Model training method, apparatus, and system

Abstract

Embodiments of the present application relate to the technical field of communications, and disclose a model training method, apparatus, and system, solving the problem of client-side privacy data being prone to leakage. The method comprises: a network data analysis network element correspondingly receives at least one first message from at least one client, the at least one first message comprising model output parameters of a teacher model trained on a corresponding client; on the basis of the model output parameters of the teacher model, the network data analysis network element trains a student model in the network data analysis network element to obtain a trained student model; and the network data analysis network element sends a second message to an artificial intelligence service control network element, the second message being used for indicating model parameters of the trained student model. The embodiments of the present application are applied to a model training process.

Description

Model training method, device and system

[0001] This application claims priority to the Chinese patent application filed with the State Intellectual Property Office on March 20, 2024, with application number "202410338861.4" and application name "Model Training Method, Device and System", the entire contents of which are incorporated by reference into this application. Technical Field

[0002] The embodiments of the present application relate to the field of communication technology, and in particular to a model training method, device, and system. Background Art

[0003] Federated learning is a distributed model training method in which a server instructs multiple clients to train based on their own models and data. Upon completion, the client sends the model information to the server, which aggregates the model information from each client to generate a global learning model. The server and clients iteratively perform this training process until the global learning model is complete. During federated learning, the client and server do not need to transmit the client's training data; only the trained model data needs to be transmitted, effectively protecting the client's data privacy.

[0004] However, by analyzing and calculating the model parameter information transmitted between the client and the server, some characteristics of the data may be inferred, leading to the indirect leakage of sensitive client information. Therefore, how to further improve the privacy and security of client data has become a technical problem that needs to be solved urgently. Summary of the Invention

[0005] The embodiments of the present application provide a model training method, device and system that can improve the privacy and security of client data.

[0006] To achieve the above objectives, this application provides the following technical solutions:

[0007] In the first aspect, a model training method is provided, which can be executed by a network data analysis network element, or by a component of the network data analysis network element, such as a processor, chip, or chip system of the network data analysis network element, or by a logic module or software that can implement all or part of the network data analysis network element. The following is an illustration of the method being executed by a network data analysis network element. The model training method includes: the network data analysis network element receives at least one first message from at least one client, and the at least one first message includes the model output parameters of the teacher model trained in the corresponding client; the network data analysis network element trains the student model in the network data analysis network element based on the model output parameters of the teacher model to obtain a trained student model; the network data analysis network element sends a second message to the artificial intelligence service control network element, and the second message is used to indicate the model parameters of the trained student model.

[0008] This application provides a model training method in which a network data analysis element receives the model output parameters of a teacher model from a client, and trains its own student model based on the model output parameters of the teacher model. The trained student model parameters are then sent to an artificial intelligence service control element. This method allows the network data analysis element and the client to train the model by transmitting model output parameters rather than a specific model. This avoids inferring data features from model parameters, thereby improving the privacy and security of client data.

[0009] In one possible implementation, the model output parameter is a probability value corresponding to the output result of the model and the data result.

[0010] Based on this, the network data analysis network element can train the student model according to the output results of the teacher model and the corresponding probability values ​​of the data results to achieve the goal of consistency between the output results and probability values ​​of the student model and the output results and probability values ​​of the teacher model.

[0011] In one possible implementation, the model output parameters include at least one of the following: the output result of the model's fully connected layer, the hard label of the model's output result, the soft label of the model's output result, or the model's hidden layer parameters.

[0012] Based on this, the network data analysis network element can train the student model according to the output results of the teacher model's fully connected layer, the hard label of the model output results, the soft label of the model output results, and at least one of the model's hidden layer parameters to achieve the goal of consistency between the output model parameters of the student model and the model output parameters of the teacher model.

[0013] In one possible implementation, the method specifically includes: the network data analysis network element trains the student model in the network data analysis network element based on the model output parameters of the teacher model, and determines whether the current training meets the preset conditions; wherein the preset conditions include: the model performance of the currently trained student model meets the requirements, or the number of iterations of the current training is greater than or equal to the preset number; if the current training does not meet the preset conditions, the network data analysis network element sends a third message to at least one client respectively; wherein the third message is used to indicate the updated model parameters, so that at least one client trains the teacher model based on the updated model parameters, and sends the model output parameters of the teacher model trained based on the updated model parameters to the network data analysis network element; the network data analysis network element iteratively executes the above training process until the current training meets the preset conditions; when the current training meets the preset conditions, the network data analysis network element determines that the currently trained student model is the trained model.

[0014] Based on this, the network data analysis network element and the client train the model in an iterative manner, which can make the performance of the final trained student model better.

[0015] In one possible implementation, the method specifically includes: the network data analysis network element iteratively trains the student model; in each iterative training, the network data analysis network element calculates the loss value between the model output parameters of the student model in the current iterative training and the output parameters of the teacher model based on the loss function; the network data analysis network element determines that the training is completed when the loss value is less than a preset threshold, and obtains the currently trained student model.

[0016] Based on this, the network data analysis network element trains the student model with the loss value between the model output parameters of the student model and the output parameters of the teacher model, which can improve the consistency between the model output parameters of the student model and the output parameters of the teacher model.

[0017] In one possible implementation, the method specifically includes: the network data analysis network element iteratively trains the student model based on a first data set preconfigured in the network data analysis network element and an acquired auxiliary data set; the first data set is a data set corresponding to a second data set preconfigured in at least one client, the second data set is used for model training of the teacher model in the corresponding client, and the auxiliary data set is used to assist the first data set in training the student model.

[0018] Based on this, the network data analysis network element trains the student model together according to the preconfigured first data set and the auxiliary data set, which can improve the accuracy of the student model training.

[0019] In one possible implementation, the method also includes: the network data analysis network element sends a first request message to the data collection coordination network element; the first request message is used to request an auxiliary data set; the network data analysis network element receives a first response message from the data collection coordination network element; the first response message includes the auxiliary data set.

[0020] Based on this, the network data analysis network element can obtain an auxiliary data set from the data collection coordination network element, so that the network data analysis network element can jointly train the student model according to the preconfigured first data set and the auxiliary data set.

[0021] In a possible implementation, the auxiliary data set in the first response message is the auxiliary data set requested by the data collection coordination network element from the data storage network element.

[0022] Based on this, the network data analysis network element can obtain the auxiliary data set from the data storage network element through the data collection coordination network element.

[0023] In one possible implementation, the method also includes: the network data analysis network element sends a fourth message to at least one client; wherein the fourth message is used to indicate initial training parameters, so that at least one client trains the corresponding teacher model based on the initial training parameters to obtain the model output parameters of the teacher model.

[0024] Based on this, the client can train the teacher model based on the initial training parameters, and then the client sends the model output parameters of the trained teacher model to the network data analysis network element.

[0025] In one possible implementation, the method also includes: the network data analysis network element receives a fifth message from the artificial intelligence service control network element, the fifth message is used to indicate the address of at least one client and the training task of model training; the network data analysis network element generates initial training parameters corresponding to the training task; the network data analysis network element generates a fourth message based on the initial training parameters; the network data analysis network element sends a fourth message to at least one client based on the address of at least one client.

[0026] Based on this, the network data analysis network element can determine at least one client based on the address of the at least one client and send initial training parameters to the at least one client.

[0027] In one possible implementation, the method also includes: the network data analysis network element sends a sixth message to the artificial intelligence service control network element, the sixth message is used to indicate the model training process parameters of the model training process, the model training process includes at least one client training teacher model training process, and the network data analysis network element training student model training process at least one; the model training process parameters include at least one of the following: model convergence rate, model accuracy improvement parameters, or resource utilization.

[0028] Based on this, the artificial intelligence service control network element can determine the training process of the network data analysis network element when performing model training based on the model training process parameters.

[0029] In one possible implementation, the method also includes: the network data analysis network element receives a seventh message from the artificial intelligence service control network element, and the seventh message is used to instruct the network data analysis network element to report the model training process parameters of the student model.

[0030] Based on this, the network data analysis network element can report the model training process parameters based on the instructions of the artificial intelligence service control network element.

[0031] In a second aspect, a model training method is provided. The method can be executed by a client, or by a component of the client, such as a processor, chip, or chip system of the client, or by a logic module or software that can implement all or part of the client. The following description takes the method executed by the client as an example. The model training method includes: the client trains a teacher model in the client and determines a trained teacher model; the client generates a first message based on the model output parameters of the trained teacher model; the first message includes the model output of the trained teacher model in the client; and the client sends the first message to a network data analysis network element.

[0032] In one possible implementation, the method specifically includes: the client receives a fourth message from the network data analysis network element; wherein the fourth message is used to indicate initial training parameters; and the client trains the teacher model in the client based on the initial training parameters.

[0033] In one possible implementation, the method specifically includes: the client receives a third message from the network data analysis network element; the third message is: when the network data analysis network element trains the student model in the network data analysis network element based on the model output parameters of the teacher model, it sends a message when it determines that the current training does not meet the preset conditions; the preset conditions include: the model performance of the student model currently trained by the network data analysis network element meets the requirements, or the number of iterations of the current training of the network data analysis network element is greater than or equal to the preset number; the client trains the teacher model in the client based on the updated model parameters.

