Joint learning method and device, network analysis function, storage medium and computer program product

Abstract

The invention discloses a joint learning method and device, a network analysis function, a storage medium and a computer program product, and the method comprises the steps: a first network analysis function carries out the joint learning based on the computing power information of the first network analysis function and the computing power information of each second network analysis function participating in the joint learning; and performing joint learning on the first model in cooperation with all the second network analysis functions participating in joint learning.

Description

Technical Field

[0001] The present application relates to the field of communication technology, and in particular to a joint learning method, device, network analysis function, storage medium and computer program product. Background Art

[0002] In the existing network intelligence architecture, the network data analysis function (NWDAF, Network Data AnalyticsFunction) can support federated learning (federated learning is also called joint learning), that is, multiple NWDAFs can jointly train the model to solve the problem of data privacy, and ensure that data in different regions does not leave the local area and is not visible to third parties; however, the existing joint learning solution has the problem of low training efficiency when the computing power of each training node is unbalanced. The training efficiency is limited by the low computing power nodes, and the resources of the remaining computing power nodes cannot be fully utilized. Summary of the invention

[0003] To solve related technical problems, the embodiments of the present application provide a joint learning method, device, network analysis function, storage medium and computer program product.

[0004] The technical solution of the embodiment of the present application is implemented as follows:

[0005] The present application embodiment provides a joint learning method, which is applied to a first network analysis function, and the method includes:

[0006] Based on the computing power information of the first network analysis function and the computing power information of each second network analysis function participating in the joint learning, the first model is jointly learned in collaboration with all the second network analysis functions participating in the joint learning.

[0007] In the above solution, the computing power information based on the first network analysis function and the computing power information of each second network analysis function participating in the joint learning, and all the second network analysis functions participating in the joint learning, jointly learning the first model, include:

[0008] Based on the first information, determine the second information and update the parameters of the sub-model of the first network analysis function; wherein the first information is obtained by forward calculation of the second network analysis function based on the sub-model of the second network analysis function, and the second information is used to update the parameters of the sub-model of the second network analysis function, and the sub-model is obtained by disassembling the first model based on the computing power information;

[0009] Second information is sent to each second network analysis function.

[0010] In the above solution, the determining of the second information based on the first information and updating the parameters of the sub-model of the first network analysis function include:

[0011] Perform reverse calculation based on the first information and the sub-model of the first network analysis function to obtain second information corresponding to the first information;

[0012] Parameters of the sub-model of the first network analysis function are updated based on all the first information.

[0013] In the above scheme, the method further includes:

[0014] The first information sent by each second network analysis function participating in the joint learning is received.

[0015] In the above solution, the computational effort of the sub-model of the second network analysis function matches the computing power of the second network analysis function; and / or

[0016] The computational effort of the sub-model of the first network analysis function matches the computing power of the first network analysis function.

[0017] In the above scheme, the method further includes:

[0018] Based on the computing power information of the first network analysis function and the computing power information of each second network analysis function participating in the joint learning, the first model is disassembled into N sub-models, where N is the total number of the first network analysis functions and all the second network analysis functions participating in the joint learning, and one sub-model corresponds to one first network analysis function or one second network analysis function;

[0019] A sub-model is sent to each second network analysis function participating in the joint learning; wherein the second network analysis function is used to perform forward calculation based on the sub-model to obtain first information, and the first network analysis function is used to perform reverse calculation based on the sub-model of the first network analysis function and the first information sent by the second network analysis function to obtain second information, and the second information is used to update the parameters of the sub-model.

[0020] In the above solution, the first model is decomposed into N sub-models based on the computing power information of the first network analysis function and the computing power information of each second network analysis function participating in the joint learning, including:

[0021] Based on the computing power information of the first network analysis function and the computing power information of each second network analysis function participating in the joint learning, dividing the first model into one or more independent layers and one or more shared layers;

[0022] Using all shared layers in the first model as submodels of the first network function;

[0023] Based on the computing power information of all second network analysis functions, the independent layers in the first model are disassembled or expanded to obtain a sub-model of each second network analysis function.

[0024] In the above scheme, the computing power information includes one or more of the following:

[0025] Processor resource utilization;

[0026] Memory usage;

[0027] Video memory usage;

[0028] The computing power of the processor.

[0029] In the above scheme, the method further includes:

[0030] Sending a first request to the network storage function, the first request being used to discover a second network analysis function;

[0031] Receiving computing power information of one or more second network analysis functions sent by the network storage function;

[0032] Based on the computing power information of the second network analysis function sent by the network storage function, a second network analysis function participating in the joint learning is determined.

[0033] In the above scheme, the method further includes:

[0034] Send a second request to the network storage function; wherein the second request carries at least the computing power information of the first network analysis function, and is used to request registration of the first network analysis function.

[0035] In the above scheme, the method further includes:

[0036] receiving a third request; the third request carrying a first identifier, for requesting to subscribe to an analysis result of an analysis service corresponding to the first identifier;

[0037] Determine the first model that matches the analysis service corresponding to the first identifier.

[0038] The present application also provides a joint learning method, which is applied to the second network analysis function, and the method includes:

[0039] Based on the computing power information of the second network analysis function and the computing power information of the first network analysis function, jointly learn the first model in collaboration with the first network analysis function.

[0040] In the above solution, the computing power information based on the second network analysis function and the computing power information of the first network analysis function, in collaboration with the first network analysis function, jointly learning the first model includes:

[0041] Performing forward calculation based on the sub-model of the second network analysis function to obtain first information, and sending the first information to the first network analysis function, wherein the sub-model is disassembled from the first model based on the computing power information;

[0042] receiving second information sent by the first network analysis function, where the second information is obtained by reverse calculation by the first network analysis function based on the first information and a sub-model of the first network analysis function;

[0043] Parameters of the sub-model of the second network analysis function are updated based on the second information.

[0044] In the above solution, before the sub-model based on the second network analysis function performs forward calculation to obtain the first information, the method further includes:

[0045] Receive the sub-model sent by the first network analysis function.

[0046] In the above scheme, the sub-model is obtained by decomposing the first model by the first network analysis function based on the computing power information of the first network analysis function and the computing power information of each second network analysis function participating in the joint learning, and one sub-model corresponds to one first network analysis function or one second network analysis function.

[0047] In the above scheme, the computing power information includes one or more of the following:

[0048] Processor resource utilization;

[0049] Memory usage;

[0050] Video memory usage;

[0051] The computing power of the processor.

[0052] In the above scheme, the method further includes:

[0053] A fourth request is sent to the network storage function, where the fourth request carries at least computing power information of the second network analysis function and is used to request registration of the second network analysis function.

[0054] The present application also provides a joint learning device, comprising:

[0055] The first training unit is used to jointly learn the first model with all the second network analysis functions participating in the joint learning based on the computing power information of the first network analysis function and the computing power information of each second network analysis function participating in the joint learning.

[0056] The present application also provides a joint learning device, comprising:

[0057] The second training unit is used to jointly learn the first model with the first network analysis function based on the computing power information of the second network analysis function and the computing power information of the first network analysis function.

[0058] The present application also provides a first network analysis function, characterized by comprising: a first processor and a first communication interface; wherein,

[0059] The first processor is used to jointly learn the first model with all the second network analysis functions participating in the joint learning based on the computing power information of the first network analysis function and the computing power information of each second network analysis function participating in the joint learning.

[0060] The present application also provides a second network analysis function, including: a second processor and a second communication interface; wherein,

[0061] The second processor is used to collaborate with the first network analysis function to jointly learn the first model based on the computing power information of the second network analysis function and the computing power information of the first network analysis function.

