI / ML member; the source AI / ML model sends the intermediate model information directly to the target AI / ML member.

[0241] Method 2: The source AI / ML member sends the intermediate model information to the AI / ML enabler server, and then the AI / ML enabler server forwards the intermediate model information to the target AI / ML member. This can be transparent transmission without limitation.

[0242] As a machine learning technology in the core network, federated learning FL supports multiple participants (for example, server NWDAF and client NWDAF) to train locally without exchanging training data and exchange model parameters through a security mechanism, thereby achieving the effect of collaborative training. For example, during the federated learning process, the client NWDAF can train a local model based on local data and send the information of the local model to the server NWDAF. The server NWDAF aggregates the local models obtained from multiple client NWDAFs, and then sends the aggregated model information to each client NWDAF participating in the federated learning, and continues to train the model based on local data, iterating repeatedly until the federated learning is completed to obtain a global model. At this point, the server NWDAF can send the global model generated by the federated learning to the client NWDAF, and the client NWDAF can store the information of the global model for further use.

[0243] However, when the client NWDAF is unable to complete the training, if the training is skipped directly or a temporary local model is reported, the overall model performance may be affected. In addition, if the client NWDAF is unable to continue the FL model training due to resource or performance issues, if the server NWDAF does not select a new client NWDAF for training, it may lead to increased overall model convergence time, decreased performance and other issues. Furthermore, when the source AI / ML member (or the source client NWDAF) is unable to complete the training task, when selecting the target AI / ML member (or the target client NWDAF), if only the data transmission requirements for transmitting the current intermediate model are considered without considering the current task training situation, an inappropriate target AI / ML member may be selected, resulting in the target AI / ML member being unable to complete the training.

[0244] In view of this, the present application provides a communication method and a communication device, wherein a federated learning server determines a second federated learning member based on the capability information of the candidate federated learning members and the first training requirement, that is, when selecting a suitable federated learning member to continue the training task, the current training situation or training requirements are taken into consideration, thereby avoiding an increase in the model convergence time and ensuring the model performance.

[0245] The communication method provided in the embodiment of the present application will be described in detail below with reference to the accompanying drawings, and can be applied to the above Figures 1 to 3 It should be understood that the embodiments of the present application can be applicable to the scenario of communication between a sending end and a receiving end.

[0246] It should also be understood that the embodiments shown below do not specifically limit the specific structure of the execution subject of the method provided in the embodiments of the present application, as long as it is possible to communicate according to the method provided in the embodiments of the present application by running the code or program that records the method provided in the embodiments of the present application. For example, the method provided in the embodiments of the present application can be executed by a federated learning server, a first federated learning member, or a second federated learning member. Unless otherwise specified, the "federated learning server, the first federated learning member, or the second federated learning member" in this application can refer to the device itself, or a component in the device (for example, a communication module, a processor, a circuit, a chip (such as a modem chip, also known as a baseband chip, or a system-on-a-chip (SoC) or system-in-a-package (SIP) chip containing a modem core, or a chip system, etc.), or it can also be a logic module or software that can realize all or part of the device functions.

[0247] Figure 5 is a flow chart of a communication method provided in an embodiment of the present application. Figure 5 As shown, the method includes the following steps, which are illustrated by taking the federated learning server, the first federated learning member, and the second federated learning member as the execution entities. For the parts not described in detail, please refer to the relevant description of the existing protocol.

[0248] S510, the first federated learning member sends a first request message to the federated learning server;

[0249] Correspondingly, the federated learning server receives the first request message from the federated learning server.

[0250] Optionally, the first request message may be an AI / ML member discovery request message, and the specific message name is not limited.

[0251] In the present application, the federated learning server may be an AI / ML server, or an AI / ML enabler server (AI / MLenabler server or AI / ML E server), or a server NWDAF, that is, a Server NWDAF including MTLF, a NWDAF with the capabilities of a federated learning server. For example, the federated learning server supports at least one of the following: discovering a federated learning member (FL member) or an AI / ML enabler client (AI / ML enabler client or AI / ML E client) or an AI / ML enabler server (enabler server) or a client device (such as a client NWDAF); requesting the client device to perform local model training and report information about the local model, specifically, sending a basic model or an initial model to the client device, thereby requesting the client device to train the basic model or the initial model using local data to obtain a local model, and report information about the local model; aggregating information about the local model from the client device and generating a global model; aggregating model information after basic model training from the client device and generating a global model; returning the global model to the client device to implement iterative training.

[0252] In the present application, the first federated learning member may be an AI / ML client, or an AI / ML enabler client (AI / ML enabler client or AI / ML E client), or a client NWDAF, or an AI / ML enabler client (enabler client), or an AI / ML enabler server (enabler server). Among them, the first federated learning member supports the execution of the training task of the first model. Here, the first federated learning member supports the execution of the training task of the first model, which can be understood as: the first federated learning member (or entity or network element or device or node, etc.) has the ability to execute the training task of the first model.

[0253] Optionally, the first federated learning member can be replaced by a first federated learning client or a first client device, and the first client device can be a client NWDAF, that is, a Client NWDAF including MTLF, a NWDAF with federated learning client capabilities. For example, the first federated learning member supports at least one of the following: using available local data for federated learning training to generate local model information, specifically, receiving information about a basic model or an initial model sent by a federated learning server, and training the basic model or the initial model using available local data to generate local model information; reporting the trained local model information to the federated learning server; reporting the trained basic model information to the federated learning server; receiving global model information from the federated learning server, and repeating training iterations.

[0254] Exemplarily, the training tasks of the first model may include at least one of the following: network performance analysis, AI model training, network service quality analysis, network abnormal behavior detection, face recognition, or cross-institutional medical data analysis and disease prediction, etc.

[0255] Optionally, federated learning can be replaced by: federated machine learning, federated learning, or federated learning.

[0256] Optionally, the training task of the first model can be replaced by other names such as federated learning task, federated learning, federated learning function, federated learning process, or federated learning activity.

[0257] Optionally, the number of the first federated learning members may be one or more, which is not limited.

[0258] In a first implementation manner, the first request message includes a first training requirement corresponding to the first model.

[0259] The first training requirement refers to the current training requirement for the training task of the first model, or in other words, the first training requirement is determined after the first federated learning member executes the training task of the first model.

[0260] Optionally, before executing step S510, the method further includes the following step S501, which is not shown in the figure.

[0261] S501: The first federated learning member determines a first training requirement according to an intermediate training result of the first model and a second training requirement corresponding to the first model.

[0262] Among them, the intermediate training result is obtained by the first federated learning member executing the training task of the first model. The specific implementation method can refer to the relevant description of the following step S502. The second training requirement is obtained from the federated learning server. The specific implementation method can refer to the relevant description of step 1 of the following step S502.

[0263] It can be understood that the first model in the embodiment of the present application is requested by the model requester. The federated learning server can determine at least one first federated learning member to participate in the training of the first model for the training of the first model. Then for the at least one first federated learning member, there is at least one local model corresponding. For example, at least one first federated learning member includes FL member#1 and FL member#2, wherein FL member#1 obtains local model #1 through local training according to the second training requirement, and the local model #1 is the intermediate training result #1 of the first model. Further, FLmember#1 can determine the first training requirement #1 based on the intermediate training result #1 and the second training requirement; similarly, FLmember#2 obtains local model #2 through local training according to the second training requirement, and the local model #2 is the intermediate training result #2 of the first model. Further, FL member#2 can determine the first training requirement #2 based on the intermediate training result #2 and the second training requirement.

[0264] As an example, when it is determined that the first federated learning member cannot continue to perform the training task of the first model, and / or it is determined that the intermediate training result of the first model does not meet the second training requirement, the first federated learning member determines the first training requirement based on the intermediate training result and the second training requirement. For example, the first federated learning member can determine that it cannot continue to perform the training task based on the current local training situation and the second training requirement, as well as its own available resources or time information available for training.

[0265] It can be understood that the second training requirement refers to the overall training requirement for the training task of the first model, which can be understood as the training requirement determined before starting to execute the training task of the first model.

[0266] In a second implementation, the first request message includes an intermediate training result of the first model.

[0267] Optionally, before executing step S510 or S501, the method further includes the following step S502, which is not shown in the figure.

[0268] S502, the first federated learning member obtains the intermediate training result of the first model, which may include the following steps.

[0269] Step 1: The model requester sends a third request message to the federated learning server, and correspondingly, the federated learning server receives the third request message from the model requester. The third request message is used to request the execution of the training task of the first model, and the third request message includes the second training requirement corresponding to the first model.

[0270] Step 2: The federated learning server determines a first federated learning member from the candidate federated learning members according to the capability information of the candidate federated learning members and the second training requirement.

[0271] Step 3: The federated learning server sends a fourth request message to the first federated learning member, and correspondingly, the first federated learning member receives the fourth request message from the federated learning server. The fourth request message is used to request execution of a training task for the first model, and the fourth request message includes the second training requirement.

[0272] Step 4: The first federated learning member performs the training task of the first model according to the second training requirement to obtain an intermediate training result.

[0273] Optionally, the number of candidate federated learning members may be one or more, and is not limited to this. For example, the candidate federated learning members include a first federated learning member and a second federated learning member.

[0274] Optionally, the third request message or the fourth request message may be an AI / ML model training request message, and the specific message name is not limited.

[0275] Optionally, the model requester may be an application (APP), a client (eg, NWDAF), a server (eg, NWDAF), or a third party (eg, NF, AF, or AS), etc.

[0276] Based on the above step S502, the first federated learning member can send a first request message carrying the intermediate training result to the federated learning server when it is determined that the first federated learning member cannot continue to perform the training task of the first model and / or determines that the intermediate training result does not meet the second training requirement, that is, execute step S510.

[0277] Optionally, based on the above second implementation manner, the method further includes the following step S520.

[0278] S520: The federated learning server determines a first training requirement according to the intermediate training result.

[0279] As an example, the federated learning server determines the first training requirement based on the intermediate training result of the first model and the second training requirement. The specific implementation method can refer to the relevant description of the first federated learning member determining the first training requirement in the above step S501. For the sake of brevity, it will not be described here.

[0280] In summary, the first request message sent by the first federated learning member to the federated learning server may carry the first training requirement or the intermediate training result, and this application does not limit this. The difference between the two is that in the first implementation, the federated learning server can directly obtain the first training requirement, while in the second implementation, the federated learning server needs to determine the first training requirement based on the obtained intermediate training result and the second training requirement.