[0034] In one possible implementation, the method also includes: the client sends a learning capability registration message to the artificial intelligence service control network element; the learning capability registration message is used to register the client's model learning capability; the client receives a learning capability registration response message from the artificial intelligence service control network element; the learning capability registration response message is used to indicate the model learning capability of the registered client.

[0035] In one possible implementation, the model learning capability of the client includes at least one of the following: an identifier of the client, a device type of the client, a computing capability of the client, a data processing capability of the client, or model information of a teacher model of the client.

[0036] On the third aspect, a model training method is provided, which can be executed by an artificial intelligence service control network element, or by a component of the artificial intelligence service control network element, such as a processor, chip, or chip system of the artificial intelligence service control network element, or can be implemented by a logic module or software that can implement all or part of the artificial intelligence service control network element. The following is an illustration of the method being executed by an artificial intelligence service control network element. The model training method includes: the artificial intelligence service control network element receives a second message from the network data analysis network element, and the second message is used to indicate the model parameters of the trained student model; the trained student model is obtained by training the student model in the network data analysis network element based on the model output parameters of the teacher model; the model output parameters of the teacher model are the model output parameters of the teacher model trained in at least one client.

[0037] In one possible implementation, the method further includes: the artificial intelligence service control network element sends a ninth message to the first device, where the ninth message is used to indicate the model parameters of the trained student model.

[0038] In one possible implementation, the method also includes: the artificial intelligence service control network element sends a fifth message to the network data analysis network element, and the fifth message is used to indicate the address of at least one client and the training task of the model training.

[0039] In one possible implementation, the method also includes: the artificial intelligence service control network element receives an eighth message from the first device, where the eighth message is used to indicate a training task for the training model; and the artificial intelligence service control network element generates a fifth message based on the eighth message.

[0040] In one possible implementation, the method also includes: the artificial intelligence service control network element receives a learning capability registration message from the target client; the learning capability registration message is used to register the model learning capability of the target client; the artificial intelligence service control network element sends a learning capability registration response message to the target client; the learning capability registration response message is used to indicate the model learning capability of the registered client.

[0041] In one possible implementation, the method also includes: the artificial intelligence service control network element sends a learning capability registration message to the unified data warehouse network element; the learning capability registration message is used to register the client's model learning capability; the artificial intelligence service control network element receives a learning capability registration response message from the unified data warehouse network element; the learning capability registration response message is used to indicate the model learning capability of the registered client.

[0042] In one possible implementation, the model learning capability of the client includes at least one of the following: an identifier of the client, a device type of the client, a computing capability of the client, a data processing capability of the client, or model information of a teacher model of the client.

[0043] In a fourth aspect, a communication device is provided for implementing the various methods described above. The communication device may be the network data analysis network element in the first aspect, or a device including the network data analysis network element, or a device included in the network data analysis network element, such as a chip. Alternatively, the communication device may be the client in the second aspect, or a device including the client, or a device included in the client, such as a chip. Alternatively, the communication device may be the artificial intelligence service control network element in the third aspect, or any implementation of the third aspect, or a device including the artificial intelligence service control network element, or a device included in the artificial intelligence service control network element, such as a chip. The communication device includes modules, units, or means corresponding to the implementation of the above methods, and the modules, units, or means may be implemented by hardware, software, or by executing corresponding software implementations through hardware. The hardware or software includes one or more modules or units corresponding to the above functions.

[0044] In some possible designs, the communication device may include a processing module and a transceiver module. The transceiver module, also referred to as a transceiver unit, is configured to implement the transmitting and / or receiving functions described in any of the above aspects and any possible implementations thereof. The transceiver module may be comprised of a transceiver circuit, a transceiver, a transceiver, or a communication interface. The processing module may be configured to implement the processing functions described in any of the above aspects and any possible implementations thereof.

[0045] In some possible designs, the transceiver module includes a sending module and a receiving module, which are respectively used to implement the sending and receiving functions in any of the above aspects and any possible implementation methods.

[0046] In a fifth aspect, a communication device is provided, comprising: at least one processor; the processor is configured to execute a computer program or instruction stored in a memory so that the communication device performs the method of any of the above aspects. The memory may be coupled to the processor, or may be independent of the processor. The communication device may be the network data analysis network element in the first aspect, or a device including the network data analysis network element, or a device included in the network data analysis network element, such as a chip. Alternatively, the communication device may be the client in the second aspect, or a device including the client, or a device included in the client, such as a chip. Alternatively, the communication device may be the artificial intelligence service control network element in the third aspect, or a device including the artificial intelligence service control network element, or a device included in the artificial intelligence service control network element, such as a chip. In some possible designs, the communication device includes a memory, which is configured to store necessary program instructions and data.

[0047] In a possible implementation, the processor includes a logic circuit and an input interface and / or an output interface, wherein the output interface is used to perform the sending action in the corresponding method, and the input interface is used to perform the receiving action in the corresponding method.

[0048] In one possible implementation, the communication device further includes a communication interface and a communication bus, and the processor, memory, and communication interface are connected via the communication bus. The communication interface is used to perform the sending and receiving actions in the corresponding method. The communication interface may also be referred to as a transceiver. Optionally, the communication interface includes a transmitter and a receiver. In this case, the transmitter is used to perform the sending action in the corresponding method, and the receiver is used to perform the receiving action in the corresponding method.

[0049] In some possible designs, the communication device may be a chip or a chip system. When the communication device is a chip system, it may be composed solely of a chip or may include a chip and other discrete components. When the communication device is a chip, the aforementioned sending action / function may be understood as output, and the aforementioned receiving action / function may be understood as input.

[0050] In a sixth aspect, a chip is provided, which includes a processor for implementing the functions involved in any of the above aspects or any of its implementation methods.

[0051] In some possible designs, the chip includes a memory for storing necessary program instructions and data.

[0052] In the seventh aspect, a computer-readable storage medium is provided, which stores a computer program or instruction. When the computer program or instruction is run on a communication device, the communication device can execute any of the above aspects or any of its implementation methods.

[0053] In an eighth aspect, a computer program product comprising instructions is provided, which, when executed on a communication device, enables the communication device to execute the method of any of the above aspects or any of its implementations.

[0054] In the ninth aspect, a communication system is provided, which includes the network data analysis network element of the first aspect, the client of the second aspect, and the artificial intelligence service control network element of the third aspect.

[0055] The technical effects brought about by any implementation method of the second to ninth aspects can refer to the technical effects brought about by the corresponding implementation method of the first aspect, and will not be repeated here.

[0056] It should be noted that various possible implementations of any of the above aspects can be combined under the premise that the solutions are not contradictory. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] FIG1 is a schematic diagram of the architecture of a communication system provided in an embodiment of the present application;

[0058] FIG2 is a schematic diagram of the hardware structure of a communication device provided in an embodiment of the present application;

[0059] FIG3 is a flow chart of a model training method provided in an embodiment of the present application;

[0060] FIG4 is a flow chart of another model training method provided in an embodiment of the present application;

[0061] FIG5 is a flow chart of another model training method provided in an embodiment of the present application;

[0062] FIG6 is a flow chart of another model training method provided in an embodiment of the present application;

[0063] FIG7 is a flow chart of another model training method provided in an embodiment of the present application;

[0064] FIG8 is a flow chart of another model training method provided in an embodiment of the present application;

[0065] FIG9 is a schematic structural diagram of a communication device provided in an embodiment of the present application;

[0066] FIG10 is a schematic diagram of the hardware structure of another communication device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0067] To facilitate understanding of the technical solutions provided by the embodiments of this application, a brief introduction to the relevant technologies of this application is first given. The brief introduction is as follows:

[0068] 1. Network data analysis network element

[0069] The network data analysis network element is used to collect data from various network functions (NFs), such as policy control network elements, session management network elements, user plane network elements, access management network elements, and application function network elements (through network capability exposure function network elements), and perform analysis and prediction. In the fifth generation (5G) communication system, the network data analysis network element can be a network data analysis function (NWDAF) network element. In future communications such as the sixth generation (6G) communication, the network data analysis network element can still be an NWDAF network element, or have other names, which is not limited in the embodiments of the present application.

[0070] 2. Federated Learning

[0071] Federated learning (FL) is a distributed model training method. Its learning process involves a server instructing multiple clients to train based on their own models and data. Upon completion, the client sends the model information to the server, which then aggregates the model information from each client to generate a global learning model. The server and clients iteratively execute this training process until the global learning model is complete. During federated learning, clients and servers do not need to transmit their own training data; only the trained model data needs to be transmitted, effectively protecting client data privacy.

[0072] The above is a brief introduction to the related technologies of this application.

[0073] In the current federated learning process, model training is performed by transmitting model parameters between multiple devices, which improves data privacy and security compared to training data between them. However, by analyzing and calculating the model parameter information transmitted between the client and server, some data characteristics may be inferred, indirectly leaking sensitive client information. Therefore, how to further improve the privacy and security of client data has become a pressing technical issue. Furthermore, federated learning requires multiple clients to train on a unified model architecture, which does not fully account for the differences in computing, data, and communication resources between clients, affecting the performance of the resulting global learning model. Furthermore, the amount of model information transmitted in federated learning is large, requiring significant transmission resources between the client and server.