[0062] The present application also provides a network analysis function, comprising a processor and a memory for storing a computer program that can be run on the processor,

[0063] Wherein, when the processor is used to run the computer program, it executes the steps of any method on the first network analysis function side, or executes the steps of any method on the second network analysis function side.

[0064] The present application also provides a storage medium having a computer program stored thereon, characterized in that when the computer program is executed by a processor, the computer program implements the steps of any method on the first network analysis function side, or implements the steps of any method on the second network analysis function side.

[0065] The present application also provides a computer program product, including a computer program, characterized in that the computer program implements the steps of any of the above methods when executed by a processor.

[0066] In the joint learning method, device, network analysis function, storage medium and computer program product provided in the embodiments of the present application, the first network analysis function cooperates with all the second network analysis functions participating in the joint learning to jointly learn the first model based on the computing power information of the first network analysis function and the computing power information of each second network analysis function participating in the joint learning; the second network analysis function cooperates with the first network analysis function to jointly learn the first model based on the computing power information of the second network analysis function and the computing power information of the first network analysis function. It can be seen that in the embodiments of the present application, the first network analysis function and the second network analysis function jointly learn the first model, and the computing power of the first network analysis function and the second network analysis function are comprehensively considered in the process of jointly training the first model, which can make full use of the computing power resources of each network analysis function, improve resource utilization, and eliminate the barrel effect; compared with the solution in the prior art in which the Client NWDAF (second network analysis function) statically performs model joint training, the above solution improves the model training efficiency while ensuring data privacy. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 A schematic diagram of an interactive process of a joint learning method according to an embodiment of the present application;

[0068] Figure 2 This is an example diagram of the first network analysis function and the second network analysis function cooperating to jointly learn the first model in an embodiment of the present application;

[0069] Figure 3 A schematic diagram of a joint learning method implementation process of an embodiment of the present application;

[0070] Figure 4 A schematic diagram of a joint learning method implementation process of an embodiment of the present application;

[0071] Figure 5 This is a schematic diagram of the structure of a joint learning device according to an embodiment of the present application;

[0072] Figure 6 This is a schematic diagram of the first network analysis function structure of an embodiment of the present application;

[0073] Figure 7 This is a schematic diagram of the structure of the second network analysis function in the embodiment of the present application;

[0074] Figure 8 This is a schematic diagram of the structure of a joint learning device according to an embodiment of the present application. DETAILED DESCRIPTION

[0075] In the related technology, the server NWDAF can select multiple different client NWDAFs; multiple different Client NWDAFs collect data from different regions and perform joint model training to obtain relevant data models. However, each Client NWDAF needs to perform the forward and backward calculations of the entire model, which has a large computing load; especially when the computing power of each Client NWDAF is different, it will lead to inconsistent completion time, obvious "barrel effect", and reduce the overall training efficiency.

[0076] Based on this, in various embodiments of the present application, the first network analysis function cooperates with all second network analysis functions participating in the joint learning to jointly learn the first model based on the computing power information of the first network analysis function and the computing power information of each second network analysis function participating in the joint learning; the second network analysis function cooperates with the first network analysis function to jointly learn the first model based on the computing power information of the second network analysis function and the computing power information of the first network analysis function. It can be seen that in the embodiments of the present application, the first network analysis function and the second network analysis function jointly learn the first model, which reduces the computing load of a single second network analysis function. In the process of jointly training the first model, the computing power of the first network analysis function and the second network analysis function are comprehensively considered, and the computing power resources of each network analysis function can be fully utilized, which improves resource utilization and eliminates the barrel effect. Compared with the solution in the prior art in which only the Client NWDAF (second network analysis function) performs joint training of the model, the above solution improves the model training efficiency while ensuring data privacy.

[0077] The present application is further described in detail below in conjunction with the accompanying drawings and embodiments.

[0078] First, in order to better illustrate the joint learning method provided in the embodiment of the present application, Figure 1 An example diagram of the interaction of the joint learning system applicable to the embodiment of the present application. Figure 1 The federated learning system shown includes a client, a first network analysis function (e.g., Server NWDAF), N-1 second network analysis functions (e.g., Client NWDAF), a network storage function (NRF, Network Repository Function) and a network function (NF, Network Function), where N is an integer greater than 1, and NF acts as a data provider to provide the input data required for model training.

[0079] like Figure 1 As shown, the joint learning method includes the following steps:

[0080] Step 1: The first network analysis function and each second network analysis function are registered to the network storage function.

[0081] Here, each network analysis function registers a network function configuration (NF profile) to a network storage function, and the NF profile includes but is not limited to computing power information, and the computing power information can be described as computing power resource information. The first network analysis function includes Server NWDAF, and the second network analysis function includes Client NWDAF.

[0082] For example, the first network analysis function sends a second request to the network storage function; the second request carries at least computing power information of the first network analysis function, and is used to request registration of the first network analysis function.

[0083] For another example, the second network analysis function sends a fourth request to the network storage function; the fourth request carries at least computing power information of the second network analysis function, and is used to request registration of the second network analysis function.

[0084] The computing power information includes one or more of the following: processor resource occupancy, memory occupancy, video memory occupancy, and processor computing power measurement. Processors include but are not limited to central processing units (CPUs) and / or graphics processing units (GPUs). The computing power measurement includes but is not limited to one or more of the following: floating-point operations per second (FLOPS), Dhrystone Million Instructions Per Second (DMIPS), and operations per second (OPS).

[0085] Step 2: The consumer (Consumer) requests or subscribes to the analysis results of the analysis service (AnalyticsService) from the first network analysis function.

[0086] Here, the consumer can send a third request to the first network analysis function; the third request carries a first identifier (ID) for requesting to subscribe to the analysis results of the analysis service corresponding to the first identifier. The first identifier refers to the identifier of the analysis service, including but not limited to Analytics ID.

[0087] The first network analysis function may determine, based on the first identifier, a first model that can provide an analysis service corresponding to the first identifier. The first model includes, but is not limited to, one or more of the following: an artificial intelligence (AI) model, a machine learning (ML) model, a federated learning model, and a joint learning model.

[0088] It should be noted that consumers include but are not limited to terminals, network functions in the core network, radio access networks (RAN), and application functions (AF) of third parties.

[0089] Step 3: The first network analysis function discovers the second network analysis function from the network storage function, and selects the second network analysis function to participate in the joint learning based on the computing power information of the discovered second network analysis function.

[0090] For example, the first network analysis function sends a first request to the network storage function, and the first request is used to discover the second network analysis function. The discovery can be described as querying or obtaining; the second network analysis function receives the first request, discovers the second network analysis function based on the first request, and returns the computing power information of the discovered second network analysis function to the first network analysis function. The first network analysis function receives the computing power information of one or more second network analysis functions sent by the network storage function, and determines the second network analysis function participating in the joint learning based on the computing power information of the second network analysis function sent by the network storage function. The second network analysis function that can meet the computing needs of the analysis service requested by the consumer can be used as the second network analysis function participating in the joint learning; or the second network analysis function participating in the joint learning can be determined from the second network analysis functions that can provide computing power services or computing services in order from high to low computing power. The ability to provide computing power services or computing services can be understood as having available computing power resources, and computing power resources include but are not limited to one or more of the following: processor resources, memory resources, and video memory resources.

[0091] It should be noted that the second network analysis function may also feed back other information (except computing power information) discovered by the second network analysis function to the first network analysis function.