[0281] Optionally, the first request message in the above step S510 may also include indication information, the indication information is used to indicate that the first federated learning member cannot continue to perform the training task of the first model, and / or the indication information is used to indicate that at least one federated learning member is found. For example, the indication information may be indicated by 1 bit, bit "1" indicates that the first federated learning member cannot continue to perform the training task of the first model, bit "0" indicates that the first federated learning member supports continuing to perform the training task of the first model, and vice versa.

[0282] It can be understood that when the first request message does not carry any indication information, it can implicitly indicate that the first federated learning member cannot continue to perform the training task of the first model, that is, the federated learning server can determine that the first federated learning member cannot continue to perform the training task of the first model based on the intermediate training result or the first training requirement carried in the first request message; in other words, when the indication information carried in the first request message indicates that the first federated learning member cannot continue to perform the training task of the first model, it can implicitly indicate that the federated learning server has discovered at least one federated learning member; in other words, when the indication information carried in the first request message indicates that at least one federated learning member has been discovered, it can implicitly indicate that the first federated learning member cannot continue to perform the training task of the first model. This application does not limit this.

[0283] Optionally, the first request message in the above step S510 may also include at least one of the following: an identifier of the first model, an identifier of the training task of the first model, an analysis identifier corresponding to the first model, a service identifier (service ID) corresponding to the first model, or an application identifier (App ID) corresponding to the first model, used to indicate the training task of the first model.

[0284] Below, examples are given for the specific contents of the intermediate training result of the first model mentioned above, the first training requirement corresponding to the first model, and the second training requirement.

[0285] Exemplarily, the intermediate training results include at least one of the following: a temporary model obtained by executing the training task of the first model; accuracy information of the temporary model; accuracy information required for the training task of the first model; training time of the temporary model; training time required to continue to execute the training task of the first model; resource usage of the temporary model; training resources required to continue to execute the training task of the first model; the number of training rounds of the temporary model; or, the number of training rounds required to continue to execute the training task of the first model, and the specific explanations are as follows.

[0286] (1) an interim model obtained by executing the training task of the first model. For example, the interim model may be a part of the first model that meets the second training requirement;

[0287] (2) Precision information of the temporary model, which is used to indicate the precision or accuracy of the temporary model. It is an indicator that measures the ability of a machine learning model to make predictions or decisions based on data. It can be used to compare the performance of different models or evaluate the effectiveness of a specific model for a given task. For example, mean square error (MSE) and mean absolute error (MAE) are often used to measure the accuracy of regression models, which are used to predict continuous values. MSE refers to the average of the squared differences between the predicted value and the true value, and MAE refers to the average of the absolute differences between the predicted value and the true value. For example, the precision or accuracy of the temporary model is 80%.

[0288] (3) the accuracy information required by the training task of the first model, such as 100%;

[0289] (4) The training time of the temporary model, for example, 40 minutes, or 10:00:00-10:40:00;

[0290] (5) Continue to execute the training task of the first model for the required training time, for example, 20 minutes;

[0291] (6) Resource usage of the temporary model, for example, the size of computing resources or graphics processing unit (GPU) resources used is 25 MB;

[0292] (7) The training resources required to continue to execute the training task of the first model, for example, the size of the GPU resources is 25 MB;

[0293] (8) The number of training rounds for the temporary model, for example, 15 rounds;

[0294] (9) Continue to execute the training task of the first model for the required number of training rounds, for example, 5 rounds.

[0295] Exemplarily, the first training requirement includes at least one of the following: accuracy information required to continue to execute the training task of the first model; training time required to continue to execute the training task of the first model; training resources required to continue to execute the training task of the first model; or the number of training rounds required to continue to execute the training task of the first model, and the specific explanations are as follows.

[0296] (1) Continue to execute the accuracy information required by the training task of the first model, for example, 100%;

[0297] (2) Continue to perform the training task of the first model for the required training time, for example, 20 minutes;

[0298] (3) The training resources required to continue to execute the training task of the first model, for example, the size of the GPU resources is 25 MB;

[0299] (4) Continue to execute the training task of the first model for the required number of training rounds, for example, 5 rounds.

[0300] Exemplarily, the second training requirement includes at least one item: an analysis identifier corresponding to the first model; a service identifier corresponding to the first model; an application identifier corresponding to the first model; the time for acquiring the first model; the training duration of the first model; the accuracy information required for the training task of the first model; the training time required for the training task of the first model; the training resources required for the training task of the first model; or the number of training rounds required for the training task of the first model, and the specific explanations are as follows.

[0301] (1) An analysis identifier corresponding to the first model, used to indicate a training task for the first model.

[0302] (2) a service identifier corresponding to the first model, used to indicate a training task of the first model;

[0303] (3) an application identifier corresponding to the first model, used to indicate a training task of the first model;

[0304] Exemplarily, the analysis identifier, service identifier, or application identifier corresponding to the first model can be used to indicate a specific function or service, which is related to the first model, that is, the first model can be used to perform the specific function or service, or in other words, the first model supports the specific function or service corresponding to the analysis identifier. Among them, the specific function or service can be face recognition or network performance analysis. For example, the analysis identifier can indicate services such as terminal anomaly detection, terminal session analysis (such as quality of service QoS analysis), etc.

[0305] The service name may be Nnwdaf_MLModelTraining service, and the specific message name is not limited.

[0306] (4) The time when the first model is obtained, for example, the time when the first model is required to be obtained is before 10:00:00;

[0307] (5) The training time of the first model, for example, the training time of the first model is required to be 100 minutes;

[0308] (6) accuracy information required by the training task of the first model, such as 100%;

[0309] (7) The training time required for the training task of the first model, for example, 120 minutes, or 10:00:00-12:00:00;

[0310] (8) The training resources required for the training task of the first model, for example, the required GPU resource size is 50 MB;

[0311] (9) The number of training rounds required for the training task of the first model, for example, 20 rounds.

[0312] It is understandable that the first training requirement can be determined based on the second training requirement and the current training situation (e.g., the intermediate training result of the first model). For example, the training time required to continue to perform the training task of the first model can be determined based on the training time required for the training task of the first model and the training time of the temporary model (or the intermediate training result) obtained by performing the training task of the first model; for another example, the number of training rounds required to continue to perform the training task of the first model is determined based on the number of training rounds required for the training task of the first model and the number of training rounds of the temporary model (or the intermediate training result) obtained by performing the training task of the first model.

[0313] As an example, if the second training requirement initially obtained by the first federated learning member includes: the training time required for the training task of the first model is 60 minutes, and the required GPU resource size is 50 MB, then the intermediate training results obtained by the first federated learning member after executing the training task according to the second training requirement include: a temporary model that has been trained locally for 30 minutes, and the model training is 50% complete. If the first federated learning member cannot continue to execute the training task of the first model, the first training requirement that can be determined includes: based on the temporary model obtained by local training, the training time required to continue to execute the training task of the first model is 30 minutes, and the required GPU resource size is 25 MB.

[0314] It should be noted that if the first federated learning member determines that its unavailable time is less than or equal to the remaining time of the overall model training, that is, for the training task of the first model, the second training requirement includes: a total of 10 model training and iterations are required, a total of 600 minutes, and each local model training is approximately 60 minutes. If the first federated learning member has performed 210 minutes of model training locally (that is, the fourth local model training took 30 minutes), the first federated learning member is unavailable within 30 minutes), then the first federated learning member can instruct the second federated learning member to perform a round of local model training. At this time, the first training requirements for the second federated learning member include: the training time information is 30 minutes, the GPU resource size is 25MB, and the number of rounds of training tasks is 1.

[0315] S530: The federated learning server determines a second federated learning member from the candidate federated learning members according to the capability information of the candidate federated learning members and the first training requirement.

[0316] Among them, the capability information indicates whether the candidate federated learning members support continuing to execute the training task of the first model, at least one of the candidate federated learning members supports continuing to execute the training task of the first model, and at least one federated learning member includes a second federated learning member.

[0317] In the present application, the second federated learning member can be an AI / ML client, or an AI / ML enabler client (AI / ML enabler client or AI / ML E client), or an AI / ML enabler server (enabler server), or a client device (such as a client NWDAF). Among them, the second federated learning member supports the continuation of the training task of the first model, and supports the continuation of the training task of the first model, which can be understood as: supporting breakpoint continuation training, that is, supporting the continuation of training the first model after obtaining the unfinished intermediate training results.

[0318] Optionally, the number of the second federated learning members may be one or more, which is not limited.

[0319] Optionally, before executing step S530, the method further includes the following step S503, which is not shown in the figure.

[0320] S503: The federated learning server obtains capability information of candidate federated learning members.

[0321] As an example, the candidate federated learning member sends the capability information of the candidate federated learning member to the federated learning server, and the capability information indicates whether the candidate federated learning member supports breakpoint continuation training, that is, whether it supports continuing model training after obtaining the unfinished intermediate training results. For example, the candidate federated learning members include the first federated learning member and the second federated learning member, then the first federated learning member can send the capability information of the first federated learning member to the federated learning server, indicating that the first federated learning member does not support breakpoint continuation training; the second federated learning member can send the capability information of the second federated learning member to the federated learning server, indicating that the second federated learning member supports breakpoint continuation training.

[0322] Optionally, in an embodiment of the present application, the federated learning server may determine to execute the training task of the first model after receiving the third request message from the model requester, and then may query the candidate federated learning member whether it supports continued training of the endpoint for the first model, that is, query whether the candidate federated learning member can / can / support continuing to execute the training task of the first model after obtaining the intermediate training result of the first model. For example, the candidate federated learning members include the first federated learning member and the second federated learning member, then the first federated learning member may feedback to the federated learning server that the first federated learning member does not support continuing to execute the training task of the first model when the intermediate training result of the first model is obtained, and the second federated learning member may feedback to the federated learning server that the second federated learning member supports continuing to execute the training task of the first model when the intermediate training result of the first model is obtained.

[0323] Optionally, the capability information of the candidate federated learning member may be reported by the candidate federated learning member to the learning server when initiating the registration process, that is, the capability information of the candidate federated learning member may be carried in the registration request message, or the capability information of the candidate federated learning member may be reported independently, that is, the capability information of the candidate federated learning member may not be carried in the registration request message, and this application does not limit this.