[0074] In order to solve the above technical problems, the present application provides a model training method, in which a network data analysis network element receives the model output parameters of a teacher model from a client, and performs model training on its own student model based on the model output parameters of the teacher model, and then sends the model parameters of the trained student model to the artificial intelligence service control network element. Based on this, the network data analysis network element and the client perform model training by transmitting model output parameters instead of transmitting specific model parameters, which can avoid inferring data features through model parameters, thereby improving the privacy security of client data. At the same time, in the present application, the network data analysis network element performs model training on its own student model based on the model output parameters of the client teacher model, rather than aggregating based on model information; therefore, the embodiment of the present application only requires the client to output the model output that the network data analysis network element can use, without requiring the client to use a unified model architecture, thereby improving the performance of the global learning model. In addition, since the amount of data of the model output parameters is usually much smaller than the amount of data of the model parameters, the embodiment of the present application can also save the transmission resources required during the transmission process between the client and the server.

[0075] The following is a detailed description of the solution provided by the embodiment of the present application. Before introducing the embodiment of the present application, the following points are explained.

[0076] In the description of this application, unless otherwise specified, " / " indicates that the objects associated before and after are in an "or" relationship, for example, A / B can represent A or B; "and / or" in this application is merely a description of the association relationship of associated objects, indicating that three relationships may exist, for example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone, where A and B can be singular or plural.

[0077] In the description of this application, A sending a message to B can be understood as A sending a message to B through one or more network elements.

[0078] In the description of this application, unless otherwise specified, "plurality" means two or more than two. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b, and / or c can mean: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or plural.

[0079] In addition, to facilitate the clear description of the technical solutions of the embodiments of the present application, in the embodiments of the present application, the words "first" and "second" are used to distinguish between identical or similar items with substantially the same functions and effects. Those skilled in the art will understand that the words "first" and "second" do not limit the quantity or execution order, and the words "first" and "second" do not necessarily mean different.

[0080] In the embodiments of this application, words such as "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary" or "for example" in the embodiments of this application should not be construed as being preferred or advantageous over other embodiments or designs. Rather, the use of words such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner to facilitate understanding.

[0081] It will be understood that the “embodiment” mentioned throughout the specification means that the specific features, structures or characteristics related to the embodiment are included in at least one embodiment of the present application. Therefore, the various embodiments throughout the specification do not necessarily refer to the same embodiment. In addition, these specific features, structures or characteristics can be combined in one or more embodiments in any suitable manner. It will be understood that in the various embodiments of the present application, the size of the sequence number of each process does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present application.

[0082] It can be understood that in this application, "when" and "if" both mean that corresponding processing will be taken under certain objective circumstances, and do not limit the time, nor do they require any judgment action when implementing it, nor do they mean that there are other limitations.

[0083] It is understood that some optional features in the embodiments of the present application may, in certain scenarios, be implemented independently of other features, such as the solution on which they are currently based, to solve corresponding technical problems and achieve corresponding effects. They may also be combined with other features in certain scenarios as needed. Accordingly, the devices provided in the embodiments of the present application may also implement these features or functions accordingly, which will not be described in detail here.

[0084] In this application, unless otherwise specified, the same or similar parts between the various embodiments can refer to each other. In the various embodiments in this application, and the various implementation methods / implementation methods / implementation methods in each embodiment, if there is no special explanation and logical conflict, the terms and / or descriptions between different embodiments and the various implementation methods / implementation methods / implementation methods in each embodiment are consistent and can be referenced to each other. The technical features in different embodiments and the various implementation methods / implementation methods / implementation methods in each embodiment can be combined to form new embodiments, implementation methods, implementation methods, or implementation methods according to their inherent logical relationships. The implementation methods of this application described below do not constitute a limitation on the scope of protection of this application.

[0085] The technical solutions provided in the embodiments of the present application can be used in various communication systems, which may be a third generation partnership project (3GPP) communication system, for example, a fourth generation (4G) long term evolution (LTE) system, a 5G new radio (NR) system, a 6G communication system, a vehicle to everything (V2X) system, a system of hybrid networking of LTE and NR, or a device to device (D2D) system, a machine to machine (M2M) communication system, an Internet of Things (IoT), and other next generation communication systems. Alternatively, the communication system may also be a non-3GPP communication system, which is not limited by the embodiments of the present application.

[0086] As shown in Figure 1, a schematic diagram of the architecture of a communication system 10 provided in an embodiment of the present application is provided. The communication system 10 includes: a network data analysis network element 101, at least one client 102, and an artificial intelligence service control network element 103. The number of network data analysis network elements 101, clients 102, and artificial intelligence service control network elements 103 in Figure 1 is only an example, and more or less may be possible.

[0087] Among them, the artificial intelligence service control network element 103 is used to select the network data analysis network element 101 and at least one client 102, and instruct the network data analysis network element 101 to interact with the at least one client 102 to perform model training. The network data analysis network element 101 sends initial training parameters to the at least one client 102. At least one client 102 performs model training on the teacher model in its respective client according to the above initial training parameters, and sends the model parameters of the trained teacher model to the network data analysis network element 101. The network data analysis network element 101 performs model training on its own student model based on the model output parameters of the teacher model, and then sends the model parameters of the trained student model to the artificial intelligence service control network element 103.

[0088] It should be pointed out that in the 5G communication system, the artificial intelligence service control network element in the embodiment of the present application can be an artificial intelligence service control function (AISCF) network element in the 5G network. In future communications such as 6G communications, the artificial intelligence service control function can still be an AISCF network element, or have other names, and the embodiment of the present application does not limit this.

[0089] Optionally, the network data analysis network element in the embodiment of the present application is configured with an analysis identifier related to model training, so that the data analysis network element can analyze the relevant data of the model training. In addition, a client-side interface (client-side interface) and a server-side interface (server-side interface) can be added to the network data analysis network element. Among them, the client interface is an interface between the network data analysis network element and the client, which is used to exchange model information (such as model output parameters and other model information in the model training process) between the network data analysis network element and the client, and the server monitors the model training progress between the network data analysis network element and the client. The server-side interface is an interface between the network data analysis network element and the artificial intelligence service control network element, which is used to obtain and store model information in the model training process (such as model knowledge of the teacher model, model parameters of the student model, etc.), and can also be used by the network data analysis network element to obtain auxiliary data sets.

[0090] In the embodiments of the present application, the client can pre-register its own learning capabilities with the artificial intelligence service control network element, so that the artificial intelligence service control network element can determine the client's computing and data capabilities and the model information in the client based on the client's learning capabilities. In this way, before subsequent model training, the artificial intelligence service control network element can select the corresponding client for the current model training based on the client's computing and data capabilities and the model information in the client. In addition, a client interface can be added to the client; for understanding the client interface, please refer to the above content, and this application will not elaborate on it.

[0091] In some embodiments, as shown in Figure 1, a communication system 10 provided by an embodiment of the present application also includes at least one of the following: a data collection and coordination network element 104, a data storage network element 105, an application network element 106, a unified data warehouse network element 107, an access network device 108, and an access and mobility management network element 109.

[0092] Among them, the data storage network element 105 is used to store the auxiliary data set required for model training; the data collection and coordination network element 104 is used to obtain the auxiliary data set required for model training from the data storage network element 105, so that the artificial intelligence service control network element can obtain the auxiliary data set required for model training from the data collection and coordination network element.

[0093] The application network element 106 and / or the data storage network element 105 are used to instruct the network data analysis network element and the client to perform model training based on the service target through the artificial intelligence service control network element.

[0094] The unified data warehouse network element 107 is used for the client to register its own model learning capability so that the artificial intelligence service control network element can obtain the client's model learning capability from the unified data warehouse network element.

[0095] It should be noted that in the embodiments of the present application, the model training process can be triggered by a network element in the network (such as an application network element or a data storage network element), or by a third-party device, and this application does not limit this. In addition, a server-side interface can be added to the artificial intelligence service control network element; the understanding of the server-side interface can refer to the above content, and this application will not elaborate on this.

[0096] Optionally, the client involved in the present application can be a user equipment (UE) in a network (such as a 5G network or a public land mobile network (PLMN) evolved after 5G), an access terminal, a terminal unit, a user station, a terminal station, a mobile station, a mobile station, a remote station, a remote terminal, a user terminal terminal equipment, TE), a mobile device, a wireless communication device, a terminal agent, a tablet computer (pad), a handheld device with wireless communication function, a computing device or other processing device connected to a wireless modem, a vehicle-mounted device, a vehicle-mounted transceiver unit, a wearable device, or a terminal device. The access terminal can be a cellular phone, a cordless phone, a session initiation protocol (SIP) phone, a wireless local loop (WLL) station, a personal digital assistant (PDA), a handheld device with wireless communication function, a computing device or other processing device connected to a wireless modem, a vehicle-mounted device, a drone, a robot, a smart point of sale (POS) machine, a customer-premises equipment (CPE) or a wearable device, a virtual reality (VR) device. The terminal device may be a wireless terminal in industrial control, a wireless terminal in self-driving, a wireless terminal in remote medical, a wireless terminal in smart grid, a wireless terminal in transportation safety, a wireless terminal in smart city, or a wireless terminal in smart home. Alternatively, the terminal device may be a terminal with communication capabilities in the Internet of Things (IoT), such as a terminal in V2X (e.g., a vehicle-to-everything (V2X) device), a terminal in D2D communication, or a terminal in M2M communication. The terminal device may be mobile or fixed.