[0092] Step 4: The first network analysis function decomposes the first model into N sub-models based on the computing power information of the first network analysis function and the computing power information of each second network analysis function participating in the joint learning, and one sub-model corresponds to one first network analysis function or one second network analysis function.

[0093] Here, the total number of the first network analysis function and the second network analysis functions participating in the joint learning is N. The first network analysis function determines the maximum amount of computing that the first network analysis function can provide and the maximum amount of computing that each second network analysis function participating in the joint learning can provide based on the computing power information of the first network analysis function and the computing power information of each second network analysis function participating in the joint learning. Based on the maximum amount of computing that the first network analysis function can provide and the maximum amount of computing that each second network analysis function participating in the joint learning can provide, the first model is disassembled into N sub-models. The disassembly can be described as splitting or dividing, and the first model can be understood as a global model. The sub-model corresponding to the first network analysis function can be called a server-side model, and the sub-model corresponding to the second network analysis function can be called a client-side model. Ideally, the N sub-models are equivalent to the first model after being merged.

[0094] It should be noted that by decomposing the first model into N sub-models, the global model can be decomposed into different computing nodes (the first network analysis function and the second network analysis function), thereby reducing the computing load of a single node.

[0095] In order to eliminate the barrel effect, in one embodiment, the computational amount of the submodel of the second network analysis function matches the computing power of the second network analysis function; and / or, the computational amount of the submodel of the first network analysis function matches the computing power of the first network analysis function. The computational amount of a submodel includes the number of layers and / or the number of parameters of the submodel, and the number of parameters can be understood as the total number of weight parameters or model parameters. For example, for a second network analysis function with higher computing power, a submodel with more layers can be allocated; for a second network analysis function with lower computing power, a submodel with fewer layers can be allocated, so that the computational amount of the submodel matches the computing power of the second network analysis function. The computational amount can be understood as the amount of computing tasks, and the computing power can be understood as computing capacity.

[0096] In order to adaptively allocate the disassembly scheme of the first model according to the computing power information of each computing node (the first network analysis function and the second network analysis function) to eliminate the "barrel effect", in one embodiment, based on the computing power information of the first network analysis function and the computing power information of each second network analysis function participating in the joint learning, the first model is disassembled into N sub-models, including:

[0097] Based on the computing power information of the first network analysis function and the computing power information of each second network analysis function participating in the joint learning, dividing the first model into one or more independent layers and one or more shared layers;

[0098] Using all shared layers in the first model as submodels of the first network function;

[0099] Based on the computing power information of all second network analysis functions, the independent layers in the first model are disassembled or expanded to obtain a sub-model of each second network analysis function.

[0100] Here, based on the computing power information of the first network analysis function and the computing power information of each second network analysis function participating in the joint learning, the first computing power ratio of the first network analysis function to all the second network analysis functions participating in the joint learning can be determined. The first computing power ratio can be understood as the ratio of computing power or computing power ratio; based on the first computing power ratio, the first model is divided into one or more independent layers and one or more shared layers; all shared layers in the first model are used as sub-models of the first network function; based on the computing power information of each second network analysis function participating in the joint learning, the second computing power ratio between the second network analysis functions is determined, and based on the second computing power ratio, the independent layers in the first model are disassembled or expanded to obtain the sub-model of each second network analysis function. The first computing power ratio can be understood as the ratio of the computing power of the first network analysis function to the total computing power, and the total computing power is the sum of the computing power of the first network analysis function and the computing power of all the second network analysis functions participating in the joint learning. The second computing power ratio can be understood as the ratio of a second network analysis function to the sum of the computing power of all the second network analysis functions participating in the joint learning.

[0101] For example, when the first computing power ratio represents that the computing power of the first network analysis function is much greater than that of the second network analysis function, the number of shared layers is greater than the number of independent layers. For example, the total number of layers of the first model is 10, and layers 1 to 3 are used as independent layers, and layers 4 to 10 are used as shared layers. The adjustment coefficient of the independent layer is determined based on the second computing power ratio between the second network analysis functions, and the independent layers in the first model are disassembled or scaled based on the adjustment coefficient of the independent layer. When the independent layers in the first model are disassembled based on the adjustment coefficient of the independent layer, the independent layers contained in the sub-models of different second network analysis functions are at least partially different. When the independent layers in the first model are scaled, the dimensions of the independent layers are expanded or reduced on the basis of the determined independent layers to increase or reduce the amount of calculation or calculation requirements of the independent layers. In other words, all sub-models of the second network analysis function are obtained by scaling the same basic model (for example, layers 1 to 3).

[0102] It should be noted that, when the first computing power ratio represents that the computing power of the first network analysis function is less than or equal to the computing power of the second network analysis function, the number of shared layers is less than or equal to the number of independent layers.

[0103] Step 5: The first network analysis function distributes the corresponding sub-model and sub-model ID to each second network analysis function participating in the joint learning.

[0104] Here, the first network analysis function sends a sub-model to each of the N-1 second network analysis functions, and the first network analysis function locally stores the sub-model of the network analysis function so that the first network analysis function and the second network analysis function collaborate to jointly learn the first model.

[0105] Figure 2 An example diagram is shown in which a first network analysis function and a second network analysis function collaborate to jointly learn a first model. The second network analysis function is used to perform forward calculations based on a submodel of the second network analysis function, and the first network analysis function is used to perform reverse calculations based on the submodel of the first network analysis function and the forward calculation results sent by the second network analysis function.

[0106] Step 6: Each second network analysis function participating in the joint learning collects local data.

[0107] Here, the second network analysis function may collect local data from the NF. The second network analysis function may save the received sub-model and the ID of the sub-model.

[0108] It should be noted that different second network analysis functions correspond to different data providers NF, which can ensure that data from different regions do not leave the local area and are only processed in the corresponding second network analysis function, thereby ensuring data privacy.

[0109] Step 7: Each second network analysis function participating in the joint learning performs forward calculation based on the corresponding sub-model and local data to obtain the first information.

[0110] Here, the first information includes the activation value obtained by the forward calculation, and the first information can be understood as the intermediate model information or the forward calculation result.

[0111] Step 8: Each second network analysis function participating in the joint learning sends corresponding first information to the first network analysis function.

[0112] Step 9: The first network analysis function receives the first information sent by each second network analysis function participating in the joint learning; based on the first information, determines the second information and updates the parameters of the sub-model of the first network analysis function.

[0113] Here, the first network analysis function determines the second information based on the first information, and updates the parameters of the sub-model of the first network analysis function based on all the first information.

[0114] Among them, based on the first information, determining the second information and updating the parameters of the sub-model of the first network analysis function can be done by performing reverse calculation based on the first information and the sub-model of the first network analysis function to obtain the second information corresponding to the first information, and updating the parameters of the sub-model of the first network analysis function based on all the first information.

[0115] The first network analysis function may determine the second information in any of the following ways:

[0116] The first network analysis function can perform reverse calculation based on the sub-model of the first network analysis function and the first information sent by a second network analysis function to obtain a second information; that is, a second information corresponds to the first information sent by a second network analysis function. The second information corresponding to different second network analysis functions can be partially the same or completely different.

[0117] The first network analysis function can also perform reverse calculation based on the sub-model of the first network analysis function and the first information sent by each second network analysis function to obtain second information, where the second information is a set of parameters, and the set of parameters includes multiple copies, which respectively correspond to the first information of each second network analysis function participating in the joint learning.