[0324] In the present application, the capability information of a candidate federated learning member is also used to indicate at least one of the following: the training time supported by the candidate federated learning member; the available training time corresponding to the candidate federated learning member; the training resources supported by the candidate federated learning member; the available training resources corresponding to the candidate federated learning member; or, the number of model training rounds supported by the candidate federated learning member, the specific meanings are as follows.

[0325] (1) The training time supported by the candidate federated learning members, for example, the training time supported by the first federated learning member is 60 minutes, or 10:00:00-11:00:00, and the training time supported by the second federated learning member is 100 minutes, or 10:00:00-11:40:00.

[0326] (2) The available training time corresponding to the candidate federated learning members, for example, the available training time corresponding to the first federated learning member is 10:30:00-11:00:00, and the available training time corresponding to the second federated learning member is 10:20:00-11:20:00.

[0327] (3) The training resources supported by the candidate federated learning members. For example, the GPU size of the training resources supported by the first federated learning member is 25 MB, and the GPU size of the training resources supported by the second federated learning member is 30 MB.

[0328] (4) Available training resources corresponding to the candidate federated learning members. For example, the GPU size of the available training resources corresponding to the first federated learning member is 10 MB, and the GPU size of the available training resources corresponding to the second federated learning member is 20 MB.

[0329] (5) The number of model training rounds supported by the candidate federated learning member. For example, the number of model training rounds supported by the first federated learning member is 20 rounds, and the number of model training rounds supported by the second federated learning member is 30 rounds.

[0330] The above is an example explanation given for the convenience of description and understanding, taking the candidate federated learning members including the first federated learning member and the second federated learning member as an example, and other schemes are not excluded. Optionally, the candidate federated learning members may also include a third federated learning member.

[0331] Based on the second federated learning member determined in step S530, the first federated learning member or the federated learning server may request the second federated learning member to continue to perform the training task of the first model to ensure that the training of the first model is completed. The following is an example of the first federated learning member and the federated learning server requesting the second federated learning member to continue to perform the training task of the first model in combination with method 1 and method 2.

[0332] Method 1: The first federated learning member requests the second federated learning member to continue to perform the training task of the first model.

[0333] S540, the federated learning server sends a first response message to the first federated learning member;

[0334] Correspondingly, the first federated learning member receives a first response message from the federated learning server.

[0335] The first response message includes information of the second federated learning member.

[0336] Optionally, the first response message may be an AI / ML member discovery response message, and the specific message name is not limited.

[0337] Exemplarily, the information of the second federated learning member includes at least one of the following: an identification (ID) of the second federated learning member, an NF instance ID of the second federated learning member (e.g., NF Instance Id(s) of the NFproducer consumer), an NF instance ID of a network element having a federated learning client function or supporting a federated learning client function (e.g., NF Instance Id(s) of the NWDAF), address information of the second federated learning member, or other information that can identify the second federated learning member. For example, the information of the second federated learning member may be at least one of the following: FL member ID, AI / ML client ID, AI / ML enabler server ID or AI / ML E server ID, AI / ML enabler client ID or AI / ML E client ID, Instance Id(s) of the AI / ML enabler server, or Instance Id(s) of the AI / ML enabler client, etc.

[0338] Optionally, the first response message may also include at least one of the following: an analysis identifier corresponding to the first model, a service identifier corresponding to the first model, or an application identifier corresponding to the first model, which is used to indicate a training task for the first model.

[0339] Optionally, the first response message may further include instruction information, where the instruction information is used to instruct the second federated learning member to continue to perform the training task.

[0340] S550, the first federated learning member sends a second request message to the second federated learning member;

[0341] Correspondingly, the second federated learning member receives the second request message from the first federated learning member.

[0342] The second request message is used to request the second federated learning member to continue to execute the training task of the first model.

[0343] Optionally, the second request message may be an FL mode transfer request message, and the specific message name is not limited.

[0344] In one implementation, the second request message includes the intermediate training result of the first model and the first training requirement. For example, the intermediate training result indicates: the training time of the local model (or temporary model) obtained by the first federated learning member for model training is 40 minutes, such as 10:00:00-10:40:00; the size of the training resources used to train the local model is 25MB; the number of training rounds for local training is 15 rounds. The first training requirement indicates: the training time for continuing to execute the first model is 30 minutes, such as 11:00:00-11:30:00, the size of the training resources required to continue to execute the training task of the first model is 15MB, and the number of training rounds required to continue to execute the training task of the first model is 10 rounds. This implementation method can implicitly indicate the second training requirement, that is, the model requester requests the relevant requirements for the total training time, training resources and number of training rounds required to execute the training task of the first model.

[0345] In another implementation, the second request message includes the intermediate training result of the first model and the second training requirement. For example, the intermediate training result indicates: the training time of the local model (or temporary model) obtained by the first federated learning member for model training is 40 minutes, such as 10:00:00-10:40:00; the size of the training resources used to train the local model is 25MB; the number of training rounds for local training is 15 rounds. The second training requirement indicates: the total training time required to execute the first model is 70 minutes, for example, it can be completed within 10:00:00-12:00:00, the size of the training resources required to execute the training task of the first model is 40MB in total, and the number of training rounds required to execute the training task of the first model is 25 rounds in total. This implementation method can implicitly indicate the first training requirement, that is, the relevant requirements for the training time, training resources and number of training rounds required for the second federated learning member to continue to execute the training task of the first model.

[0346] In another implementation, the second request message includes the intermediate training result of the first model, the first training requirement, and the second training requirement. For example, the intermediate training result indicates: the training time of the local model (or temporary model) obtained by the first federated learning member for model training is 40 minutes, such as 10:00:00-10:40:00; the size of the training resources used to train the local model is 25MB; the number of training rounds for local training is 15 rounds. The first training requirement indicates: the training time for continuing to execute the first model is 30 minutes, such as 11:00:00-11:30:00, the size of the training resources required to continue to execute the training task of the first model is 15MB, and the number of training rounds required to continue to execute the training task of the first model is 10 rounds. The second training requirement indicates: the total training time required to execute the first model is 70 minutes, such as it can be completed within 10:00:00-12:00:00, the size of the training resources required to execute the training task of the first model is 40MB in total, and the number of training rounds required to execute the training task of the first model is 25 rounds in total.

[0347] The above implementations are merely examples given for ease of understanding and do not exclude other solutions.

[0348] Optionally, the second request message may also carry information of the federated learning server, such as an identifier and / or address information of the federated learning server, so that the second federated learning member can subsequently feed back the training results obtained from local training to the federated learning server.

[0349] Method 2: The federated learning server requests the second federated learning member to continue to execute the training task of the first model.

[0350] S560, the federated learning server sends a second request message to the second federated learning member;

[0351] Correspondingly, the second federated learning member receives a second request message from the federated learning server.

[0352] The second request message is used to request the second federated learning member to continue to perform the training task of the first model, and the second request message includes the intermediate training result and the first training requirement.

[0353] Optionally, the second request message may be an ML mode training request message, and the specific message name is not limited.

[0354] It can be understood that the above method 1 can be performed on the basis of the first implementation method of step S510, that is, when the first federated learning server carries the first training requirement in the first request message, or does not carry the intermediate training result of the first model, the federated learning server can perform the discovery process of the second federated learning server (the specific discovery process can be found in Figure 3 ), and feeds back the information of the second federated learning member to the first federated learning member through method one, and the first federated learning member requests the second federated learning member to continue to perform the training task of the first model. Similarly, the above method two can be performed on the basis of the second implementation method of step S510, that is, when the first federated learning server can carry the intermediate training result of the first model in the first request message, the federated learning server can request the second federated learning member to continue to perform the training task of the first model through method two after determining the first training requirement and executing the discovery process of the second federated learning server.

[0355] Based on the above method 1 or method 2, the second federated learning member can continue to perform the training task of the first model according to the intermediate training results and the first training requirements, and feedback the training results to the federated learning server. For details, please refer to the following steps S570-S580.

[0356] S570, the second federated learning member continues to perform the training task according to the intermediate training result and the first training requirement, and obtains the first model and the training information of the first model.

[0357] Exemplarily, the training information of the first model includes at least one of the following: an identifier of the first model (for example, the ID of the first model); address information of the first model (address of model); analysis identifier corresponding to the first model; service identifier corresponding to the first model; application identifier corresponding to the first model; time of acquiring the first model; training duration of the first model; or accuracy information of the first model. For the incomplete parts, please refer to the relevant description above.

[0358] Optionally, before executing step S570, the second federated learning member may determine whether to continue to execute the training task, that is, the method further includes the following step S504, which is not shown in the figure.

[0359] S504: The second federated learning member determines whether to continue to perform the training task based on the intermediate training result and the first training requirement.

[0360] As an example, the second federated learning member determines to continue the training task based on its own capability information, intermediate training results and the first training requirements. For example, the second federated learning member supports continuing the training task of the first model, the available training time is 120 minutes, the available GPU resource size is 50MB, the intermediate training results include: the first federated learning member has trained 50% of the first model, the GPU resource size used is 20MB, and the training time is 60 minutes. The second training requirement indicates that the overall training requirements of the first model include: the training time of the first model is required to be 100 minutes, and the required GPU resource size is 60MB. It can be seen that the second federated learning member has the ability to continue the training task of the first model.

[0361] As another example, the second federated learning member may receive instruction information from the first federated learning member, and determine to continue to perform the training task according to the instruction information. The instruction information is used to indicate that the second federated learning member can continue to perform the training task. The instruction information may be explicitly carried in the first response message in the above step S540, or the instruction information may be determined by the first federated learning member according to the information of the second federated learning member carried in the first response message, without limitation.

[0362] It can be understood that only when it is determined that the second federated learning member continues to perform the training task, steps S570-S580 are executed, and then the training of the first model can be completed.

[0363] It can be understood that since the federated learning server selects or determines the second federated learning member from the candidate federated learning members according to the intermediate training results and the second training requirements in step S530 to continue the training task of the first model, it can be said that the second federated learning member supports continuing the training task of the first model, or in other words, the second federated learning member has the ability to continue training the first model after obtaining the unfinished intermediate training results. Therefore, the above-mentioned step S504 may not be executed. This is because, usually, after executing step S504, the second federated learning member can determine to participate in continuing the training task of the first model.