[0097] The embodiments of this application do not limit the form of the terminal device. The device used to implement the functions of the terminal device can be the terminal device; it can also be a device that supports the terminal device to implement the functions, such as a chip system. The device can be installed in the terminal device or used in conjunction with the terminal device. In the embodiments of this application, the chip system can be composed of a chip or include a chip and other discrete devices.

[0098] In one possible implementation, the network data analysis network element, client, and artificial intelligence service control network element in the embodiment of the present application can also be referred to as a communication device, which can be a general device or a dedicated device. The embodiment of the present application does not make specific limitations on this.

[0099] In one possible implementation, the relevant functions of the network data analysis network element, client or artificial intelligence service control network element in the embodiment of the present application can be implemented by one device, or can be implemented by multiple devices together, or can be implemented by one or more functional modules within a device, and the embodiment of the present application does not specifically limit this. It is understandable that the above functions can be network elements in hardware devices, software functions running on dedicated hardware, or a combination of hardware and software, or virtualization functions instantiated on a platform (for example, a cloud platform).

[0100] In specific implementation, the network data analysis network element, client, and artificial intelligence service control network element shown in Figure 1 can all adopt the composition structure shown in Figure 2, or include the components shown in Figure 2. Figure 2 is a schematic diagram of the hardware structure of a communication device 200 provided in an embodiment of the present application. The communication device 200 includes a processor 201, a communication line 202, a memory 203, and at least one communication interface (Figure 2 is only exemplary and is illustrated by including a communication interface 204 as an example).

[0101] The processor 201 may be a general-purpose central processing unit (CPU), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits for controlling the execution of the program of the present application.

[0102] The communication link 202 may include a pathway for transmitting information between the aforementioned components.

[0103] The communication interface 204 uses any transceiver or other device for communicating with other devices or communication networks, such as Ethernet, RAN, wireless local area networks (WLAN), etc.

[0104] The memory 203 may be a read-only memory (ROM) or other type of static storage device that can store static information and instructions, a random access memory (RAM) or other type of dynamic storage device that can store information and instructions, or an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disc storage, an optical disc storage (including a compact disc, laser disc, optical disc, digital versatile disc, Blu-ray disc, etc.), a magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto. The memory may exist independently and be connected to the processor via the communication line 202. The memory may also be integrated with the processor.

[0105] The memory 203 is used to store computer-executable instructions for executing the solution of the present application, and the execution is controlled by the processor 201. The processor 201 is used to execute the computer-executable instructions stored in the memory 203, thereby implementing the communication method provided in the following embodiments of the present application.

[0106] Optionally, the computer-executable instructions in the embodiments of the present application may also be referred to as application code, which is not specifically limited in the embodiments of the present application.

[0107] In a specific implementation, as an embodiment, the processor 201 may include one or more CPUs, such as CPU0 and CPU1 in FIG. 2 .

[0108] In a specific implementation, as an embodiment, the communication device 200 may include multiple processors, such as the processor 201 and the processor 207 in FIG2 . Each of these processors may be a single-core (single-CPU) processor or a multi-core (multi-CPU) processor. The processor herein may refer to one or more devices, circuits, and / or processing cores for processing data (e.g., computer program instructions).

[0109] In a specific implementation, as an embodiment, the communication device 200 may further include an output device 205 and an input device 206. The output device 205 communicates with the processor 201 and can display information in a variety of ways. For example, the output device 205 can be a liquid crystal display (LCD), a light emitting diode (LED) display device, a cathode ray tube (CRT) display device, or a projector. The input device 206 communicates with the processor 201 and can receive user input in a variety of ways. For example, the input device 206 can be a mouse, a keyboard, a touch screen device, or a sensor device.

[0110] The communication device 200 described above can be a general-purpose device or a dedicated device. In a specific implementation, the communication device 200 can be a desktop computer, a portable computer, a network server, a personal digital assistant (PDA), a mobile phone, a tablet computer, a wireless terminal, an embedded device, or a device with a structure similar to that shown in FIG2 . The embodiments of the present application do not limit the type of the communication device 200.

[0111] In addition, the composition structure shown in FIG2 does not constitute a limitation on the communication device. In addition to the components shown in FIG2, the communication device may include more or fewer components than shown, or combine certain components, or arrange the components differently.

[0112] In the embodiment of the present application, the chip system can be composed of chips, or can include chips and other discrete devices.

[0113] In addition, the actions and terms involved in the various embodiments of this application can refer to each other without limitation. The message names or parameter names in the messages exchanged between the various devices in the embodiments of this application are only examples, and other names can also be used in specific implementations without limitation.

[0114] The model training method provided in the embodiment of the present application is described below in conjunction with Figures 1 and 2.

[0115] It should be noted that in the following embodiments of the present application, the message names, parameter names, or information names between network elements are only examples. In other embodiments, they may also be other names. The model training method provided in this application does not make specific limitations on this.

[0116] It is understood that in the embodiments of the present application, each network element may perform some or all of the steps in the embodiments of the present application. These steps or operations are merely examples, and the embodiments of the present application may also perform other operations or variations of various operations. In addition, the steps may be performed in a different order than those presented in the embodiments of the present application, and it is possible that not all operations in the embodiments of the present application need to be performed.

[0117] Figure 3 is a flow chart of a model training method provided in an embodiment of the present application. In an embodiment of the present application, the client trains its own teacher model and sends the model output parameters of the trained teacher model to the network data analysis network element; the network data analysis network element trains the student model in the network data analysis network element based on the model output parameters of the teacher model to obtain the trained student model; the network data analysis network element sends the model parameters of the trained student model to the artificial intelligence service control network element. Below, in combination with the above-mentioned model training process, the functions and execution actions of each device in the communication system provided in an embodiment of the present application are introduced. As shown in Figure 3, the model training method includes the following steps:

[0118] Step 301: At least one client sends a first message to a network data analysis network element. Correspondingly, the network data analysis network element receives at least one first message from the at least one client.

[0119] Among them, at least one first message includes the model output parameters of the teacher model trained in the corresponding client.

[0120] In one possible implementation, the model output parameters in the embodiment of the present application are used to characterize the corresponding probability values ​​of the model output results and data results.

[0121] In one example, the model output parameter includes at least one of the following: an output result of a fully connected layer of the model, a hard label of the model output result, a soft label of the model output result, or a hidden layer parameter of the model. Optionally, the output result of the fully connected layer of the model is the data / value output by the model.

[0122] Taking at least one client including client 1 and client 2 as an example, client 1 performs model training on teacher model 1 in client 1 to obtain model output parameters of teacher model 1, and client 1 sends the model output parameters of teacher model 1 to the network data analysis network element. Client 2 performs model training on teacher model 2 in client 2 to obtain model output parameters of teacher model 2, and client 2 sends the model output parameters of teacher model 2 to the network data analysis network element. Accordingly, the network data analysis network element receives the model output parameters of teacher model 1 from client 1 and the model output parameters of teacher model 2 from client 2.

[0123] In one example, the first message is: LocalLearning_Notify.

[0124] Step 302: The network data analysis network element trains the student model in the network data analysis network element based on the model output parameters of the teacher model to obtain a trained student model.

[0125] In one possible implementation, the network data analysis network element can use federated learning, knowledge distillation, etc. to train the student model in the network data analysis network element based on the model output parameters of the teacher model. The embodiment of this application does not limit this.

[0126] In combination with the example in the above step 301, the network data analysis network element performs model training on the student model in the network data analysis network element based on the model output parameters of the teacher model 1 in the client 1 and the model output parameters of the teacher model 2 in the client 2 to obtain the trained student model.

[0127] Step 303: The network data analysis network element sends a second message to the artificial intelligence service control network element. Correspondingly, the artificial intelligence service control network element receives the second message from the network data analysis network element.

[0128] The second message is used to indicate the model parameters of the trained student model.

[0129] In one possible implementation, after the network data analysis network element completes training of the student model, it determines the model parameters of the student model and generates a second message based on the model parameters of the student model. Thereafter, the network data analysis network element sends the second message to the artificial intelligence service control network element, so that the artificial intelligence service control network element can process the data in the artificial intelligence service control network element based on the model parameters of the student model.

[0130] In one example, the second message is: Nnwdaf_AnalyticsSubscription_Notify.

[0131] This application provides a model training method in which a network data analysis element receives the model output parameters of a teacher model from a client, and trains its own student model based on the model output parameters of the teacher model. The trained student model parameters are then sent to an artificial intelligence service control element. This method allows the network data analysis element and the client to train the model by transmitting model output parameters rather than a specific model. This avoids inferring data features from model parameters, thereby improving the privacy and security of client data.