[0118] It should be noted that the second information includes the intermediate gradient obtained by reverse calculation, and the second information can be understood as intermediate model information or reverse calculation results.

[0119] In actual application, the implementation process of determining the second information based on the first information and updating the parameters of the sub-model of the first network analysis function can be: reverse calculation is performed based on each first information and the sub-model of the first network analysis function to obtain the third information corresponding to each first information, and the second information corresponding to each first information is determined based on the third information corresponding to each first information; and the parameters of the sub-model of the first network analysis function are updated based on all the third information. The third information includes the intermediate gradient obtained by reverse calculation.

[0120] Step 10: The first network analysis function sends second information to each second network analysis function participating in the joint learning.

[0121] Step 11: Each second network analysis function participating in the joint learning receives the second information, and updates the parameters of the sub-model of the second network analysis function based on the received second information.

[0122] Here, the second network analysis function locally updates the parameters of the sub-model of the second network analysis function based on the received second information.

[0123] It should be noted that after the parameters of the sub-model are updated, the second network analysis function can directly enter the next round of training, for example, performing forward calculations again based on local data and the updated sub-model until the set convergence conditions are reached or a stop training instruction sent by the first network analysis function is received.

[0124] Step 12: The first network analysis function reports the current training status information to the consumer, requesting the consumer to make a decision on the training strategy.

[0125] Here, the training state information may include but is not limited to the precision and / or training progress and / or accuracy of the first model. The training progress includes but is not limited to the training rounds.

[0126] Step 13: The consumer determines whether the first model meets the requirements, and if the first model does not meet the requirements, modifies the subscription requirements to the first network analysis function to update or terminate model training.

[0127] Here, modifying the subscription requirement can be understood as modifying the precision and / or accuracy and / or training rounds and / or convergence conditions of the first model. Modifying the subscription requirement can be described as modifying the model subscription.

[0128] Step 14: The first network analysis function updates or terminates the training process based on the modified subscription requirements.

[0129] For example, based on the modified precision and / or accuracy and / or training rounds and / or convergence conditions of the first model, the first model is retrained with the second network analysis function.

[0130] In an embodiment of the present application, each of the N-1 second network analysis functions performs forward calculation based on its own sub-model to obtain first information, and sends the first information to the first network analysis function; the first network analysis function performs reverse calculation based on the sub-model of the first network analysis function and the received first information to obtain second information, and sends the corresponding second information to the second network analysis function; the first network analysis function also updates the parameters of the sub-model of the first network analysis function based on all the received first information, and the second network analysis function updates the parameters of its own sub-model based on the received second information, thereby realizing that the first network analysis function and the second network analysis function collaborate to jointly learn the first model, thereby improving the efficiency of model training.

[0131] The above introduces the interaction process of the joint learning system. The following introduces the implementation process of the first network analysis function and the second network analysis function for joint learning of the first model.

[0132] The present application embodiment provides a joint learning method, which is applied to a first network analysis function, and the first network analysis function is not limited to Server NWDAF. Figure 3 As shown, the method includes:

[0133] Step 301: Based on the computing power information of the first network analysis function and the computing power information of each second network analysis function participating in the joint learning, jointly learn the first model with all the second network analysis functions participating in the joint learning.

[0134] Here, the first network analysis function and all the second network analysis functions participating in the joint learning cooperate to perform joint learning on the first model, including the first network analysis function and all the second network analysis functions participating in the joint learning jointly sharing the calculation task of the first model.

[0135] The first network analysis function can decompose the computing tasks of the first model based on the computing power information of the first network analysis function and the computing power information of each second network analysis function participating in the joint learning, so as to determine the computing tasks that the first network analysis function is responsible for and the computing tasks that each second network analysis function participating in the joint learning is responsible for in the process of training the first model, and the computing amount of the computing tasks matches the computing power of the network analysis function.

[0136] The first network analysis function can also decompose the first model into N sub-models based on the computing power information of the first network analysis function and the computing power information of the N-1 second network analysis functions participating in the joint learning. One sub-model corresponds to one first network analysis function or one second network analysis function. Each network analysis function participating in the joint learning is responsible for the computing task of one sub-model. The number of layers and / or the amount of computation of the sub-model matches the computing power of the network analysis function. The number of layers of the sub-model is positively correlated with the amount of computation of the sub-model.

[0137] In order to facilitate the first network analysis function to evaluate the computing power of each network analysis function, in one embodiment, the computing power information includes one or more of the following:

[0138] Processor resource utilization;

[0139] Memory usage;

[0140] Video memory usage;

[0141] The computing power of the processor.

[0142] In order to improve the model training efficiency, in one embodiment, the computing power information based on the first network analysis function and the computing power information of each second network analysis function participating in the joint learning, and all the second network analysis functions participating in the joint learning, jointly learn the first model, including:

[0143] Based on the first information, determine the second information and update the parameters of the sub-model of the first network analysis function; wherein the first information is obtained by forward calculation of the second network analysis function based on the sub-model of the second network analysis function, and the second information is used to update the parameters of the sub-model of the second network analysis function, and the sub-model is obtained by disassembling the first model based on the computing power information;

[0144] Second information is sent to each second network analysis function.

[0145] Here, the second information is determined based on the first information, and the parameters of the sub-model of the first network analysis function are updated based on all the first information.

[0146] In order to improve the model training efficiency, in one embodiment, determining the second information based on the first information and updating the parameters of the sub-model of the first network analysis function include:

[0147] Perform reverse calculation based on the first information and the sub-model of the first network analysis function to obtain second information corresponding to the first information;

[0148] Parameters of the sub-model of the first network analysis function are updated based on all the first information.

[0149] Here, the first information can be understood as the forward calculation result. The second information can be understood as the reverse calculation result or the intermediate gradient obtained by reverse calculation; one second network analysis function corresponds to one second information, that is, one second information is obtained by reverse calculation based on the submodel of the first network analysis function and the first information sent by the second network analysis function. Please refer to the relevant description in step 9 for the implementation process of determining the second information based on the first information, which will not be repeated here.

[0150] In order to improve the model training efficiency, before determining the second information based on the first information and updating the parameters of the sub-model of the first network analysis function, it is necessary to first receive the first information. Based on this, in one embodiment, the method further includes:

[0151] The first information sent by each second network analysis function participating in the joint learning is received.

[0152] In order to eliminate the barrel effect and improve the model training efficiency, in one embodiment, the computational load of the sub-model of the second network analysis function matches the computing power of the second network analysis function; and / or

[0153] The computational effort of the sub-model of the first network analysis function matches the computing power of the first network analysis function.

[0154] Here, the computational complexity of the sub-model includes but is not limited to the number of layers and / or the number of parameters of the sub-model.

[0155] In order to fully utilize the computing resources of each network analysis function and eliminate the barrel effect, thereby improving the model training efficiency, in one embodiment, the method further includes:

[0156] Based on the computing power information of the first network analysis function and the computing power information of each second network analysis function participating in the joint learning, the first model is disassembled into N sub-models, where N is the total number of the first network analysis functions and all the second network analysis functions participating in the joint learning, and one sub-model corresponds to one first network analysis function or one second network analysis function;

[0157] A sub-model is sent to each second network analysis function participating in the joint learning; wherein the second network analysis function is used to perform forward calculation based on the sub-model to obtain first information, and the first network analysis function is used to perform reverse calculation based on the sub-model of the first network analysis function and the first information sent by the second network analysis function to obtain second information, and the second information is used to update the parameters of the sub-model.

[0158] Here, the implementation process of decomposing the first model into N sub-models, please refer to Figure 1 The relevant description of step 4 in the embodiment is not repeated here.