[0364] Optionally, if the second federated learning member is participating in or is about to participate in other federated learning (i.e., different from the training task of the first model) before receiving the second request message, which causes the second federated learning member to be unable to participate in or complete the training task of the first model, then the above step S504 can be executed, and the second federated learning member can determine not to participate in continuing to execute the training task of the first model. In this case, the second federated learning member does not execute steps S570-S580, that is, the second federated learning member can reject the message to the federated learning server, and optionally, can carry a reason value, which can indicate that the second federated learning member is participating in or is about to participate in other federated learning, etc.

[0365] S580, the second federated learning member sends a second response message to the federated learning server;

[0366] Correspondingly, the federated learning server receives a second response message from the second federated learning member.

[0367] Among them, the second response message includes the first model and the training information of the first model, that is, the second federated learning member can feedback the local training results to the federated learning server, that is, including the first model and the training information of the first model.

[0368] Optionally, for the above-mentioned method one, if the second request message does not carry the information of the federated learning server, the second federated learning member may send the first model and the training information of the first model to the first federated learning member, and then the first federated learning member sends the first model and the training information of the first model to the federated learning server; if the second request message carries the information of the federated learning server, the second federated learning member may send the first model and the training information of the first model to the federated learning server based on the information of the federated learning server, without forwarding through the first federated learning member, thereby saving signaling overhead.

[0369] Optionally, for the above-mentioned method 2, the second federated learning member may send the first model and the training information of the first model to the federated learning server after completing local training.

[0370] Optionally, the first model and training information of the first model may be carried in a FL model report (eg, FL localmode report).

[0371] Optionally, the second federated learning member may send a second response message to the first federated learning member or the federated learning server to indicate that the second federated learning member confirms to continue the training task of the first model, or to indicate that the second federated learning member has completed the training of the first model.

[0372] Optionally, the second response message may be an FL mode transfer response message or an ML mode training response message, and the specific message name is not limited.

[0373] That is, the second federated learning member can send the trained local model (i.e., the first model) to the federated learning server. After obtaining the local models of all the first federated learning members and the second federated learning members, the federated learning server can perform model aggregation and calculate the global model data. Optionally, the federated learning server can further instruct the federated learning member to perform a new round of local model training, which is not limited.

[0374] Based on the above scheme, the federated learning server can select at least one second federated learning member from the candidate federated learning members based on the first training requirement or the intermediate training result of the first model carried in the received first request message to continue the training task of the first model and ensure the completion of the training of the first model. That is, when the federated learning server selects a suitable second federated learning member to continue the training task, it can take into account the current training situation or training requirements, thereby avoiding an increase in the model convergence time and a decrease in the model performance.

[0375] For ease of understanding, the following Figure 6 and Figure 7 , the specific process applicable to the embodiment of the present application is introduced for different scenarios. In the example below, the federated learning server is an AI / ML server (or replaced by an AI / ML enablerserver or an AI / ML E server), the first federated learning member is an AI / ML client#1 (or replaced by an AI / MLenabler client#1 or an AI / ML E client#1), the second federated learning member is an AI / ML client#2 (or replaced by an AI / ML enabler client#2 or an AI / ML E client#2), and the model requester is an App. Optionally, other federated learning members may also be included, such as an AI / ML client#3 (or replaced by an AI / ML enablerclient#3 or an AI / ML E client#3). The present application does not limit the number of federated learning members that perform the training task of the first model. It can be understood that the process described below is only an example description, and the embodiment of the present application is not limited thereto. The content not described in detail below can refer to the description in method 500, which will not be repeated below.

[0376] It should be noted that the AI / ML E client in the embodiment of the present application can be replaced by a UE, a wireless access controller (AC), or a VAL server, that is, the UE, AC, or VAL server can serve as a federated learning member to perform federated learning tasks.

[0377] It should be pointed out that the names of the messages transmitted between the execution entities in the embodiments of the present application are only examples given for ease of understanding, and the present application does not make any specific limitations on the message names.

[0378] Figure 6 is a flow chart of the communication method provided in the embodiment of the present application. Figure 6 As shown, the implementation method is that the first federated learning member requests the second federated learning member to continue to perform the training task of the first model, which specifically includes the following steps. For the parts not described in detail, please refer to the relevant description of the above existing protocol.

[0379] S601, the App client (for example, the App client) sends a registration request message to AI / ML E client#1 (ie, the first federated learning member), and correspondingly, AI / ML E client#1 receives the registration request message from the App client.

[0380] The registration request message carries at least one of a service ID, an analytics ID, or an APP ID, which is used to indicate the first model or a training task of the first model.

[0381] S602, AI / ML E client#1 sends a registration response message to the App client, and correspondingly, the App client receives a registration response message from AI / ML E client#1.

[0382] The registration response message indicates whether the registration is successful or failed.

[0383] S603, AI / MLE client#1 sends a registration request message to the AI / MLE server (i.e., the federated learning server), and correspondingly, the AI / MLE server receives the registration request message from the AI / MLE client#1.

[0384] Among them, the registration request message carries the service ID, analytics ID, APP ID, or capability of AI / ML Eclient#1 (i.e., capability information of the first federated learning member, including compute capability or storage capability). For the specific interpretation of the capability information of the first federated learning member, please refer to the relevant description of steps S530 and S503 of the above method 500.

[0385] Further, the AI / MLE server sends a registration response message to the AI / MLE client#1, and correspondingly, the AI / MLE client#1 receives a registration response message from the AI / MLE server, where the registration response message is used to indicate whether the registration is successful or failed.

[0386] Similarly, AI / ML E client#2 sends a registration request message to the AI / ML E server to request registration with the AI / ML E server. The registration request message may carry the capability of AI / ML E client#2 (i.e., the capability information of the second federated learning member). For a specific interpretation of the capability information of the second federated learning member, reference may be made to the description of steps S530 and S503 of the above method 500.

[0387] S604, the App (i.e., the model requester) sends an AI / ML model training request message (i.e., the third request message) to the AI / ML E server. Correspondingly, the AI / ML E server receives the AI / ML model training request message from the App.

[0388] Among them, the AI / ML model training request message includes a second training requirement (ML model training request), and optionally, may also carry at least one of a service ID, an analytics ID, or an APP ID. For a specific interpretation of the second training requirement, please refer to the relevant description of step S510 of the above method 500.

[0389] Optionally, the above step S604 can be replaced by: APP can also send a request message to AI / ML E client#1 and / or AI / ML E client#2 to request to execute the training task of the first model. Further, AI / ML E client#1 and / or AI / ML E client#2 send an AI / ML model training request message (i.e., a third request message) to AI / ML E server. Correspondingly, AI / ML E server receives the AI / ML model training request message from AI / ML E client#1 and / or AI / ML E client#2. In other words, AI / ML E server can directly or indirectly receive the AI / ML model training request message (i.e., the third request message) from App, which is not limited in this application.

[0390] S605, AI / MLE server determines AI / MLE client #1.

[0391] Exemplarily, the AI / ML E server determines or selects AI / ML Eclient#1 to perform the training task of the first model according to the second training requirement and the capability information of AI / ML E client#1. Specifically, the AI / ML E server and the NRF may exchange information and execute the process of discovering AI / ML Eclient#1. The specific discovery process may refer to the above Figure 3 The relevant description is not given here.

[0392] S606, the AI / ML E server sends an AI / ML model training request message (i.e., a fourth request message) to the AI / ML E client #1. Correspondingly, the AI / ML E client #1 receives the AI / ML model training request message from the AI / ML E server.

[0393] The AI / ML model training request message is used to request AI / ML E client#1 to perform the training task of the first model. The AI / ML model training request message includes the second training requirement, and optionally, may also carry at least one of a service ID, an analytics ID, or an APP ID.

[0394] S607, when AI / ML E client #1 determines that it cannot continue to perform the training task and / or determines that the intermediate training result cannot meet the second training requirement, AI / ML E client #1 determines the first training requirement. Specifically, AI / ML E client #1 determines the first training requirement according to the second training requirement and the intermediate training result.

[0395] Exemplarily, AI / ML E client#1 performs local model training according to the second training requirement to obtain an intermediate training result. For specific interpretations of the first training requirement and the intermediate training result, reference may be made to the relevant description of step S510 of the above method 510.

[0396] It can be understood that during the training process, after the AI / ML E client has trained the local model, it can interact with the AI / ML E server, that is, AI / ML E client#1 can send the trained local model to the AI / ML E server, and the AI / ML E server can calculate the global model data after obtaining the local models of all AI / ML E clients. Optionally, the AI / ML E server can further instruct the AI / ML E client to perform a new round of local model training.

[0397] Exemplarily, when the AI / ML E client detects that the current resources cannot continue model training, and / or the currently trained model cannot meet the required accuracy, the AI / ML E client #1 can determine the first training requirement based on the current FL model training situation, that is, determine the capabilities of the target FL member, that is, the federated learning capability that the AI / ML E client #2 needs to have. For the specific implementation method, please refer to the relevant description of step S501 of the above method 500.

[0398] S608, AI / ML E client#1 sends a discovery request message (ie, a first request message, such as AI / ML E member discovery request) to the AI / ML E server. Correspondingly, the AI / ML E server receives the discovery request message from AI / ML E client#1.

[0399] The discovery request message is used to request the discovery of at least one second federated learning member for continuing the training task of the first model. The discovery request message carries the first training requirement. Optionally, the discovery request message may also include at least one of a service ID, an analytics ID, or an APP ID.

[0400] S609, the AI / ML E server determines AI / ML E client #2 based on the first training requirement and the capability information of the candidate AI / ML E clients (for example, including AI / ML E client #1 and AI / ML E client #2), that is, selects a suitable AI / ML E client #2 as the federated learning member that continues to execute the training task of the first model. For the specific implementation method, please refer to the relevant description of step S530 of the above method 500.

[0401] S610, the AI / ML E server sends a discovery response message (AI / ML member discovery response) to the AI / ML E client #1, and correspondingly, the AI / ML E client #1 receives the discovery response message from the AI / ML E server.

[0402] The discovery response message includes information of AI / ML E client#2, such as ID or address information of AI / ML E client#2. Optionally, the discovery response message may also include at least one of a service ID, an analytics ID, or an APP ID.