[0132] As an embodiment, in step 302, the process of the network data analysis network element training the student model in the network data analysis network element based on the model output parameters of the teacher model includes the following scenarios 1 and 2, namely: scenario 1, iterative training; scenario 2, single training. Scenario 1 and scenario 2 are described below.

[0133] Scenario 1: Iterative training

[0134] In conjunction with FIG3 , as shown in FIG4 , in scenario 1, the above step 302 can be specifically implemented by the following steps 401 to 406 , which are described in detail below:

[0135] Step 401: The network data analysis network element trains the student model in the network data analysis network element based on the model output parameters of the teacher model.

[0136] In one possible implementation, the network data analysis element iteratively trains the student model. During each training iteration, the network data analysis element calculates the loss between the output parameters of the student model and the output parameters of the teacher model in the current training iteration based on a loss function. The network data analysis element determines that training is complete when the loss value is less than a preset threshold, thereby obtaining the currently trained student model.

[0137] It should be pointed out that in the above step 302, the network data analysis network element can also train the student model based on this process, and this application does not limit this.

[0138] It can be understood that the process of training the student model in the network data analysis network element based on the model output parameters of the teacher model can also refer to other model training processes in related technologies, and this application does not limit this.

[0139] Step 402: The network data analysis network element determines whether the current training meets the preset conditions.

[0140] Among them, the preset conditions include: the model performance of the currently trained student model meets the requirements, or the number of iterations of the current training is greater than or equal to the preset number.

[0141] In one possible implementation, after each iteration is completed, the network data analysis network element determines whether the model performance of the currently trained student model meets the requirements. If so, it determines that the current training meets the preset conditions. If not, it determines that the current training does not meet the preset conditions, and continues with the next round of iterative training. In addition, after each iteration is completed, the network data analysis network element records the number of iterations plus 1, and determines whether the number of iterations is greater than the preset number. If the number of iterations is greater than the preset number, the network data analysis network element determines that the current training meets the preset conditions. When either the model performance of the student model or the number of iterations of the current training meets the preset conditions, the network data analysis network element determines that the current training meets the preset conditions, and further determines that the training of the student model is completed.

[0142] Step 403: If the current training does not meet the preset condition, the network data analysis network element sends a third message to at least one client. Correspondingly, at least one client receives the third message from the network data analysis network element.

[0143] Among them, the third message is used to indicate the updated model parameters, so that at least one client trains the teacher model based on the updated model parameters, and sends the model output parameters of the teacher model trained based on the updated model parameters to the network data analysis network element.

[0144] In one possible implementation, the network data analysis network element determines that the current training does not meet the preset conditions when the model performance of the student model currently being trained does not meet the requirements and the number of iterations of the current training is less than a preset number. The network data analysis network element generates a third message based on the model parameters of the student model obtained during the current training, and sends the third message to at least one client.

[0145] In one example, the third message is: LocalLearning_Subscribe.

[0146] Step 404: At least one client trains the teacher model based on the updated model parameters.

[0147] Step 405: At least one client sends the model output parameters of the teacher model trained based on the updated model parameters to the network data analysis network element. Correspondingly, the network data analysis network element receives the model output parameters of the teacher model from the at least one client.

[0148] In one possible implementation, in the above steps 404 and 405, at least one client trains the teacher model based on the updated model parameters, and generates an updated first message based on the model output parameters of the updated teacher model. At least one client sends the updated first message to the network data analysis network element, so that the network data analysis network element continues to train the student model according to the updated first message.

[0149] In other words, if the current training does not meet the preset conditions, the network data analysis network element and at least one client iteratively execute the above steps 401 to 405 until the current training meets the preset conditions, and then execute the following step 406. Thereafter, the network data analysis network element executes the following step 406.

[0150] Step 406: The network data analysis network element determines that the currently trained student model is a trained model.

[0151] In other words, the network data analysis network element determines that the currently trained student model is a trained model when the model performance of the currently trained student model meets the requirements or the number of iterations of the currently trained model is greater than or equal to the preset number. Thereafter, the network data analysis network element executes the above step 303.

[0152] Scenario 2: Single training

[0153] During the single training process in Scenario 2, the network data analysis element only needs to execute steps 401 and 406 to complete the single training, without having to determine the preset conditions and iteratively execute steps 401 to 405. The specific implementation of steps 401 and 406 can refer to Scenario 1 above and will not be repeated here.

[0154] The training process in step 302 is described in detail above.

[0155] In one possible implementation, as shown in FIG5 , the process of the network data analysis network element in the above S401 training the student model in the network data analysis network element based on the model output parameters of the teacher model can be specifically implemented through the following steps 501 .

[0156] Step 501: The network data analysis network element iteratively trains the student model based on a first data set preconfigured in the network data analysis network element and an acquired auxiliary data set.

[0157] Among them, the first data set is a data set corresponding to the second data set pre-configured in at least one client, the second data set is used for model training of the teacher model in the corresponding client, and the auxiliary data set is used to assist the first data set in training the student model.

[0158] In one possible implementation, the auxiliary dataset may be an auxiliary dataset obtained by the network data analysis network element from other network elements, or the auxiliary dataset may be an auxiliary dataset determined by the network data analysis network element after performing data analysis on local model knowledge and existing datasets. This application does not limit this.

[0159] As shown in FIG5 , before step 501, the network data analysis network element may collect auxiliary data sets through the following steps 502 and 503, which are described in detail below:

[0160] Step 502: The network data analysis network element sends a first request message to the data collection and coordination network element. Correspondingly, the data collection and coordination network element receives the first request message from the network data analysis network element.

[0161] The first request information is used to request the auxiliary data set.

[0162] In one example, the first request information is: Ndccf_DataManagement_Subscribe.

[0163] In a possible implementation, the first request information includes a target address and a correlation ID.

[0164] Optionally, the target address is the address of the artificial intelligence service control network element. The relevant identifier is used to indicate an identifier related to this model training, such as the request identifier, service identifier, analysis identifier, etc. recorded below. This application does not limit this. The explanation of the request identifier, service identifier, and analysis identifier is detailed below and will not be repeated here.

[0165] It should be pointed out that in the 5G communication system, the data collection coordination network element can be a data collection coordination function (DCCF) network element. In future communications such as 6G communications, the data collection coordination network element can still be a DCCF network element, or have other names. The embodiments of the present application do not limit this.

[0166] Step 503: The data collection and coordination network element sends a first response message to the network data analysis network element. Correspondingly, the network data analysis network element receives the first response message from the data collection and coordination network element.

[0167] The first response information includes the auxiliary data set.

[0168] In one example, the first response information is: Ndccf_DataManagement_Notify.

[0169] In a possible implementation, the auxiliary data set in the first response message is the auxiliary data set requested by the data collection coordination network element from the data storage network element. In this case, before the above step 503, the method further includes the following steps 504 and 505:

[0170] Step 504: The data collection and coordination network element sends a second request message to the data storage network element. Correspondingly, the data storage network element receives the second request message from the data collection and coordination network element.

[0171] The second request information is used to request the auxiliary data set.

[0172] In one example, the second request information is: Ndsf_DataExposure_Request.

[0173] It should be pointed out that in the 5G communication system, the data storage network element can be a data storage function (DSF) network element. In future communications such as 6G communications, the data storage network element can still be a DSF network element, or have other names. The embodiments of the present application do not limit this.

[0174] Step 505: The data storage network element sends a second response message to the data collection and coordination network element. Correspondingly, the data collection and coordination network element receives the second response message from the data storage network element.

[0175] The second response information includes the auxiliary data set.

[0176] In one example, the second response information is: Ndsf_DataExposure_Response.

[0177] Optionally, after the data collection coordination network element receives the second response information from the data storage network element, the data collection coordination network element generates the first response information according to the auxiliary data set in the second response message.

[0178] The above introduces the model training process involved in the embodiments of the present application.

[0179] In conjunction with FIG3 , as shown in FIG6 , in one possible implementation, the training process of the network data analysis network element and the client is model training based on instructions from the first device and the artificial intelligence service control network element in the network, which is described in detail below:

[0180] Step 601: The first device sends an eighth message to the artificial intelligence service control network element. Correspondingly, the artificial intelligence service control network element receives the eighth message from the first device.

[0181] In one possible implementation, the eighth message is used to indicate a training task for the training model.

[0182] In one example, the eighth message is: EnsembleDistillation_Subscribe.

[0183] Optionally, the eighth message is used to indicate the following contents: a requester ID and a service target.

[0184] In one example, the request identifier is the identifier of the requester of the current service target, such as the identifier of the first device mentioned above.

[0185] The service objectives include at least one of the following: service types, service performance, data availability and quality, or computing requirements.

[0186] The service type includes at least one of the following: classification, word generation, or prediction. The service performance includes at least one of the following: accuracy, precision, recall, F1 score, or mean square error.

[0187] In a possible implementation, the first device is an application network element or a data storage network element.

[0188] It should be pointed out that in the 5G communication system, the application network element can be an application function (AF) network element. In future communications such as 6G communications, the application network element can still be an AF network element, or have other names. The embodiments of the present application do not limit this.

[0189] Step 602: The artificial intelligence service control network element performs service analysis to determine the client and network data analysis network element corresponding to the service.