[0159] In one embodiment, the first model is decomposed into N sub-models based on the computing power information of the first network analysis function and the computing power information of each second network analysis function participating in the joint learning, including:

[0160] Based on the computing power information of the first network analysis function and the computing power information of each second network analysis function participating in the joint learning, dividing the first model into one or more independent layers and one or more shared layers;

[0161] Using all shared layers in the first model as submodels of the first network function;

[0162] Based on the computing power information of all second network analysis functions, the independent layers in the first model are disassembled or expanded to obtain a sub-model of each second network analysis function.

[0163] Here, for the implementation process of dividing the first model into one or more independent layers and one or more shared layers, and disassembling or scaling the independent layers in the first model, please refer to the relevant description above and will not be repeated here.

[0164] Before the first network analysis function and the second network analysis function jointly learn the first model, the first network analysis function needs to obtain the computing power information of the second network analysis function from the network storage function, so as to select an available second network analysis function to participate in the joint learning. Based on this, in one embodiment, the method further includes:

[0165] Sending a first request to the network storage function, the first request being used to discover a second network analysis function;

[0166] Receiving computing power information of one or more second network analysis functions sent by the network storage function;

[0167] Based on the computing power information of the second network analysis function sent by the network storage function, a second network analysis function participating in the joint learning is determined.

[0168] Here, the second network analysis function that can meet the computing requirements of the analysis service requested by the consumer can be used as the second network analysis function participating in the joint learning; the second network analysis function participating in the joint learning can also be determined from the second network analysis functions that can provide computing power services or computing services in order of computing power from high to low. The ability to provide computing power services or computing services can be understood as having available computing power resources, and computing power resources include but are not limited to one or more of the following: processor resources, memory resources, and video memory resources. The number of second network analysis functions participating in the joint learning is one or more.

[0169] Before the first network analysis function and the second network analysis function jointly learn the first model, the first network analysis function needs to register with the network storage function. Based on this, in one embodiment, the method further includes:

[0170] Send a second request to the network storage function; wherein the second request carries at least the computing power information of the first network analysis function, and is used to request registration of the first network analysis function.

[0171] Before the first network analysis function and the second network analysis function jointly learn the first model, the first network analysis function needs to determine the first model that matches the analysis service. Based on this, in one embodiment, the method further includes:

[0172] receiving a third request; the third request carrying a first identifier, for requesting to subscribe to an analysis result of an analysis service corresponding to the first identifier;

[0173] Determine the first model that matches the analysis service corresponding to the first identifier.

[0174] Here, the first model can be understood as a model that can provide the analysis service corresponding to the first identifier.

[0175] Correspondingly, the embodiment of the present application further provides a joint learning method, which is applied to a second network analysis function, and the second network analysis function includes but is not limited to Client NWDAF, such as Figure 4 As shown, the method includes:

[0176] Step 401: Based on the computing power information of the second network analysis function and the computing power information of the first network analysis function, jointly learn the first model with the first network analysis function.

[0177] Here, the first network analysis function assigns computing tasks or sub-models for the first model to the first network analysis function and all second network analysis functions participating in the joint learning based on the computing power information of the first network analysis function and the computing power information of all second network analysis functions participating in the joint learning; the second network analysis function performs joint learning on the first model in collaboration with the first network analysis function based on the computing tasks or sub-models for the first model assigned to the second network analysis functions by the first network analysis function.

[0178] In order to facilitate the first network analysis function to evaluate the computing power of each network analysis function, in one embodiment, the computing power information includes one or more of the following:

[0179] Processor resource utilization;

[0180] Memory usage;

[0181] Video memory usage;

[0182] The computing power of the processor.

[0183] In order to improve the model training efficiency, in one embodiment, the computing power information based on the second network analysis function and the computing power information of the first network analysis function, and the first network analysis function, jointly learning the first model, include:

[0184] Performing forward calculation based on the sub-model of the second network analysis function to obtain first information, and sending the first information to the first network analysis function, wherein the sub-model is disassembled from the first model based on the computing power information;

[0185] receiving second information sent by the first network analysis function, where the second information is obtained by reverse calculation by the first network analysis function based on the first information and a sub-model of the first network analysis function;

[0186] Parameters of the sub-model of the second network analysis function are updated based on the second information.

[0187] Here, the computational effort of the sub-model of the second network analysis function matches the computing power of the second network analysis function.

[0188] In order to realize the joint learning of the first model by the second network analysis function and the first network analysis function in collaboration, so as to improve the model training efficiency, in one embodiment, before the sub-model based on the second network analysis function performs forward calculation to obtain the first information, the method further includes:

[0189] Receive the sub-model sent by the first network analysis function.

[0190] Here, one second network analysis function corresponds to one sub-model.

[0191] In order to make full use of the computing resources of the first network function and the second network function, in one embodiment, the sub-model is obtained by decomposing the first model by the first network analysis function based on the computing power information of the first network analysis function and the computing power information of each second network analysis function participating in the joint learning, and one sub-model corresponds to one first network analysis function or one second network analysis function.

[0192] Before the second network analysis function and the first network analysis function jointly learn the first model, the second network analysis function needs to register with the network storage function so that the network storage function provides the discovered second network analysis function to the first network analysis function. Based on this, in one embodiment, the method further includes:

[0193] A fourth request is sent to the network storage function, where the fourth request carries at least computing power information of the second network analysis function and is used to request registration of the second network analysis function.

[0194] In order to implement the method of the first network analysis function side of the embodiment of the present application, the embodiment of the present application also provides a joint learning device, which is set on the first network analysis function, such as Figure 5 As shown, the device comprises:

[0195] The first training unit 501 is used to jointly learn the first model with all the second network analysis functions participating in the joint learning based on the computing power information of the first network analysis function and the computing power information of each second network analysis function participating in the joint learning.

[0196] In one embodiment, the first training unit 501 is specifically used to determine the second information and update the parameters of the sub-model of the first network analysis function based on the first information; and to send the second information to each second network analysis function; wherein the first information is obtained by the second network analysis function through forward calculation based on the sub-model of the second network analysis function, and the second information is used to update the parameters of the sub-model of the second network analysis function, and the sub-model is obtained by disassembling the first model based on the computing power information.

[0197] In one embodiment, the first training unit 501 is specifically used to perform reverse calculation based on the first information and the sub-model of the first network analysis function to obtain second information corresponding to the first information; and to update the parameters of the sub-model of the first network analysis function based on all the first information.

[0198] In one embodiment, the device further comprises:

[0199] The first receiving unit is used to receive first information sent by each second network analysis function participating in the joint learning.

[0200] In one embodiment, the computational effort of the sub-model of the second network analysis function matches the computing power of the second network analysis function; and / or

[0201] The computational effort of the sub-model of the first network analysis function matches the computing power of the first network analysis function.

[0202] In one embodiment, the device further comprises:

[0203] a splitting unit, configured to split the first model into N sub-models based on the computing power information of the first network analysis function and the computing power information of each second network analysis function participating in the joint learning, where N is the total number of the first network analysis function and all the second network analysis functions participating in the joint learning, and one sub-model corresponds to one first network analysis function or one second network analysis function;

[0204] A first sending unit is used to send a sub-model to each second network analysis function participating in the joint learning; wherein the second network analysis function is used to perform forward calculation based on the sub-model to obtain first information, and the first network analysis function is used to perform reverse calculation based on the sub-model of the first network analysis function and the first information sent by the second network analysis function to obtain second information, and the second information is used to update the parameters of the sub-model.