[0403] S611, AI / ML E client#1 sends a FL model transfer request message (i.e., a second request message, such as FL model transfer request) to AI / ML E client#2. Correspondingly, AI / ML E client#2 receives the FL model transfer request message from AI / ML E client#1.

[0404] The FL model transmission request message is used to request AI / ML E client#2 to continue to execute the training task of the first model.

[0405] Exemplarily, the FL model transmission request message carries the intermediate training result and the first training requirement. Optionally, the FL model transmission request message may also include at least one of a service ID, an analytics ID, or an APP ID. Optionally, the FL model transmission request message may also include information of an AI / ML E server, such as an ID and / or address of the AI / ML E server.

[0406] Optionally, the FL model transmission request message may also carry indication information for indicating whether AI / ML E client#2 can continue to perform model training. For the manner of determining the indication information, reference may be made to the relevant description of step S504 of the above method 500.

[0407] S612, AI / ML E client#2 continues to execute the training task according to the intermediate training result and the first training requirement, and obtains the first model and the training information of the first model.

[0408] For a specific explanation of the training information of the first model, please refer to the relevant description of step S570 of the above method 500.

[0409] Optionally, before executing step S612, AI / ML E client#2 may determine whether to continue training the model according to the intermediate training result and the first training requirement. For the specific implementation method, please refer to the relevant description of step S504 of the above method 500.

[0410] S613, AI / ML E client#2 sends a FL model transfer response message (FL model transfer response) to AI / ML E client#1, and correspondingly, AI / ML E client#1 receives the FL model transfer response message from AI / ML E client#2.

[0411] The FL model transmission response message is used to indicate whether AI / ML E client#2 can continue to perform model training (success) or not (failure). If model training can continue, AI / ML E client#2 executes the above step S612.

[0412] Optionally, the present application does not limit the order in which steps S612 and S613 are executed.

[0413] S614, AI / ML E client#2 sends a FL model report (e.g., FL local mode report) to the AI / ML E server, and correspondingly, the AI / ML E server receives the FL model report from the AI / ML E client#2, wherein the FL model report includes the first model and the training information of the first model.

[0414] S615, AI / ML E server performs model aggregation.

[0415] For the specific implementation of the above steps S614-S615, please refer to the relevant description of step S580 of the above method 500.

[0416] Based on the above solution, AI / ML E client#1 can determine the capabilities of the target AI / ML E client according to the local model training situation and the model training requirements required by the AI / ML E server, and feed back to the AI / ML E server, so that the AI / ML E server can select a suitable AI / ML E client#2 to continue to perform the training task. Furthermore, AI / ML E client#1 can send the intermediate training results and the first training requirement to AI / ML E client#2, requesting AI / ML E client#2 to continue to perform the model training without skipping or discarding the training when AI / ML E client#1 cannot continue to perform the training task, thereby avoiding increasing the model convergence time and avoiding the problem of model performance degradation.

[0417] Figure 7 Schematic diagram of the communication method provided in the embodiment of the present application. Figure 6 The difference is that in this implementation, the federated learning server requests the second federated learning member to continue to perform the training task of the first model, such as Figure 7 As shown, it specifically includes the following multiple steps, and the parts not fully described can refer to the relevant description of the above-mentioned existing protocol.

[0418] S701, the App client (for example, the App client) sends a registration request message to AI / ML E client#1 (ie, the first federated learning member), and correspondingly, AI / ML E client#1 receives the registration request message from the App client.

[0419] S702, AI / ML E client#1 sends a registration response message to the App client, and correspondingly, the App client receives a registration response message from AI / ML E client#1.

[0420] S703, AI / MLE client#1 sends a registration request message to the AI / MLE server (i.e., the federated learning server), and correspondingly, the AI / MLE server receives the registration request message from the AI / MLE client#1.

[0421] Similarly, AI / MLE client#2 sends a registration request message to the AI / MLE server to request to register with the AI / MLE server.

[0422] S704, the App (i.e., the model requester) sends an AI / ML model training request message (i.e., the third request message) to the AI / ML E server. Correspondingly, the AI / ML E server receives the AI / ML model training request message from the App.

[0423] The AI / ML model training request message includes a second training request (ML model training request).

[0424] S705, AI / MLE server determines AI / MLE client #1.

[0425] Exemplarily, the AI / ML E server determines or selects AI / ML Eclient#1 to perform the training task of the first model based on the second training requirement and the capability information of AI / ML E client#1.

[0426] S706, the AI / ML E server sends an AI / ML model training request message (i.e., a fourth request message) to the AI / ML E client#1. Correspondingly, the AI / ML E client#1 receives the AI / ML model training request message from the AI / ML E server.

[0427] The AI / ML model training request message is used to request the AI / ML E client#1 to perform the training task of the first model. The AI / ML model training request message includes a second training requirement.

[0428] S707, when AI / ML E client #1 determines that it cannot continue to perform the training task and / or determines that the intermediate training result cannot meet the second training requirement, AI / ML E client #1 determines the first training requirement. Specifically, AI / ML E client #1 determines the first training requirement according to the second training requirement and the intermediate training result.

[0429] Exemplarily, AI / ML E client #1 performs local model training according to the second training requirement to obtain an intermediate training result.

[0430] S708, AI / ML E client#1 sends a discovery request message (ie, a first request message, such as AI / ML member discovery request) to the AI / ML E server. Correspondingly, the AI / ML E server receives the discovery request message from AI / ML E client#1.

[0431] The discovery request message is used to request the discovery of at least one second federated learning member for continuing the training task of the first model. The discovery request message carries the intermediate training result.

[0432] The specific implementation of the above steps S701-S708 can refer to the relevant description of steps S601-S608 of the above method 600, and for the sake of brevity, it will not be repeated here. The difference between the two is that: Figure 6 The discovery request message of step S608 carries the first training requirement, Figure 7 The discovery request message of step S708 carries the intermediate training result.

[0433] S709: The AI / MLE server determines AI / MLE client#2 according to the first training requirement and the capability information of the candidate AI / MLE clients (for example, including AI / MLE client#1 and AI / MLE client#2).

[0434] Specifically, the AI / ML E server determines the first training requirement based on the intermediate training result obtained in step S708 and the second training requirement obtained in step S704, and then determines AI / ML E client#2 based on the first training requirement and the capability information of the candidate AI / ML E client, that is, selects a suitable AI / ML E client#2 as the federated learning member that continues to execute the training task of the first model. For the specific implementation method, please refer to the relevant description of steps S520-S530 of the above method 500.

[0435] S710, AI / ML E server sends a model training request message (i.e., a second request message, such as ML model training request) to AI / ML E client#2. Correspondingly, AI / ML E client#2 receives the model training request message from AI / ML E client#1.

[0436] The model training request message is used to request AI / ML E client#2 to continue to execute the training task of the first model.

[0437] Exemplarily, the model training request message carries the intermediate training result and the first training requirement. Optionally, the model training request message may also include at least one of a service ID, an analytics ID, or an APP ID.

[0438] Optionally, the model training request message may also carry indication information for indicating whether AI / ML E client#2 can continue to perform model training. For the manner of determining the indication information, reference may be made to the relevant description of step S504 of the above method 500.

[0439] S711, AI / ML E client#2 continues to execute the training task according to the intermediate training result and the first training requirement, and obtains the first model and the training information of the first model.

[0440] S712, AI / ML E client#2 sends a model training response message (i.e., a second response message, such as ML model training response) to AI / ML E client#1, and correspondingly, AI / ML E client#1 receives a model training response message from AI / ML E client#2.

[0441] S713, AI / ML E client#2 sends a FL model report (e.g., FL local mode report) to the AI / ML E server, and correspondingly, the AI / ML E server receives the FL model report from the AI / ML E client#2, wherein the FL model report includes the first model and the training information of the first model.

[0442] S714, the AI / ML E server performs model aggregation.

[0443] For the specific implementation of the above steps S711-S714, please refer to the relevant description of steps S612-S615 of the above method 600, which will not be repeated here for the sake of brevity.

[0444] Based on the above solution, the AI / ML E server can determine the capabilities of the target AI / ML E client according to the local model training situation of AI / ML E client #1 and the model training requirement parameters required by the AI / ML E server, and then select AI / ML E client #2 to continue to perform the training task. Furthermore, the AI / ML E server can send the intermediate training results and the first training requirement to AI / ML E client #2, requesting AI / ML E client #2 to continue to perform the model training without skipping or discarding the training when AI / ML E client #1 cannot continue to perform the training task, thereby avoiding increasing the model convergence time and avoiding the problem of model performance degradation.

[0445] The above combination Figures 5 to 7The technical solution of the present application is described, that is, when the first federated learning member determines that the first federated learning member cannot continue to perform the training task of the first model, and / or the intermediate training result of the first model does not meet the second training requirement corresponding to the first model, the first federated learning member interacts with the federated learning server to find a suitable second federated learning member for continuing to perform the training task of the first model, and then the first federated learning member or the federated learning server subsequently requests the second federated learning member to continue to perform the training task to ensure the completion of the training of the first model and guarantee the model performance.

[0446] Combine the following Figure 8 and Fig. 9 The technical solution of this application is described. Figures 5 to 7 The difference is that when the first federated learning member determines that the first federated learning member cannot continue to perform the training task of the first model, and / or the intermediate training result of the first model does not meet the second training requirement corresponding to the first model, it is not necessary to interact with the federated learning server, which can improve the model training efficiency. This is because when the federated learning server requests the first federated learning member to perform the training task of the first model, the information of the standby federated learning member (i.e., the second federated learning member) can be fed back to the first federated learning member.

[0447] Figure 8 is a flow chart of a communication method provided in an embodiment of the present application. Figure 8 As shown, the method includes the following steps, and the federated learning server, the first federated learning member, the second federated learning member, and the model requester are used as the execution subjects for example description. For the parts not described in detail, please refer to the above Figure 5 Or a relevant description of an existing agreement.

[0448] S810, the model requesting party sends a third request message to the federated learning server;

[0449] Correspondingly, the federated learning server receives a third request message from the model requester.

[0450] Among them, the third request message is used to request the execution of the training task of the first model. The third request message includes the second training requirement corresponding to the first model. The specific interpretation of the second training requirement can refer to the relevant description of step S510 of the above method 500.