[0190] In one possible implementation, the artificial intelligence service control network element pre-stores the model learning capabilities of each client and the model learning capabilities of the network data analysis network element. After receiving the service target indicated by the eighth message, the artificial intelligence service control network element determines the client and network data analysis network element that can provide model training services for the service target based on the model learning capabilities of each client and the model learning capabilities of the network data analysis network element.

[0191] Step 603: The artificial intelligence service control network element sends a fifth message to the network data analysis network element. Correspondingly, the network data analysis network element receives the fifth message from the artificial intelligence service control network element.

[0192] The fifth message is used to indicate the address of at least one client and the training task of the model training.

[0193] In one possible implementation, the fifth message further includes at least one of the following: an analytics ID, a service ID of a service target, initial model training parameters, client addresses, or a target address.

[0194] The analysis identifier is used to indicate a model training process parameter in the network data analysis network element statistical model training process. The model training process parameter includes at least one of the following: model convergence rate, model accuracy improvement parameter, or resource utilization.

[0195] The initial model training parameters are used to indicate the training target of the model training corresponding to the service target, the model architecture of the network data analysis network element, and hyperparameters. Hyperparameters include but are not limited to at least one of the following: learning rate or number of iterations (epochs).

[0196] Optionally, the artificial intelligence service control network element generates a fifth message based on the eighth message.

[0197] In one example, the fifth message is Nnwdaf_AnalyticsSubscription_Subscribe.

[0198] Step 604: The network data analysis network element sends a fourth message to at least one client. Correspondingly, the client receives the fourth message from the network data analysis network element.

[0199] Among them, the fourth message is used to indicate the initial training parameters, so that at least one client trains the corresponding teacher model based on the initial training parameters to obtain the model output parameters of the teacher model.

[0200] Optionally, the fourth message further includes: a service identifier of the service target, so that the network data analysis network element determines the service target corresponding to the local model training based on the service identifier. As an example, the fourth message is: LocalLearning_Subscribe.

[0201] In one possible implementation, after the network data analysis network element receives the fifth message from the artificial intelligence service control network element, the network data analysis network element generates initial training parameters corresponding to the training task; the network data analysis network element generates a fourth message based on the initial training parameters; and the network data analysis network element sends the fourth message to at least one client based on the address of the at least one client.

[0202] It should be noted that after step 604, the network data analysis network element, the client, and the artificial intelligence service control network element execute the relevant model training process in the above steps 301 to 303.

[0203] In addition, as shown in FIG6 , after step 303 , the artificial intelligence service control network element reports the model parameters of the student model to the first device through the following step 605 .

[0204] Step 605: The artificial intelligence service control network element sends a ninth message to the first device.

[0205] The ninth message is used to indicate the model parameters of the trained student model.

[0206] As an example, the ninth message is: EnsembleDistillation_Notify.

[0207] In one possible implementation, the network data analysis network element also needs to report the model training process parameters to the artificial intelligence service control network element. Specifically, it includes:

[0208] Step 606: The network data analysis network element sends a sixth message to the artificial intelligence service control network element. Correspondingly, the artificial intelligence service control network element receives the sixth message from the network data analysis network element.

[0209] The sixth message is used to indicate model training process parameters of the model training process, which includes at least one of a training process in which a client trains a teacher model and a process in which a network data analysis network element trains a student model. It should be noted that the model training process parameters can be referred to the description in step 603 above and are not further described here.

[0210] It is understandable that the network data analysis network element can determine to report the model training process parameters to the artificial intelligence service control network element based on the analysis identifier in the fifth message in step 603. Alternatively, the network data analysis network element can actively report the model training process parameters to the artificial intelligence service control network element. Alternatively, the network data analysis network element can report the model training process parameters to the artificial intelligence service control network element based on the seventh message sent by the artificial intelligence service control network element. This application does not limit this, wherein the seventh message is used to instruct the network data analysis network element to report the model training process parameters of the student model, and the seventh message is a message independent of the fifth message.

[0211] In one possible implementation, before the artificial intelligence service control network element performs service analysis in step 602 and determines the client and network data analysis network element corresponding to the service, the client can pre-register its model training capability with the core network. The process in which the client can pre-register its model training capability with the core network includes the following cases 1 and 2, respectively: Case 1, the client registers learning capability with the artificial intelligence service control network element; Case 2, the client registers learning capability with the unified data warehouse network element. The following are detailed descriptions of each:

[0212] Case 1: The client registers learning capabilities with the AI ​​service control network element.

[0213] As shown in FIG7 , in case 1, the process of the client registering learning capabilities with the artificial intelligence service control network element includes the following steps 701 and 702:

[0214] Step 701: The client sends a learning capability registration request message to the artificial intelligence service control network element. Correspondingly, the artificial intelligence service control network element receives the learning capability registration request message from the client.

[0215] In one possible implementation, the learning capability registration request message includes at least one of the following: client identifier, client type, client computing capability, or client data processing capability. Optionally, the learning capability registration request message may also include client model information.

[0216] The client identifier is used to uniquely identify the client, and each client has a different client identifier. The client type includes but is not limited to at least one of the following: a car, a mobile phone, a robot, or a radio access network (RAN) device. The client computing capability includes but is not limited to at least one of the following: processing power, memory, and speed. The client data capability includes but is not limited to at least one of the following: data type or data size. The client model information includes but is not limited to at least one of the following: model architecture, model size, or model target. The model architecture includes but is not limited to at least one of the following: a convolutional neural network model (CNN), a recurrent neural network model (RNN), or a converter. The model size includes the number of parameters of the model. The model target includes but is not limited to at least one of the following: classification or analysis.

[0217] As an example, the learning capability registration request message is: learningcapability_Registration_Request.

[0218] After registration, if the client needs to update its learning capability, the client may also send a learning capability registration update message: learningcapability_Update_Request to the artificial intelligence service control network element.

[0219] Step 702: The AI ​​service control network element sends a learning capability registration response message to the client. Correspondingly, the client receives the learning capability registration response message from the AI ​​service control network element.

[0220] Among them, the learning capability registration response message is used to indicate that the artificial intelligence service control network element has successfully registered the client's learning capability.

[0221] As an example, the learning capability registration response message is: learningcapability_Registration_Response.

[0222] In addition, when the client sends a learning capability registration update message, the corresponding learning capability registration response message is: learningcapability_Update_Response.

[0223] Case 2: The client registers learning capabilities with the unified data warehouse network element.

[0224] As shown in FIG8 , in case 2, the process of registering the learning capability of the client with the unified data warehouse network element includes the following steps 801 to 804:

[0225] Step 801: The client sends a learning capability registration request message to the artificial intelligence service control network element. Correspondingly, the artificial intelligence service control network element receives the learning capability registration request message from the client.

[0226] The specific implementation of step 801 may refer to the above step 701 and will not be repeated here.

[0227] It should be pointed out that in the 5G communication system, the unified data repository network element can be a unified data repository (UDR) network element. In future communications such as 6G communications, the unified data repository network element can still be a UDR network element, or have other names. The embodiments of this application do not limit this.

[0228] Step 802: The artificial intelligence service control network element sends a learning capability registration request message to the unified data storage network element. Correspondingly, the unified data storage network element receives the learning capability registration request message from the artificial intelligence service control network element.

[0229] In one possible implementation, after the artificial intelligence service control network element receives the learning capability registration request message from the client, it transparently transmits the learning capability registration request message to the unified data warehouse network element.

[0230] Step 803: The unified data storage network element sends a learning capability registration response message to the artificial intelligence service control network element. Correspondingly, the artificial intelligence service control network element receives the learning capability registration response message from the unified data storage network element.

[0231] Step 804: The artificial intelligence service control network element sends a learning capability registration response message to the client. Correspondingly, the client receives the learning capability registration response message from the artificial intelligence service control network element.

[0232] In one possible implementation, after the artificial intelligence service control network element receives the learning capability registration request message from the unified data warehouse network element, it transparently transmits the learning capability registration request message to the client.

[0233] It should be noted that in case 2, since the client registers its learning capabilities with the unified data warehouse network element, when the artificial intelligence service control network element needs to obtain the client's learning capabilities, it needs to request the client's learning capabilities from the unified data warehouse network element.

[0234] The above mainly introduces the solutions provided by the embodiments of the present application from the perspective of interaction between network elements. Accordingly, the embodiments of the present application also provide a communication device, which is used to implement the various methods described above. The communication device can be the network data analysis network element in the above method embodiments, or a device including the above network data analysis network element, or a component that can be used for the network data analysis network element; or the communication device can be the client in the above method embodiments, or a device including the above client, or a component that can be used for the client; or the communication device can be the artificial intelligence service control network element in the above method embodiments, or a device including the above artificial intelligence service control network element, or a component that can be used for the artificial intelligence service control network element. It is understandable that in order to implement the above functions, the communication device includes the corresponding hardware structure and / or software modules for performing each function. It should be readily appreciated by those skilled in the art that, in combination with the various exemplary units and algorithm steps described in the embodiments disclosed herein, the present application can be implemented in the form of hardware or a combination of hardware and computer software. Whether a function is executed in hardware or in a manner where computer software drives hardware depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0235] In the embodiment of the present application, the communication device can be divided into functional modules according to the above method embodiment. For example, each functional module can be divided according to each function, or two or more functions can be integrated into one processing module. The above integrated modules can be implemented in the form of hardware or in the form of software functional modules. It should be understood that the division of modules in the embodiment of the present application is schematic and is only a logical functional division. In actual implementation, there may be other division methods.