[0205] In one embodiment, the splitting unit is used to:

[0206] Based on the computing power information of the first network analysis function and the computing power information of each second network analysis function participating in the joint learning, dividing the first model into one or more independent layers and one or more shared layers;

[0207] Using all shared layers in the first model as submodels of the first network function;

[0208] Based on the computing power information of all second network analysis functions, the independent layers in the first model are disassembled or expanded to obtain a sub-model of each second network analysis function.

[0209] In one embodiment, the computing power information includes one or more of the following:

[0210] Processor resource utilization;

[0211] Memory usage;

[0212] Video memory usage;

[0213] The computing power of the processor.

[0214] In one embodiment, the device further comprises:

[0215] A second sending unit, configured to send a first request to the network storage function, wherein the first request is used to discover a second network analysis function;

[0216] A second receiving unit, configured to receive computing power information of one or more second network analysis functions sent by the network storage function;

[0217] The first determination unit is used to determine the second network analysis function participating in the joint learning based on the computing power information of the second network analysis function sent by the network storage function.

[0218] In one embodiment, the device further comprises:

[0219] The third sending unit is used to send a second request to the network storage function; wherein the second request carries at least the computing power information of the first network analysis function, and is used to request registration of the first network analysis function.

[0220] In one embodiment, the device further comprises:

[0221] A third receiving unit, configured to receive a third request; the third request carries a first identifier, and is configured to request to subscribe to an analysis result of an analysis service corresponding to the first identifier;

[0222] The second determining unit is used to determine the first model that matches the analysis service corresponding to the first identifier.

[0223] In actual application, the first training unit 501, the first receiving unit, the first sending unit, the second sending unit, the second receiving unit, the third sending unit, and the third receiving unit can be implemented by the processor in the joint learning device combined with the communication interface, and the first determination unit and the second determination unit can be implemented by the processor in the joint learning device.

[0224] In order to implement the method of the second network analysis function side of the embodiment of the present application, the embodiment of the present application also provides a joint learning device, which is set on the second network analysis function, such as Figure 6 As shown, the device comprises:

[0225] The second training unit 601 is used to jointly learn the first model with the first network analysis function based on the computing power information of the second network analysis function and the computing power information of the first network analysis function.

[0226] In one embodiment, the second training unit 601 is specifically configured to:

[0227] Performing forward calculation based on the sub-model of the second network analysis function to obtain first information, and sending the first information to the first network analysis function, wherein the sub-model is disassembled from the first model based on the computing power information;

[0228] receiving second information sent by the first network analysis function, where the second information is obtained by reverse calculation by the first network analysis function based on the first information and a sub-model of the first network analysis function;

[0229] Parameters of the sub-model of the second network analysis function are updated based on the second information.

[0230] In one embodiment, the device further comprises:

[0231] The fourth receiving unit is used to receive the sub-model sent by the first network analysis function.

[0232] In one embodiment, the sub-model is obtained by decomposing the first model by the first network analysis function based on the computing power information of the first network analysis function and the computing power information of each second network analysis function participating in the joint learning, and one sub-model corresponds to one first network analysis function or one second network analysis function.

[0233] In one embodiment, the computing power information includes one or more of the following:

[0234] Processor resource utilization;

[0235] Memory usage;

[0236] Video memory usage;

[0237] The computing power of the processor.

[0238] In one embodiment, the device further comprises:

[0239] The fourth sending unit is used to send a fourth request to the network storage function, where the fourth request carries at least the computing power information of the second network analysis function and is used to request registration of the second network analysis function.

[0240] In actual application, the second training unit 601, the fourth receiving unit and the fourth sending unit can be implemented by a processor in the joint learning device, the splitting unit and the determining unit.

[0241] It should be noted that: the joint learning device provided in the above embodiment only uses the division of the above program modules as an example when performing joint learning. In practical applications, the above processing can be assigned to different program modules as needed, that is, the internal structure of the device is divided into different program modules to complete all or part of the processing described above. In addition, the joint learning device provided in the above embodiment and the joint learning method embodiment belong to the same concept. The specific implementation process is detailed in the method embodiment and will not be repeated here.

[0242] Based on the hardware implementation of the above program modules, and in order to implement the method of the first network analysis function side of the embodiment of the present application, the embodiment of the present application also provides a first network analysis function, the first network analysis function includes ServerNWDAF, such as Figure 7 As shown, the first network analysis function 700 includes:

[0243] The first communication interface 701 is capable of exchanging information with other network nodes;

[0244] The first processor 702 is connected to the first communication interface 701 to implement information exchange with other network nodes, and is used to execute the method provided by one or more technical solutions of the first network analysis function side when running a computer program. The computer program is stored in the first memory 703.

[0245] Specifically, the first processor 702 is used to perform joint learning on the first model in collaboration with all second network analysis functions participating in the joint learning based on the computing power information of the first network analysis function and the computing power information of each second network analysis function participating in the joint learning.

[0246] The first processor 702 is specifically configured to determine the second information based on the first information and update the parameters of the sub-model of the first network analysis function; wherein the first information is obtained by forward calculation of the second network analysis function based on the sub-model of the second network analysis function, and the second information is used to update the parameters of the sub-model of the second network analysis function, and the sub-model is obtained by disassembling the first model based on the computing power information;

[0247] The first communication interface 701 is used to send second information to each second network analysis function.

[0248] In one embodiment, the first processor 702 is specifically used to perform reverse calculation based on the first information and the sub-model of the first network analysis function to obtain second information corresponding to the first information; and to update the parameters of the sub-model of the first network analysis function based on all the first information.

[0249] In one embodiment, the first communication interface 701 is further used to receive first information sent by each second network analysis function participating in the joint learning.

[0250] In one embodiment, the computational effort of the sub-model of the second network analysis function matches the computing power of the second network analysis function; and / or

[0251] The computational effort of the sub-model of the first network analysis function matches the computing power of the first network analysis function.

[0252] In one embodiment, the first processor 702 is further configured to decompose the first model into N sub-models based on the computing power information of the first network analysis function and the computing power information of each second network analysis function participating in the joint learning, where N is the total number of the first network analysis function and all second network analysis functions participating in the joint learning, and one sub-model corresponds to one first network analysis function or one second network analysis function;

[0253] The first communication interface 701 is also used to send a sub-model to each second network analysis function participating in the joint learning; wherein the second network analysis function is used to perform forward calculation based on the sub-model to obtain first information, and the first network analysis function is used to perform reverse calculation based on the sub-model of the first network analysis function and the first information sent by the second network analysis function to obtain second information, and the second information is used to update the parameters of the sub-model.

[0254] In one embodiment, the first processor 702 is further configured to:

[0255] Based on the computing power information of the first network analysis function and the computing power information of each second network analysis function participating in the joint learning, dividing the first model into one or more independent layers and one or more shared layers;

[0256] Using all shared layers in the first model as submodels of the first network function;

[0257] Based on the computing power information of all second network analysis functions, the independent layers in the first model are disassembled or expanded to obtain a sub-model of each second network analysis function.

[0258] In one embodiment, the computing power information includes one or more of the following:

[0259] Processor resource utilization;

[0260] Memory usage;

[0261] Video memory usage;

[0262] The computing power of the processor.

[0263] In one embodiment, the first communication interface 701 is further used to send a first request to the network storage function, the first request being used to discover the second network analysis function; and to receive computing power information of one or more second network analysis functions sent by the network storage function;

[0264] The first processor 702 is further configured to determine a second network analysis function that participates in the joint learning based on the computing power information of the second network analysis function sent by the network storage function.