[0451] S820: The federated learning server determines a first federated learning member and a second federated learning member from the candidate federated learning members according to the capability information and the second training requirement of the candidate federated learning members.

[0452] It can be understood that the second federated learning member is an alternative (or standby) federated learning member for executing the training task of the first model. Specifically, the second federated learning member can be regarded as an alternative federated learning member of the first federated learning member. Among them, the alternative (or standby) federated learning member can be understood as: the second federated learning member can continue to execute the training task of the first model when the first federated learning member cannot continue to execute the training task of the first model to ensure that the first model can be trained to completion, that is, the second federated learning member supports the continued execution of the training task of the first model, or the second federated learning member has the ability to continue to execute the training task of the first model.

[0453] For a specific interpretation of the capability information of the candidate federated learning member, please refer to the relevant description of the above method 500.

[0454] S830, the federated learning server sends a fourth request message to the first federated learning member;

[0455] Correspondingly, the first federated learning member receives a fourth request message from the federated learning server.

[0456] The fourth request message is used to request execution of the training task of the first model, and the fourth request message includes the second training requirement and information of the second federated learning member.

[0457] S840, the first federated learning member performs the training task of the first model according to the second training requirement, and obtains an intermediate training result of the first model.

[0458] For a specific interpretation of the intermediate training results, please refer to the relevant description of the above method 500.

[0459] The specific implementation of the above steps S810-S830 can refer to the relevant description of steps 1 to 4 in step S502 of the above method 500, which will not be repeated here for the sake of brevity. The difference is that in step S820, the federated learning server also determines a spare second federated learning member, and in step S830, the information of the second federated learning member is fed back to the first federated learning member.

[0460] S850, when it is determined that the first federated learning member cannot continue to perform the training task of the first model, and / or when it is determined that the intermediate training result of the first model does not meet the second training requirement, the first federated learning member determines the first training requirement based on the intermediate training result and the second training requirement corresponding to the first model.

[0461] For a specific explanation of the first training requirement, please refer to the relevant description of the above method 500.

[0462] Regarding the specific implementation manner in which the first federated learning member determines the first training requirement according to the intermediate training result and the second training requirement corresponding to the first model, reference may be made to the relevant description of step S501 of the above method 500.

[0463] Since the first federated learning member has obtained the information of the alternative (or standby) federated learning member in step S830, when it is determined that the first federated learning member cannot continue to perform the training task of the first model, and / or the intermediate training result of the first model does not meet the second training requirements, the first federated learning member can directly request the second federated learning member to continue to perform the training task of the first model, that is, execute the following step S860, without interacting with the federated learning server to discover alternative federated learning members, which can improve the efficiency of model training.

[0464] S860, the first federated learning member sends a second request message to the second federated learning member;

[0465] Correspondingly, the second federated learning member receives the second request message from the first federated learning member.

[0466] The second request message is used to request the second federated learning member to continue to perform the training task of the first model, and the second request message includes the intermediate training result of the first model and the first training requirement.

[0467] S870, the second federated learning member continues to perform the training task according to the intermediate training result and the first training requirement, and obtains the first model and the training information of the first model.

[0468] For a specific explanation of the training information of the first model, please refer to the relevant description of step S570 of the above method 500.

[0469] Optionally, before executing step S870, the second federated learning member may determine whether to continue to execute the training task. For specific implementation methods, reference may be made to the relevant description of step S504 of the above method 500.

[0470] S880, the second federated learning member sends the first model and training information of the first model to the federated learning server;

[0471] Correspondingly, the federated learning server receives the first model from the second federated learning member and the training information of the first model.

[0472] For the specific implementation of the above steps S860-S880, please refer to the relevant description of steps S550, S570 and S580 of the above method 500, which will not be repeated here for the sake of brevity.

[0473] Based on the above scheme, the federated learning server can select the first federated learning member and the backup learning member of the first federated learning member, that is, the second federated learning member, from the candidate federated learning members based on the second training requirement carried in the received third request message and the capability information of the candidate federated learning members, to continue to perform the training task of the first model and ensure the completion of the training of the first model. That is, when the first federated learning member determines that the first federated learning member cannot continue to perform the training task of the first model, and / or the intermediate training result of the first model does not meet the second training requirement corresponding to the first model, it can directly request the second federated learning member to continue to perform the training task of the first model without information interaction with the federated learning server, which can improve the efficiency of model training, avoid the increase of model convergence time, and avoid the degradation of model performance.

[0474] For ease of understanding, the following Fig. 9 , taking the federated learning server as AI / ML E server (or replaced by AI / ML enabler server or AI / ML E server), the first federated learning member as AI / ML E client#1 (or replaced by AI / ML enabler client#1 or AI / ML E client#1), the second federated learning member as AI / ML Eclient#2 (or replaced by AI / ML enabler client#2 or AI / ML Eclient#2), and the model requester as App as an example for explanation. Optionally, other federated learning members may also be included, such as AI / ML Eclient#3 (or replaced by AI / ML enabler client#3 or AI / ML E client#3). The present application does not limit the number of federated learning members that perform the training task of the first model. It will be understood that the process described below is only an example description, and the embodiments of the present application are not limited to this. For the content not described in detail below, please refer to the description in method 800, which will not be repeated below.

[0475] It should be pointed out that the AI / ML E client in the embodiment of the present application can be replaced by UE, AC, or VAL server, that is, UE, AC, or VAL server can serve as a federated learning member to perform federated learning tasks.

[0476] It should be pointed out that the names of the messages transmitted between the execution entities in the embodiments of the present application are only examples given for ease of understanding, and the present application does not make any specific limitations on the message names.

[0477] Fig. 9 is a flow chart of the communication method provided in the embodiment of the present application, such as Fig. 9 As shown, it includes the following steps. For the parts not described in detail, please refer to the above Figure 8 Or a relevant description of an existing agreement.

[0478] S901, the App client (for example, the App client) sends a registration request message to AI / ML E client#1 (ie, the first federated learning member), and correspondingly, AI / ML E client#1 receives the registration request message from the App client.

[0479] S902, AI / ML E client#1 sends a registration response message to the App client, and correspondingly, the App client receives a registration response message from AI / ML E client#1.

[0480] S903, AI / MLE client#1 sends a registration request message to the AI / MLE server (i.e., the federated learning server), and correspondingly, the AI / MLE server receives the registration request message from the AI / MLE client#1.

[0481] Similarly, AI / MLE client#2 sends a registration request message to the AI / MLE server to request to register with the AI / MLE server.

[0482] S904, the App (i.e., the model requester) sends an AI / ML model training request message (i.e., the third request message) to the AI / ML E server. Correspondingly, the AI / ML E server receives the AI / ML model training request message from the App.

[0483] The AI / ML model training request message includes a second training request (ML model training request).

[0484] For the specific implementation of the above steps S901-S904, please refer to the relevant description of steps S601-S604 of the above method 600, which will not be repeated here for the sake of brevity.

[0485] S905, the AI / MLE server determines the AI / MLE client#1 and the AI / MLE client#2.

[0486] For the specific implementation, please refer to the relevant description of step S820 of the above method 800.

[0487] Among them, AI / ML E client#2 can be regarded as an alternative federated learning member of AI / ML E client#1, that is, AI / ML E client#2 can continue to execute the training task of the first model when AI / ML E client#1 cannot continue to execute the training task of the first model to ensure that the first model can be trained to completion, that is, AI / ML E client#2 supports continuing to execute the training task of the first model, or in other words, AI / ML E client#2 has the ability to continue to execute the training task of the first model.

[0488] Exemplarily, the AI / ML E server determines that AI / ML E client #1 performs the training task of the first model, and selects AI / ML E client #2 as a candidate federated learning member of AI / ML E client #1 based on the second training requirement and the capability information of the candidate AI / ML E clients (for example, including at least AI / ML E client #1 and AI / ML E client #2).

[0489] S906, the AI / ML E server sends an AI / ML model training request message (i.e., a fourth request message) to the AI / ML E client#1. Correspondingly, the AI / ML E client#1 receives the AI / ML model training request message from the AI / ML E server.

[0490] The AI / ML model training request message is used to request AI / ML E client#1 to perform the training task of the first model. The AI / ML model training request message includes the second training requirement and information of AI / ML E client#2.

[0491] S907, when AI / ML E client #1 determines that it cannot continue to perform the training task and / or determines that the intermediate training result cannot meet the second training requirement, it determines the first training requirement. Specifically, AI / ML E client #1 determines the first training requirement according to the second training requirement and the intermediate training result. For a specific implementation method, refer to the relevant description of step S850 of the above method 800.

[0492] Exemplarily, AI / ML E client #1 performs local model training according to the second training requirement to obtain an intermediate training result.

[0493] S908, AI / ML E client#1 sends a model training request message (i.e., a second request message, such as ML model training request) or a FL model transfer request message (i.e., a second request message, such as FL model transfer request) to AI / ML E client#2, and correspondingly, AI / ML E client#2 receives the model training request message or the FL model transfer request message from AI / ML E client#1.

[0494] The model training request message or the FL model transmission request message is used to request AI / ML E client#2 to continue to perform the training task of the first model. Exemplarily, the model training request message or the FL model transmission request message carries the intermediate training result and the first training requirement.

[0495] S909, AI / ML E client#2 continues to execute the training task according to the intermediate training result and the first training requirement, and obtains the first model and the training information of the first model.

[0496] Optionally, before executing step S909, AI / ML E client#2 may determine whether to continue training the model according to the intermediate training result and the first training requirement.

[0497] S910, AI / ML E client#2 sends a model training response message (i.e., a second response message, such as an ML model training response) or a FL model transfer response message (i.e., a second response message, such as an FL model transfer response) to AI / ML E client#1. Correspondingly, AI / ML E client#1 receives a model training response message or a FL model transfer response message from AI / ML E client#2.

[0498] S911, AI / ML E client#2 sends a FL model report (e.g., FL local mode report) to the AI / ML E server, and correspondingly, the AI / ML E server receives the FL model report from the AI / ML E client#2, wherein the FL model report includes the first model and the training information of the first model.

[0499] S912, AI / ML E server performs model aggregation.

[0500] For the specific implementation of the above steps S908-S912, please refer to the relevant description of steps S611-S615 of the above method 600.