[0236] 9 shows a schematic structural diagram of a communication device 90 , which includes a processing module 901 and a communication module 902 . Optionally, the communication device 90 further includes a storage module 903 . The storage module 903 is used to store program codes and data of the communication device 90 .

[0237] Taking the communication device 90 as an example, which can be the network data analysis network element in the above-mentioned method embodiment, or a device including the above-mentioned network data analysis network element, or a component that can be used for the network data analysis network element, then: the communication module 902 is used to receive at least one first message from at least one client, and the at least one first message includes the model output parameters of the teacher model trained in the corresponding client; the processing module 901 is used to train the student model in the network data analysis network element based on the model output parameters of the teacher model to obtain the trained student model; the communication module 902 is also used to send a second message to the artificial intelligence service control network element, and the second message is used to indicate the model parameters of the trained student model.

[0238] In a possible implementation, the model output parameter is a probability value corresponding to the output result of the model and the data result.

[0239] In one possible implementation, the model output parameters include at least one of the following: the output result of the model's fully connected layer, the hard label of the model's output result, the soft label of the model's output result, or the model's hidden layer parameters.

[0240] In one possible implementation, the processing module 901 is specifically used to: train the student model in the network data analysis network element based on the model output parameters of the teacher model, and determine whether the current training meets the preset conditions; wherein the preset conditions include: the model performance of the currently trained student model meets the requirements, or the number of iterations of the current training is greater than or equal to the preset number; if the current training does not meet the preset conditions, instruct the communication module 902 to send a third message to at least one client respectively; wherein the third message is used to indicate the updated model parameters, so that at least one client trains the teacher model based on the updated model parameters, and sends the model output parameters of the teacher model trained based on the updated model parameters to the network data analysis network element; iteratively execute the above training process until the current training meets the preset conditions; when the current training meets the preset conditions, determine that the currently trained student model is the trained model.

[0241] In one possible implementation, the processing module 901 is specifically used to: iteratively train the student model; in each iterative training, calculate the loss value between the model output parameters of the student model in the current iterative training and the output parameters of the teacher model based on the loss function; determine that the training is completed when the loss value is less than a preset threshold, and obtain the currently trained student model.

[0242] In one possible implementation, the processing module 901 is specifically used to: iteratively train the student model based on a first data set preconfigured in the network data analysis network element and an obtained auxiliary data set; the first data set is a data set corresponding to a second data set preconfigured in at least one client, the second data set is used for model training of the teacher model in the corresponding client, and the auxiliary data set is used to assist the first data set in training the student model.

[0243] In one possible implementation, the communication module 902 is further used to: send a first request message to the data collection coordination network element; the first request message is used to request an auxiliary data set; receive a first response message from the data collection coordination network element; the first response message includes the auxiliary data set.

[0244] In a possible implementation, the auxiliary data set in the first response message is the auxiliary data set requested by the data collection coordination network element from the data storage network element.

[0245] In one possible implementation, the communication module 902 is also used to send a fourth message to at least one client; wherein the fourth message is used to indicate initial training parameters, so that at least one client trains the corresponding teacher model based on the initial training parameters to obtain the model output parameters of the teacher model.

[0246] In one possible implementation, the communication module 902 is also used to receive a fifth message from the artificial intelligence service control network element, where the fifth message is used to indicate the address of at least one client and the training task of the model training; the processing module 901 is also used to generate initial training parameters corresponding to the training task; the processing module 901 is also used to generate a fourth message based on the initial training parameters; the communication module 902 is also used to send a fourth message to at least one client based on the address of at least one client.

[0247] In one possible implementation, the communication module 902 is also used to send a sixth message to the artificial intelligence service control network element, where the sixth message is used to indicate model training process parameters of the model training process. The model training process includes at least one training process of a client training a teacher model and at least one of the processes of a network data analysis network element training a student model; the model training process parameters include at least one of the following: model convergence rate, model accuracy improvement parameters, or resource utilization.

[0248] In one possible implementation, the communication module 902 is also used to receive a seventh message from the artificial intelligence service control network element, and the seventh message is used to instruct the network data analysis network element to report the model training process parameters of the student model.

[0249] Among them, all relevant contents of each step involved in the above method embodiment can be referred to the functional description of the corresponding functional module and will not be repeated here.

[0250] Taking the communication device 90 as an example, which can be the client in the above-mentioned method embodiment, or a device including the above-mentioned client, or a component that can be used for the client, then: the processing module 901 is used to train the teacher model in the client and determine the trained teacher model; the processing module 901 is also used to generate a first message based on the model output parameters of the trained teacher model; the first message includes the model output of the trained teacher model in the client; the communication module 902 is used to send the first message to the network data analysis network element.

[0251] In one possible implementation, the communication module 902 is further used to receive a fourth message from the network data analysis network element; wherein the fourth message is used to indicate initial training parameters; and the client trains the teacher model in the client based on the initial training parameters.

[0252] In one possible implementation, the communication module 902 is also used to receive a third message from the network data analysis network element; the third message is: when the network data analysis network element trains the student model in the network data analysis network element based on the model output parameters of the teacher model, it sends a message when it determines that the current training does not meet the preset conditions; the preset conditions include: the model performance of the student model currently trained by the network data analysis network element meets the requirements, or the number of iterations of the current training of the network data analysis network element is greater than or equal to the preset number; the processing module 901 is also used to train the teacher model in the client based on the updated model parameters.

[0253] In one possible implementation, the communication module 902 is also used to: send a learning capability registration message to the artificial intelligence service control network element; the learning capability registration message is used to register the client's model learning capability; receive a learning capability registration response message from the artificial intelligence service control network element; the learning capability registration response message is used to indicate the model learning capability of the registered client.

[0254] In one possible implementation, the model learning capability of the client includes at least one of the following: an identifier of the client, a device type of the client, a computing capability of the client, a data processing capability of the client, or model information of a teacher model of the client.

[0255] Among them, all relevant contents of each step involved in the above method embodiment can be referred to the functional description of the corresponding functional module and will not be repeated here.

[0256] Taking the communication device 90 as an example, which can be the artificial intelligence service control network element in the above-mentioned method embodiment, or a device including the above-mentioned artificial intelligence service control network element, or a component that can be used for the artificial intelligence service control network element, then: the communication module 902 is used to receive a second message from the network data analysis network element, and the second message is used to indicate the model parameters of the trained student model; the trained student model is obtained by training the student model in the network data analysis network element based on the model output parameters of the teacher model; the model output parameters of the teacher model are the model output parameters of the teacher model that has been trained in at least one client.

[0257] In a possible implementation, the communication module 902 is further configured to send a ninth message to the first device, where the ninth message is configured to indicate model parameters of the trained student model.

[0258] In one possible implementation, the communication module 902 is further used to send a fifth message to the network data analysis network element, where the fifth message is used to indicate the address of at least one client and a training task for model training.

[0259] In one possible implementation, the communication module 902 is further used to receive an eighth message from the first device, where the eighth message is used to indicate a training task for the training model; and the processing module 901 is used to generate a fifth message based on the eighth message.

[0260] In one possible implementation, the communication module 902 is also used to: receive a learning capability registration message from the target client; the learning capability registration message is used to register the model learning capability of the target client; and send a learning capability registration response message to the target client; the learning capability registration response message is used to indicate the model learning capability of the registered client.

[0261] In one possible implementation, the communication module 902 is also used to: send a learning capability registration message to the unified data warehouse network element; the learning capability registration message is used to register the client's model learning capability; receive a learning capability registration response message from the unified data warehouse network element; the learning capability registration response message is used to indicate the model learning capability of the registered client.

[0262] In one possible implementation, the model learning capability of the client includes at least one of the following: an identifier of the client, a device type of the client, a computing capability of the client, a data processing capability of the client, or model information of a teacher model of the client.

[0263] Among them, all relevant contents of each step involved in the above method embodiment can be referred to the functional description of the corresponding functional module and will not be repeated here.

[0264] Alternatively, the modules in FIG9 may also be referred to as units. For example, the processing module may be referred to as a processing unit, and the transceiver module may be referred to as a transceiver unit. In addition, in the embodiment shown in FIG9 , the names of the units may not be those shown in the figure. For example, the transceiver module may also be referred to as a communication module or a communication unit.

[0265] If the various units in Figure 9 are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for enabling a computer device (which can be a personal computer, a server, or an access network device, etc.) or a processor (processor) to execute all or part of the steps of the method described in each embodiment of the present application. The storage medium for storing computer software products includes: various media that can store program codes, such as a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0266] In the embodiment of the present application, the communication device 90 is presented in the form of various functional modules divided in an integrated manner. The "module" here can refer to a specific ASIC, circuit, processor and memory that executes one or more software or firmware programs, integrated logic circuit, and / or other devices that can provide the above functions. In a simple embodiment, those skilled in the art will appreciate that the communication device 90 can take the form of the communication device shown in Figure 10.