[0265] In one embodiment, the first communication interface 701 is further used to send a second request to the network storage function; wherein the second request carries at least computing power information of the first network analysis function, and is used to request registration of the first network analysis function.

[0266] In one embodiment, the first communication interface 701 is further used to receive a third request; the third request carries a first identifier and is used to request to subscribe to an analysis result of an analysis service corresponding to the first identifier;

[0267] The first processor 702 is further configured to determine the first model that matches the analysis service corresponding to the first identifier.

[0268] It should be noted that the specific processing process of the first processor 702 and the first communication interface 701 can be understood by referring to the above method.

[0269] Of course, in actual application, the various components in the first network analysis function 700 are coupled together through the bus system 704. It can be understood that the bus system 704 is used to realize the connection and communication between these components. In addition to the data bus, the bus system 704 also includes a power bus, a control bus, and a status signal bus. However, for the sake of clarity, Figure 7 Various buses are labeled as bus system 704 .

[0270] The first memory 703 in the embodiment of the present application is used to store various types of data to support the operation of the first network analysis function 700. Examples of such data include: any computer program used to operate on the first network analysis function 700.

[0271] The method disclosed in the above embodiment of the present application can be applied to the first processor 702, or implemented by the first processor 702. The first processor 702 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by an integrated logic circuit of hardware or software instructions in the first processor 702. The above-mentioned first processor 702 may be a general-purpose processor, a digital signal processor (DSP, Digital Signal Processor), or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The first processor 702 can implement or execute the various methods, steps and logic block diagrams disclosed in the embodiments of the present application. A general-purpose processor may be a microprocessor or any conventional processor, etc. In combination with the steps of the method disclosed in the embodiment of the present application, it can be directly embodied as a hardware decoding processor to execute, or it can be executed by a combination of hardware and software modules in the decoding processor. The software module may be located in a storage medium, which is located in the first memory 703, and the first processor 702 reads the information in the first memory 703 and completes the steps of the above method in combination with its hardware.

[0272] In an exemplary embodiment, the first network analysis function 700 can be implemented by one or more application specific integrated circuits (ASIC), DSP, programmable logic device (PLD), complex programmable logic device (CPLD), field programmable gate array (FPGA), general processor, controller, microcontroller (MCU), microprocessor, or other electronic components to execute the aforementioned method.

[0273] Based on the hardware implementation of the above program module, and in order to implement the method of the second network analysis function side of the embodiment of the present application, the embodiment of the present application also provides a second network analysis function, and the second network analysis function includes ClientNWDAF. Figure 8 As shown, the second network analysis function 800 includes:

[0274] The second communication interface 801 is capable of exchanging information with other network nodes;

[0275] The second processor 802 is connected to the second communication interface 801 to implement information exchange with other network nodes, and is used to execute the method provided by one or more technical solutions of the second network analysis function side when running the computer program. The computer program is stored in the second memory 803.

[0276] Specifically, the second processor 802 is used to collaborate with the first network analysis function to jointly learn the first model based on the computing power information of the second network analysis function and the computing power information of the first network analysis function.

[0277] In one embodiment, the second processor 802 is specifically configured to perform forward calculation based on the sub-model of the second network analysis function to obtain the first information, and to update the parameters of the sub-model of the second network analysis function based on the second information;

[0278] The second communication interface 801 is used to send the first information to the first network analysis function, and to receive second information sent by the first network analysis function. The sub-model is obtained by disassembling the first model based on the computing power information; the second information is obtained by reverse calculation by the first network analysis function based on the first information and the sub-model of the first network analysis function.

[0279] In one embodiment, the second communication interface 801 is further used to receive the sub-model sent by the first network analysis function.

[0280] In one embodiment, the sub-model is obtained by decomposing the first model by the first network analysis function based on the computing power information of the first network analysis function and the computing power information of each second network analysis function participating in the joint learning, and one sub-model corresponds to one first network analysis function or one second network analysis function.

[0281] In one embodiment, the computing power information includes one or more of the following:

[0282] Processor resource utilization;

[0283] Memory usage;

[0284] Video memory usage;

[0285] The computing power of the processor.

[0286] In one embodiment, the second communication interface 801 is further used to send a fourth request to the network storage function, wherein the fourth request carries at least computing power information of the second network analysis function and is used to request registration of the second network analysis function.

[0287] It should be noted that the specific processing process of the second processor 802 and the second communication interface 801 can be understood by referring to the above method.

[0288] Of course, in actual application, the various components in the second network analysis function 800 are coupled together through the bus system 804. It can be understood that the bus system 804 is used to realize the connection and communication between these components. In addition to the data bus, the bus system 804 also includes a power bus, a control bus, and a status signal bus. However, for the sake of clarity, Figure 8 Various buses are labeled as bus system 804 .

[0289] The second memory 803 in the embodiment of the present application is used to store various types of data to support the operation of the second network analysis function 800. Examples of such data include: any computer program used to operate on the second network analysis function 800.

[0290] The method disclosed in the above embodiment of the present application can be applied to the second processor 802, or implemented by the second processor 802. The second processor 802 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the hardware integrated logic circuit or software instructions in the second processor 802. The above-mentioned second processor 802 may be a general-purpose processor, DSP, or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The second processor 802 can implement or execute the various methods, steps and logic block diagrams disclosed in the embodiments of the present application. A general-purpose processor may be a microprocessor or any conventional processor, etc. In combination with the steps of the method disclosed in the embodiment of the present application, it can be directly embodied as a hardware decoding processor to execute, or it can be executed by a combination of hardware and software modules in the decoding processor. The software module can be located in a storage medium, which is located in the second memory 803. The second processor 802 reads the information in the second memory 803 and completes the steps of the above method in combination with its hardware.

[0291] In an exemplary embodiment, the second network analysis function 800 may be implemented by one or more ASICs, DSPs, PLDs, CPLDs, FPGAs, general purpose processors, controllers, MCUs, Microprocessors, or other electronic components to perform the aforementioned method.

[0292] It can be understood that the memory (first memory 703 and second memory 803) of the embodiment of the present application can be a volatile memory or a non-volatile memory, and can also include both volatile and non-volatile memories. Among them, the non-volatile memory can 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), a magnetic random access memory (FRAM), a flash memory, a magnetic surface memory, an optical disc, or a compact disc read-only memory (CD-ROM); the magnetic surface memory can be a disk memory or a tape memory. The volatile memory can be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of RAM are available, such as static random access memory (SRAM), synchronous static random access memory (SSRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct RAM bus random access memory (DRRAM).The memories described in the embodiments of the present application are intended to include, but are not limited to, these and any other suitable types of memories.

[0293] In an exemplary embodiment, the embodiment of the present application further provides a storage medium, namely a computer storage medium, specifically a computer-readable storage medium, for example, including a first memory 703 storing a computer program, and the computer program can be executed by the first processor 702 of the first network analysis function 700 to complete the steps of the method on the first network analysis function side. For another example, including a second memory 803 storing a computer program, the computer program can be executed by the second processor 802 of the second network analysis function 800 to complete the steps of the method on the second network analysis function side. The computer-readable storage medium can be a memory such as FRAM, ROM, PROM, EPROM, EEPROM, Flash Memory, magnetic surface memory, optical disk, or CD-ROM.