[0501] Based on the above scheme, the AI / ML E server can select AI / ML Eclient#1 and a backup learning member of AI / ML E client#1, that is, AI / ML E client#2, from the candidate AI / ML E clients based on the second training requirement carried in the received third request message and the capability information of the candidate AI / ML Eclient, to continue to perform the training task of the first model and ensure that the training of the first model is completed. That is, when AI / ML Eclient#1 determines that AI / ML Eclient#1 cannot continue to perform the training task of the first model and / or the intermediate training result of the first model does not meet the second training requirement corresponding to the first model, it can directly request AI / ML Eclient#2 to continue to perform the training task of the first model without exchanging information with the AI / ML E server, which can improve the model training efficiency, avoid the increase of model convergence time, and avoid the degradation of model performance.

[0502] It should be understood that the sequence numbers of the above processes do not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0503] It should also be understood that the present application will present various aspects, embodiments, or features around a system that may include multiple devices, components, modules, etc. It should be understood and appreciated that each system may include additional devices, components, modules, etc., and / or may not include all of the devices, components, modules, etc. discussed in conjunction with the figures. In addition, combinations of these schemes may also be used.

[0504] It should also be understood that in some of the above embodiments, the devices in the existing network architecture are mainly used as examples for exemplary description (it should be understood that the specific form of the device is not limited in the embodiments of the present application. For example, devices that can achieve the same function in the future are applicable to the embodiments of the present application.

[0505] It can be understood that in the above-mentioned method embodiments, the methods and operations implemented by the device (for example, the second federated learning member, the federated learning server, the first federated learning member) can also be implemented by components of the device (for example, a chip or a circuit).

[0506] Above, combined Figures 1 to 9The communication method provided in the embodiment of the present application is described in detail. The above communication method is mainly introduced from the perspective of interaction between the second federated learning member, the federated learning server, and the first federated learning member. It can be understood that in order to realize the above functions, the second federated learning member, the federated learning server, and the first federated learning member include hardware structures and / or software modules corresponding to the execution of each function.

[0507] Those skilled in the art should be aware that, in combination with the units and algorithm steps of each example described in the embodiments disclosed herein, the present application can be implemented in the form of hardware or a combination of hardware and computer software. Whether a function is performed in the form of hardware or computer software driving hardware depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0508] The following, combined Fig.10 and Fig.11 The communication device provided in the embodiment of the present application is described in detail. The description of the device embodiment corresponds to the description of the method embodiment, so the contents not described in detail can be referred to the above method embodiment, and some contents are not described for brevity.

[0509] The embodiment of the present application can divide the functional modules of the communication device according to the above method example. For example, each functional module can be divided corresponding to each function, or two or more functions can be integrated into one processing module. The above integrated module can be implemented in the form of hardware, or in the form of software functional modules, or a combination of software and hardware. The division of modules in the embodiment of the present application is schematic, which is only a logical function division. There may be other division methods in actual implementation. The following is an example of dividing each functional module corresponding to each function.

[0510] Fig.10 1 is a schematic block diagram of a communication device 1000 provided in an embodiment of the present application. Fig.10 As shown, the communication device 1000 includes a processing module 1010 and a communication module 1020. The communication device 1000 may be a transmitting end device, or a communication device applied to a transmitting end device or used in combination with a transmitting end device and capable of implementing a method executed by the transmitting end device, such as a chip, a chip system or a circuit; or the communication device 1000 may be a receiving end device, or a communication device applied to a receiving end device or used in combination with a receiving end device and capable of implementing a method executed by the receiving end device, such as a chip, a chip system or a circuit.

[0511] The communication module may also be referred to as a transceiver module, a transceiver, a transceiver, a transceiver unit or a transceiver device, etc. The processing module may also be referred to as a processor, a processing board, a processing unit or a processing device, etc. Optionally, the communication module is used to perform the sending operation and the receiving operation of the sending end device and the receiving end device in the above method, and the device used to implement the receiving function in the communication module may be regarded as a receiving unit, and the device used to implement the sending function in the communication module may be regarded as a sending unit, that is, the communication module includes a receiving unit and a sending unit.

[0512] Optionally, the communication device 1000 may further include a storage module 1030 for storing device program code and / or data.

[0513] In one example, when the communication device 1000 is applied to a federated learning server, the processing module 1010 can be used to implement the processing function of the federated learning server in the above embodiment, and the communication module 1020 can be used to implement the sending and receiving function of the federated learning server in the above embodiment.

[0514] In another example, when the communication device 1000 is applied to the first federated learning member, the processing module 1010 can be used to implement the processing function of the first federated learning member in the above embodiment, and the communication module 1020 can be used to implement the sending and receiving function of the first federated learning member in the above embodiment.

[0515] In another example, when the communication device 1000 is applied to a second federated learning member, the processing module 1010 can be used to implement the processing function of the second federated learning member in the above embodiment, and the communication module 1020 can be used to implement the sending and receiving function of the second federated learning member in the above embodiment.

[0516] In addition, it should be noted that the aforementioned communication module and / or processing module can be implemented by a virtual module, for example, the processing module can be implemented by a software function unit or a virtual device, and the communication module can be implemented by a software function or a virtual device. Alternatively, the processing module or the communication module can also be implemented by a physical device, for example, if the device is implemented using a chip / circuit (such as an integrated circuit or a logic circuit, etc.). The communication module can be an input-output circuit and / or a communication interface, performing input operations (corresponding to the aforementioned receiving operations) and output operations (corresponding to the aforementioned sending operations); the processing module is an integrated processor or microprocessor or circuit (such as an integrated circuit or a logic circuit, etc.).

[0517] The division of modules in this application is schematic and is only a logical function division. There may be other division methods in actual implementation. In addition, each functional module in each example of this application may be integrated into one processor, or may exist physically separately, or two or more modules may be integrated into one module. The above-mentioned integrated modules may be implemented in the form of hardware or in the form of software functional modules.

[0518] In one example, the functional unit in any of the above devices can be one or more integrated circuits configured to implement the above methods, such as: one or more application specific integrated circuits (ASICs), or, one or more central processing units (CPUs), one or more microprocessors (MCUs), one or more digital signal processors (DSPs), or, one or more field programmable gate arrays (FPGAs), or a combination of at least two of these integrated circuit forms.

[0519] In an example, the storage module 1030 may include a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory and / or a register, etc.

[0520] Fig.11 2 is a schematic block diagram of a communication device 2000 provided in an embodiment of the present application. Optionally, the communication device 2000 may be a chip or a chip system. Optionally, in the present application, the chip system may be composed of a chip, or may include a chip and other discrete devices.

[0521] like Fig.11 As shown, the communication device 2000 can be used to implement the functions of any device (e.g., terminal device, network device) in the communication system described in the above examples. The communication device 2000 may include at least one processor 2010. Optionally, the processor 2010 is coupled to a memory, and the memory may be located within the device, or the memory may be integrated with the processor, or the memory may be located outside the device. For example, the communication device 2000 may also include at least one memory 2020. The memory 2020 stores the necessary computer programs, computer programs or instructions and / or data for implementing any of the above examples; the processor 2010 may execute the computer program stored in the memory 2020 to complete the method in any of the above examples.

[0522] The communication device 2000 may also include a communication interface 2030, and the communication device 2000 may exchange information with other devices through the communication interface 2030. Exemplarily, the communication interface 2030 may be a transceiver, a circuit, a bus, a module, a pin, or other types of communication interfaces. When the communication device 2000 is a chip-type device or circuit, the communication interface 2030 in the device 2000 may also be an input-output circuit, which may input information (or receive information) and output information (or send information), and the processor 2010 may be an integrated processor, a microprocessor, an integrated circuit or a logic circuit, etc. The processor may determine the output information based on the input information.

[0523] In one example, when the communication device 2000 is applied to a federated learning server, the processor 2010 may be used to implement the processing function of the federated learning server in the above embodiment, and the communication interface 2030 may be used to implement the sending and receiving function of the federated learning server in the above embodiment.

[0524] In another example, when the communication device 2000 is applied to the first federated learning member, the processor 2010 can be used to implement the processing function of the first federated learning member in the above embodiment, and the communication interface 2030 can be used to implement the sending and receiving function of the first federated learning member in the above embodiment.

[0525] In another example, when the communication device 2000 is applied to a second federated learning member, the processor 2010 may be used to implement the processing function of the second federated learning member in the above embodiment, and the communication interface 2030 may be used to implement the transceiver function of the second federated learning member in the above embodiment.

[0526] The coupling in this application is an indirect coupling or communication connection between devices, units or modules, which can be electrical, mechanical or other forms, and is used for information exchange between devices, units or modules. The processor 2010 may cooperate with the memory 2020 and the communication interface 2030. The specific connection medium between the above-mentioned processor 2010, memory 2020 and communication interface 2030 is not limited in this application.

[0527] Alternatively, if Fig.11 As shown in , the processor 2010, the memory 2020 and the communication interface 2030 are connected to each other via a bus 2040. Optionally, the bus may include an address bus, a data bus, a control bus and other types of buses. In addition, for ease of representation, Fig.11 Although one bus 2040 is shown in FIG. 1 , this does not mean that there is only one bus or only one type of bus.

[0528] It should be understood that the processor mentioned in the embodiments of the present application can be the following devices or part of the circuit used for processing functions in the following devices: CPU, and can also be other general-purpose processors, DSP, ASIC, FPGA or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc.

[0529] It should also be understood that the memory mentioned in the embodiments of the present application may be a volatile memory and / or a 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). For example, RAM can be used as an external cache. By way of example and not limitation, RAM includes the following forms: static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct rambus RAM (DR RAM).

[0530] It should be noted that when the processor is a general-purpose processor, DSP, ASIC, FPGA or other programmable logic device, discrete gate or transistor logic device, discrete hardware component, the memory (storage module) can be integrated into the processor.

[0531] It should also be noted that the memory described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0532] An embodiment of the present application also provides a computer-readable storage medium on which computer instructions are stored for implementing the methods performed by a communication device (e.g., a federated learning server, a first federated learning member, or a second federated learning member) in the above-mentioned method embodiments.

[0533] An embodiment of the present application also provides a computer program product, comprising instructions, which, when executed by a computer, implement the methods performed by a communication device (e.g., a federated learning server, a first federated learning member, or a second federated learning member) in the above-mentioned method embodiments.