[0267] 10 , the communication device includes a processor 1001 and a transceiver 1002 , and optionally, further includes a memory 1003 connected to the processor 1001 .

[0268] Processor 1001 may be a general-purpose central processing unit (CPU), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits for controlling the execution of the program of the present application. Processor 1001 may also include multiple CPUs, and processor 1001 may be a single-core (single-CPU) processor or a multi-core (multi-CPU) processor. The processor here may refer to one or more devices, circuits, or processing cores for processing data (such as computer program instructions).

[0269] The processor 1001, memory 1003, and transceiver 1002 are connected via a bus. The transceiver 1002 is used to communicate with other communication devices. Optionally, the transceiver 1002 may include a transmitter and a receiver. The device used to implement the receiving function in the transceiver 1002 can be considered a receiver, and the receiver is used to perform the receiving steps in the embodiments of the present application. The device used to implement the transmitting function in the transceiver 1002 can be considered a transmitter, and the transmitter is used to perform the transmitting steps in the embodiments of the present application.

[0270] In a first possible implementation, referring to FIG10 , the communication device further includes a memory 1003. The memory 1003 may be a ROM or other type of static storage device that can store static information and instructions, a RAM or other type of dynamic storage device that can store information and instructions, or an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical disc, laser disc, optical disc, digital versatile disc, Blu-ray disc, etc.), a magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and can be accessed by a computer, and the present embodiment of the application does not impose any restrictions on this. The memory 1003 may exist independently or be integrated with the processor 1001. Among them, the memory 1003 may contain computer program code. The processor 1001 is used to execute the computer program code stored in the memory 1003, thereby implementing the method provided in the embodiment of the present application.

[0271] An embodiment of the present application also provides a computer-readable storage medium, comprising instructions, which, when executed on a computer, enables the computer to execute any of the above methods.

[0272] An embodiment of the present application also provides a computer program product comprising instructions, which, when executed on a computer, enables the computer to execute any of the above methods.

[0273] An embodiment of the present application further provides a chip, including: a processor and an interface, wherein the processor is coupled to a memory via the interface, and when the processor executes a computer program or instruction in the memory, any one of the methods provided in the above embodiments is executed.

[0274] An embodiment of the present application also provides a communication system, including: a network data analysis network element, a client, and an artificial intelligence service control network element in the above embodiment.

[0275] In the above embodiments, all or part of the embodiments can be implemented by software, hardware, firmware, or any combination thereof. When implemented using a software program, all or part of the embodiments can be implemented in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of the present application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via a wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) method. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that includes one or more media integrated therein. The available media may be magnetic media (eg, floppy disks, hard disks, magnetic tapes), optical media (eg, DVDs), or semiconductor media (eg, solid state disks (SSDs)).

[0276] Although the present application is described herein in conjunction with various embodiments, in the process of implementing the claimed application, those skilled in the art may understand and implement other variations of the disclosed embodiments by reviewing the drawings, the disclosure, and the appended claims. In the claims, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude multiple situations. A single processor or other unit may implement several functions listed in the claims. Certain measures are recorded in different dependent claims, but this does not mean that these measures cannot be combined to produce good results.

[0277] Although the present application has been described in conjunction with features and embodiments thereof, it is apparent that various modifications and combinations may be made thereto without departing from the spirit and scope of the present application. Accordingly, this specification and drawings are merely illustrative of the present application as defined by the appended claims and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, those skilled in the art may make various modifications and variations to the present application without departing from the spirit and scope of the present application. Thus, the present application is intended to include such modifications and variations as fall within the scope of the claims of the present application and their equivalents.

[0278] The above is merely an embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions within the technical scope disclosed in the present application shall be included in the scope of protection of the present application. Therefore, the scope of protection of the present application shall be based on the scope of protection of the claims.

Claims

1. A model training method, characterized in that: The method comprises: The network data analysis network element receives at least one first message from at least one client, wherein the at least one first message includes a model output parameter of a teacher model trained in the corresponding client; The network data analysis network element trains the student model in the network data analysis network element based on the model output parameters of the teacher model to obtain a trained student model; The network data analysis network element sends a second message to the artificial intelligence service control network element, where the second message is used to indicate the model parameters of the trained student model.

2. The method according to claim 1, characterized in that The model output parameters are probability values ​​corresponding to the output results of the model and the data results.

3. The method according to claim 1 or 2, characterized in that The model output parameter includes at least one of the following: the output result of the model's fully connected layer, the hard label of the model's output result, the soft label of the model's output result, or the model's hidden layer parameter.

4. The method according to any one of claims 1 to 3, characterized in that The network data analysis network element trains a student model in the network data analysis network element based on the model output parameters in the at least one first message to obtain a trained student model, including: The network data analysis network element trains the student model in the network data analysis network element based on the model output parameters of the teacher model, and determines whether the current training meets the preset conditions; wherein the preset conditions include: the model performance of the currently trained student model meets the requirements, or the number of iterations of the current training is greater than or equal to the preset number; If the current training does not meet the preset condition, the network data analysis network element sends a third message to each of the at least one client; wherein the third message is used to indicate the updated model parameters, so that the at least one client trains the teacher model based on the updated model parameters, and sends the model output parameters of the teacher model trained based on the updated model parameters to the network data analysis network element; The network data analysis network element iteratively performs the above training process until the current training meets the preset conditions; When the current training meets the preset conditions, the network data analysis network element determines that the student model of the current training is the trained model.

5. The method according to claim 4, characterized in that The network data analysis network element trains the student model in the network data analysis network element based on the model output parameters of the teacher model, including: The network data analysis network element iteratively trains the student model; In each iterative training, the network data analysis network element calculates the loss value between the model output parameters of the student model in the current iterative training and the output parameters of the teacher model based on the loss function; The network data analysis network element determines that the training is completed when the loss value is less than a preset threshold value, and obtains the currently trained student model.

6. The method according to any one of claims 1 to 5, characterized in that The network data analysis network element iteratively trains the student model, including: The network data analysis network element iteratively trains the student model based on the first data set preconfigured in the network data analysis network element and the obtained auxiliary data set; the first data set is a data set corresponding to the second data set preconfigured in the at least one client, the second data set is used for model training of the teacher model in the corresponding client, and the auxiliary data set is used to assist the first data set in training the student model.

7. The method according to claim 6, characterized in that The method further comprises: The network data analysis network element sends a first request message to the data collection coordination network element; the first request message is used to request the auxiliary data set; The network data analysis network element receives first response information from the data collection coordination network element; the first response information includes the auxiliary data set.

8. The method according to claim 7, characterized in that The auxiliary data set in the first response message is the auxiliary data set requested by the data collection coordination network element from the data storage network element.

9. The method according to any one of claims 1 to 8, characterized in that Before the network data analysis network element receives at least one first message from at least one client, the method further includes: The network data analysis network element sends a fourth message to the at least one client; wherein, the fourth message is used to indicate initial training parameters, so that the at least one client trains the corresponding teacher model based on the initial training parameters to obtain the model output parameters of the teacher model.

10. The method according to claim 9, characterized in that Before the network data analysis network element sends the fourth message to each of the clients, the method further includes: The network data analysis network element receives a fifth message from the artificial intelligence service control network element, where the fifth message is used to indicate the address of the at least one client and the training task of the model training; The network data analysis network element generates the initial training parameters corresponding to the training task; The network data analysis network element generates the fourth message based on the initial training parameters; The network data analysis network element sends the fourth message to the at least one client based on the address of the at least one client.

11. The method according to any one of claims 1 to 10, characterized in that The method further comprises: The network data analysis network element sends a sixth message to the artificial intelligence service control network element, and the sixth message is used to indicate the model training process parameters of the model training process, the model training process including the training process of the at least one client training the teacher model and at least one of the processes of the network data analysis network element training the student model; the model training process parameters include at least one of the following: model convergence rate, model accuracy improvement parameters, or resource utilization.

12. The method according to claim 11, characterized in that Before the network data analysis network element sends the sixth message to the artificial intelligence service control network element, the method further includes: The network data analysis network element receives a seventh message from the artificial intelligence service control network element, and the seventh message is used to instruct the network data analysis network element to report the model training process parameters of the student model.

13. A communication device, characterized in that: include: A functional unit for executing the method according to any one of claims 1 to 12; wherein the actions executed by the functional unit are implemented by hardware or the corresponding software is implemented by hardware.

14. A communication device, characterized in that: include: processor; The processor is connected to a memory, the memory is used to store computer-executable instructions, and the processor executes the computer-executable instructions stored in the memory, so that the communication device implements the method according to any one of claims 1 to 12.

15. A computer-readable storage medium, characterized in that The method comprises instructions, which, when executed on a computer, cause the computer to execute the method according to any one of claims 1 to 12.

16. A chip, characterized in that: The chip includes a processor; the processor is connected to a memory, the memory is used to store computer-executable instructions, and the processor executes the computer-executable instructions stored in the memory to enable the communication device to implement the method according to any one of claims 1 to 12.

17. A computer program product comprising instructions, characterized in that When the method is run on a communication device, the communication device is enabled to implement the method according to any one of claims 1 to 12.

Images