[0294] Exemplarily, the embodiment of the present application further provides a computer program product, including a computer program, which can be executed by the first processor 702 of the first network analysis function 700 to complete the steps of the first network analysis function side method. The computer program can be executed by the second processor 802 of the second network analysis function 800 to complete the steps of the second network analysis function side method.

[0295] It should be noted that: "first", "second", etc. are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. Multi-items can be two or more items, and a plurality can be two or more. The term "and / or" in this article is only a description of 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. In addition, the term "one or more" in this article means any combination of at least two of any one or more of a plurality. For example, including one or more of A, B, and C can mean including any one or at least two or more elements selected from the set consisting of A, B, and C.

[0296] In addition, the technical solutions described in the embodiments of the present application can be combined arbitrarily without conflict.

[0297] The above description is only a preferred embodiment of the present application and is not intended to limit the protection scope of the present application.

Claims

1. A joint learning method, characterized in that: Applied to a first network analysis function, the method comprises: Based on the computing power information of the first network analysis function and the computing power information of each second network analysis function participating in the joint learning, the first model is jointly learned in collaboration with all the second network analysis functions participating in the joint learning.

2. The method according to claim 1, characterized in that The method of performing joint learning on the first model in collaboration with all the second network analysis functions participating in the joint learning based on the computing power information of the first network analysis function and the computing power information of each second network analysis function participating in the joint learning includes: Based on the first information, determine the second information and update the parameters of the sub-model of the first network analysis function; wherein the first information is obtained by forward calculation of the second network analysis function based on the sub-model of the second network analysis function, and the second information is used to update the parameters of the sub-model of the second network analysis function, and the sub-model is obtained by disassembling the first model based on the computing power information; Second information is sent to each second network analysis function.

3. The method according to claim 2, characterized in that The determining the second information based on the first information and updating the parameters of the sub-model of the first network analysis function include: Perform reverse calculation based on the first information and the sub-model of the first network analysis function to obtain second information corresponding to the first information; Parameters of the sub-model of the first network analysis function are updated based on all the first information.

4. The method according to claim 2, characterized in that: The method further comprises: The first information sent by each second network analysis function participating in the joint learning is received.

5. The method according to claim 2, characterized in that: The computational effort of the sub-model of the second network analysis function matches the computing power of the second network analysis function; and / or The computational effort of the sub-model of the first network analysis function matches the computing power of the first network analysis function.

6. The method according to any one of claims 1 to 5, characterized in that: The method further comprises: Based on the computing power information of the first network analysis function and the computing power information of each second network analysis function participating in the joint learning, the first model is disassembled into N sub-models, where N is the total number of the first network analysis functions and all the second network analysis functions participating in the joint learning, and one sub-model corresponds to one first network analysis function or one second network analysis function; A sub-model is sent to each second network analysis function participating in the joint learning; wherein the second network analysis function is used to perform forward calculation based on the sub-model to obtain first information, and the first network analysis function is used to perform reverse calculation based on the sub-model of the first network analysis function and the first information sent by the second network analysis function to obtain second information, and the second information is used to update the parameters of the sub-model.

7. The method according to claim 6, characterized in that The first model is decomposed into N sub-models based on the computing power information of the first network analysis function and the computing power information of each second network analysis function participating in the joint learning, including: Based on the computing power information of the first network analysis function and the computing power information of each second network analysis function participating in the joint learning, dividing the first model into one or more independent layers and one or more shared layers; Using all shared layers in the first model as submodels of the first network function; Based on the computing power information of all second network analysis functions, the independent layers in the first model are disassembled or expanded to obtain a sub-model of each second network analysis function.

8. The method according to any one of claims 1 to 5 and 7, characterized in that: The computing power information includes one or more of the following: Processor resource utilization; Memory usage; Video memory usage; The computing power of the processor.

9. The method according to any one of claims 1 to 5 and 7, characterized in that: The method further comprises: Sending a first request to the network storage function, the first request being used to discover a second network analysis function; Receiving computing power information of one or more second network analysis functions sent by the network storage function; Based on the computing power information of the second network analysis function sent by the network storage function, a second network analysis function participating in the joint learning is determined.

10. The method according to any one of claims 1 to 5 and 7, characterized in that: The method further comprises: Send a second request to the network storage function; wherein the second request carries at least the computing power information of the first network analysis function, and is used to request registration of the first network analysis function.

11. The method according to any one of claims 1 to 5 and 7, characterized in that: The method further comprises: receiving a third request; the third request carrying a first identifier, for requesting to subscribe to an analysis result of an analysis service corresponding to the first identifier; Determine the first model that matches the analysis service corresponding to the first identifier.

12. A joint learning method, characterized in that: Applied to the second network analysis function, the method comprises: Based on the computing power information of the second network analysis function and the computing power information of the first network analysis function, jointly learn the first model with the first network analysis function.

13. The method according to claim 12, characterized in that The step of jointly learning the first model based on the computing power information of the second network analysis function and the computing power information of the first network analysis function in collaboration with the first network analysis function includes: Performing forward calculation based on the sub-model of the second network analysis function to obtain first information, and sending the first information to the first network analysis function, wherein the sub-model is disassembled from the first model based on the computing power information; receiving second information sent by the first network analysis function, where the second information is obtained by reverse calculation by the first network analysis function based on the first information and a sub-model of the first network analysis function; Parameters of the sub-model of the second network analysis function are updated based on the second information.

14. The method according to claim 13, characterized in that Before the sub-model based on the second network analysis function performs forward calculation to obtain the first information, the method further includes: Receive the sub-model sent by the first network analysis function.

15. The method according to any one of claims 12 to 14, characterized in that The sub-model is obtained by decomposing the first model based on the computing power information of the first network analysis function and the computing power information of each second network analysis function participating in the joint learning. One sub-model corresponds to one first network analysis function or one second network analysis function.

16. The method according to any one of claims 12 to 14, characterized in that: The computing power information includes one or more of the following: Processor resource utilization; Memory usage; Video memory usage; The computing power of the processor.

17. The method according to any one of claims 12 to 14, characterized in that The method further comprises: A fourth request is sent to the network storage function, where the fourth request carries at least computing power information of the second network analysis function and is used to request registration of the second network analysis function.

18. A joint learning device, characterized in that: include: The first training unit is used to jointly learn the first model with all the second network analysis functions participating in the joint learning based on the computing power information of the first network analysis function and the computing power information of each second network analysis function participating in the joint learning.

19. A joint learning device, characterized in that: include: The second training unit is used to jointly learn the first model with the first network analysis function based on the computing power information of the second network analysis function and the computing power information of the first network analysis function.

20. A first network analysis function, characterized in that: include: A first processor and a first communication interface; wherein, The first processor is used to jointly learn the first model with all the second network analysis functions participating in the joint learning based on the computing power information of the first network analysis function and the computing power information of each second network analysis function participating in the joint learning.

21. A second network analysis function, characterized in that: include: A second processor and a second communication interface; wherein, The second processor is used to collaborate with the first network analysis function to jointly learn the first model based on the computing power information of the second network analysis function and the computing power information of the first network analysis function.

22. A network analysis function, characterized in that: comprising a processor and a memory for storing a computer program capable of being executed on the processor, Wherein, when the processor is used to run the computer program, it executes the steps of the method described in any one of claims 1 to 11, or executes the steps of the method described in any one of claims 12 to 17.

23. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method described in any one of claims 1 to 11 are implemented, or the steps of the method described in any one of claims 12 to 17 are implemented.

24. A computer program product comprising a computer program, characterized in that The computer program, when executed by a processor, performs the steps of the method according to any one of claims 1 to 17.

Images