[0534] An embodiment of the present application also provides a communication system, which includes the federated learning server and / or the first federated learning member in the above embodiment.

[0535] Optionally, the communication system further includes the second federated learning member in the above embodiment.

[0536] The explanation of the relevant contents and beneficial effects of any of the above-mentioned devices can be referred to the corresponding method embodiments provided above, which will not be repeated here.

[0537] In various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0538] The present application will present various aspects, embodiments or features around a system that may include multiple devices, components, modules, etc. It should be understood and appreciated that each system may include additional devices, components, modules, etc., and / or may not include all of the devices, components, modules, etc. discussed in conjunction with the figures. In addition, combinations of these schemes may also be used.

[0539] In the present application, under the premise of no logical contradiction, the examples may reference each other, for example, the methods and / or terms between method embodiments may reference each other, for example, the functions and / or terms between device embodiments may reference each other, for example, the functions and / or terms between device examples and method examples may reference each other.

[0540] It should be understood that in some of the above embodiments, the devices in the existing network architecture are mainly used as examples for exemplary description, and the specific form of the device is not limited in the embodiments of the present application. For example, devices that can achieve the same function in the future are applicable to the embodiments of the present application.

[0541] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0542] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be described again here.

[0543] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0544] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0545] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0546] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the existing solution or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard drives, ROM, RAM, magnetic disks, or optical disks.

[0547] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A communication method, characterized in that: Applied to federated learning servers, including: Receiving a first request message from a first federated learning member, where the first request message includes a first training requirement corresponding to the first model; According to the capability information of the candidate federated learning members and the first training requirement, a second federated learning member is determined from the candidate federated learning members, the capability information indicates whether the candidate federated learning members support continuing to execute the training task of the first model, at least one of the candidate federated learning members supports continuing to execute the training task of the first model, and the at least one federated learning member includes the second federated learning member.

2. A communication method, characterized in that: Applied to federated learning servers, including: receiving a first request message from a first federated learning member, where the first request message includes an intermediate training result of the first model, where the intermediate training result is obtained by the first federated learning member executing a training task of the first model; Determining a first training requirement corresponding to the first model according to the intermediate training result; According to the capability information of the candidate federated learning members and the first training requirement, a second federated learning member is determined from the candidate federated learning members, the capability information indicates whether the candidate federated learning members support continuing to execute the training task of the first model, at least one of the candidate federated learning members supports continuing to execute the training task of the first model, and the at least one federated learning member includes the second federated learning member.

3. The method according to claim 1 or 2, characterized in that: The first request message also includes indication information, where the indication information is used to indicate that the first federated learning member cannot continue to perform the training task of the first model, and / or the indication information is used to indicate that at least one federated learning member is found.

4. The method according to any one of claims 1 to 3, characterized in that The method further comprises: A first response message is sent to the first federated learning member, where the first response message includes information about the second federated learning member.

5. The method according to any one of claims 1 to 3, characterized in that The method further comprises: A second request message is sent to the second federated learning member, where the second request message is used to request to continue the training task of the first model, and the second request message includes the first training requirement and the intermediate training result of the first model.

6. The method according to any one of claims 1 to 5, characterized in that The intermediate training result of the first model includes at least one of the following: A temporary model obtained by executing the training task of the first model; The accuracy information of the temporary model; The accuracy information required by the training task of the first model; The training time of the temporary model; Continue to perform the training task of the first model for a training time required; Resource usage of the temporary model; Continue to execute the training resources required for the training task of the first model; The number of training rounds for the temporary model; or, Continue to perform the training task of the first model for the number of training rounds required.

7. The method according to any one of claims 1 to 6, characterized in that Before determining a second federated learning member from the candidate federated learning members according to the capability information of the candidate federated learning members and the first training requirement, the method further includes: Acquire capability information of the candidate federated learning member.

8. The method according to any one of claims 1 to 7, characterized in that The capability information is also used to indicate at least one of the following: The training time supported by the candidate federated learning member; The available training time corresponding to the candidate federated learning member; The training resources supported by the candidate federated learning member; The available training resources corresponding to the candidate federated learning member; or, The number of model training rounds supported by the candidate federated learning member.

9. The method according to any one of claims 1 to 8, characterized in that The first training requirement includes at least one of the following: Continue to execute the accuracy information required by the training task of the first model; Continue to perform the training task of the first model for a training time required; Continue to execute the training resources required for the training task of the first model; or, Continue to perform the training task of the first model for the number of training rounds required.

10. The method according to any one of claims 1 to 9, characterized in that Before receiving the first request message from the first federated learning member, the method further includes: receiving a third request message from a model requester, where the third request message is used to request execution of a training task for the first model, and the third request message includes a second training requirement corresponding to the first model; Determining the first federated learning member from the candidate federated learning members according to the capability information of the candidate federated learning members and the second training requirement; Sending a fourth request message to the first federated learning member, where the fourth request message is used to request execution of a training task for the first model, and the fourth request message includes the second training requirement; The first training requirement is determined according to the second training requirement and the intermediate training result of the first model.

11. The method according to claim 10, characterized in that The second training requirement includes at least one of: an analysis identifier corresponding to the first model; A service identifier corresponding to the first model; An application identifier corresponding to the first model; Acquire the time of the first model; the training duration of the first model; The accuracy information required by the training task of the first model; The training time required for the training task of the first model; The training resources required by the training task of the first model; or The number of training rounds required for the training task of the first model.

12. The method according to any one of claims 1 to 11, characterized in that The first federated learning member or the second federated learning member is an enabled client or an enabled server.

13. A communication method, characterized in that: Applicable to First Federation Learning members, including: When it is determined that the first federated learning member cannot continue to perform the training task of the first model, and / or when it is determined that the intermediate training result of the first model does not meet the second training requirement corresponding to the first model, determine the first training requirement corresponding to the first model according to the intermediate training result and the second training requirement, wherein the intermediate training result is obtained by the first federated learning member performing the training task of the first model; Sending a first request message to a federated learning server, where the first request message includes the first training requirement; A first response message is received from the federated learning server, where the first response message includes information of a second federated learning member, and the second federated learning member supports continuing to execute the training task of the first model.

14. A communication method, characterized in that: Applicable to First Federation Learning members, including: When it is determined that the first federated learning member cannot continue to perform the training task of the first model, and / or when it is determined that the intermediate training result of the first model does not meet the second training requirement corresponding to the first model, a first request message is sent to the federated learning server, wherein the first request message includes the intermediate training result, the intermediate training result is used to determine the first training requirement corresponding to the first model, and the intermediate training result is obtained by the first federated learning member performing the training task of the first model; A first response message is received from the federated learning server, where the first response message includes information of a second federated learning member, and the second federated learning member supports continuing to execute the training task of the first model.

15. The method according to claim 13 or 14, characterized in that The method further comprises: A second request message is sent to the second federated learning member, where the second request message is used to request to continue the training task of the first model, and the second request message includes the first training requirement and the intermediate training result.

16. The method according to any one of claims 13 to 15, characterized in that The first request message also includes indication information, where the indication information is used to indicate that the first federated learning member cannot continue to perform the training task of the first model, and / or the indication information is used to indicate that at least one federated learning member is found.

17. The method according to any one of claims 13 to 16, characterized in that Before sending the first request message to the federated learning server, the method further includes: Sending capability information of the first federated learning member to the federated learning server, where the capability information indicates that the first federated learning member supports executing the training task of the first model.

18. The method according to any one of claims 13 to 17, characterized in that Before determining the first training requirement corresponding to the first model according to the intermediate training result and the second training requirement, or before sending a first request message to the federated learning server, the method further includes: receiving a fourth request message from the federated learning server, where the fourth request message is used to request execution of a training task for the first model, and the fourth request message includes the second training requirement; The training task of the first model is performed according to the second training requirement to obtain the intermediate training result.

19. The method according to any one of claims 13 to 18, characterized in that The first training requirement includes at least one of the following: Continue to execute the accuracy information required by the training task of the first model; Continue to perform the training task of the first model for a training time required; Continue to execute the training resources required for the training task of the first model; or, Continue to perform the training task of the first model for the number of training rounds required.

20. The method according to any one of claims 13 to 19, characterized in that The intermediate training result includes at least one of the following: A temporary model obtained by executing the training task of the first model; The accuracy information of the temporary model; The accuracy information required by the training task of the first model; The training time of the temporary model; Continue to perform the training task of the first model for a training time required; Resource usage of the temporary model; Continue to execute the training resources required for the training task of the first model; The number of training rounds for the temporary model; or, Continue to perform the training task of the first model for the number of training rounds required.

21. A communication device, characterized in that: The method comprises a module for implementing the method according to any one of claims 1 to 12, or a module for implementing the method according to any one of claims 13 to 20.

22. A communication device, characterized in that: The invention comprises at least one processor configured to execute a computer program or instructions so that the method as claimed in any one of claims 1 to 12 is executed, or so that the method as claimed in any one of claims 15 to 19 is executed.

23. The communication device according to claim 22, characterized in that: The communication device further comprises a memory, wherein the memory is used to store the computer program or instruction; and / or, The communication device further comprises a communication interface, which is coupled to the at least one processor and is used to input and / or output information.

24. A communication system, characterized in that: It includes a federated learning server and / or a first federated learning member, wherein the federated learning server is used to execute the method as described in any one of claims 1 to 12, and the first federated learning member is used to execute the method as described in any one of claims 13 to 20.

25. The communication system according to claim 24, characterized in that The communication system further includes a second federated learning member, wherein the second federated learning member is used to: receiving a second request message from the first federated learning member or the federated learning server, where the second request message is used to request to continue to perform the training task of the first model, and the second request message includes an intermediate training result of the first model and a first training requirement corresponding to the first model; The training task is continued to be performed according to the intermediate training result and the first training requirement to obtain the first model and the training information of the first model.

26. A computer-readable storage medium, characterized in that: The computer-readable storage medium is used to store a computer program, and when the computer program is run on a computer, the method according to any one of claims 1 to 12 is executed, or the method according to any one of claims 13 to 20 is executed.

27. A computer program product, characterized in that The method comprises a computer program or an instruction, which, when executed by a processor, causes the method according to any one of claims 1 to 12 to be executed, or causes the method according to any one of claims 13 to 20 to be executed.

Images