Communication method and device

Abstract

The communication method and device provided by the embodiment of the invention are used for improving the training performance. The method comprises the following steps: receiving first capability information from first management equipment, wherein the first capability information indicates training capability supported by the first management equipment and / or network element equipment; and sending a training request to the first management device according to the first capability information, wherein the training request is used for requesting to execute a first reasoning function by adopting a first training technology. According to the mode, the second management equipment can obtain the training technology supported by the first management equipment and / or the network element equipment, so that the second management equipment can select the appropriate training technology for the specific reasoning function according to the requirement to trigger the first management equipment and / or the network element equipment to execute training.

Description

Technical Field

[0001] This application relates to the field of communication technology, and in particular to a communication method and apparatus. Background Technology

[0002] Artificial intelligence / machine learning (AI / ML) technologies and related applications are being increasingly adopted by a wider range of industries. Currently, AI / ML technologies have introduced new training techniques, including federated learning, generative AI learning, reinforcement learning, and distributed learning. How to uniformly manage and control these training techniques has become a pressing issue. Summary of the Invention

[0003] This application provides a communication method and apparatus for improving training performance.

[0004] In a first aspect, this application provides a communication method, the execution subject of which may be a second management device, or a chip or circuit on the second management device side. Taking the second management device as an example, the method includes: receiving first capability information from a first management device, the first capability information indicating the training capabilities supported by the first management device and / or network element devices; and sending a training request to the first management device according to the first capability information, the training request being used to request the use of a first training technique to execute a first inference function.

[0005] This application proposes a unified management and control scheme for training technologies, enabling a second management device (e.g., a cross-domain management unit) to acquire training technologies supported by a first management device (e.g., a domain management entity) and / or network element devices. This allows the second management device (e.g., the cross-domain management unit) to select appropriate training technologies for specific inference functions as needed, triggering the first management device (e.g., the domain management entity) and / or network element devices to perform training. This helps reduce the training overhead of inference functions (or inference services) and improves training speed and accuracy.

[0006] In one possible design, the training technique includes at least one of the following: federated training, distributed training, reinforcement training, or generative AI.

[0007] In one possible design, the first capability information includes at least one of the following: the name of the training technology supported by the first management device and / or network element device; the role corresponding to each training technology supported by the first management device and / or network element device; the entity to which the role corresponding to each training technology supported by the first management device and / or network element device belongs; the inference type applicable to each training technology supported by the first management device and / or network element device; the inference function applicable to each training technology supported by the first management device and / or network element device; or the model capability applicable to each training technology supported by the first management device and / or network element device. This design, by refining the reported training capabilities, allows the second management device to select a more reasonable training technology, thereby further reducing the training overhead of the inference function (or inference service) and improving training speed and accuracy.

[0008] In one possible design, the first capability information includes the names of subclasses of training technologies supported by the first management device and / or network element device. This design, by refining the reported training capabilities, allows the second management device to select more appropriate training technologies, thereby further reducing the training overhead of the inference function (or inference service) and improving training speed and accuracy.

[0009] In one possible design, the first capability information includes at least one of the following: the name of at least one subclass of federated training: vertical federated training or horizontal federated training; the name of at least one subclass of distributed training: data distributed learning or model distributed learning; and the name of at least one subclass of generative AI: pre-training, fine-tuning training, incremental pre-training, cue-based fine-tuning, retrieval enhancement techniques, or generative adversarial network techniques.

[0010] In one possible design, the first capability information includes at least one of the following: at least one role in federated training: federated training server role or federated training agent role; at least one role in reinforcement training: reinforcement learning agent or reinforcement learning environment; at least one role in distributed training: distributed learning principal or distributed learning agent; at least one role in generative AI: generator or discriminator.

[0011] In one possible design, the inference type includes at least one of the following: Management Data Analysis (MDA) inference function, Access Network (RAN) intelligent inference function, or Network Data Analysis Function (NWDAF) inference function.

[0012] In one possible design, the inference function includes at least one of the following: at least one of the following inference functions of the MDA inference function: coverage analysis capability or mobility analysis capability; at least one of the following inference functions of the RAN intelligent inference function: random access channel (RACH) optimization, mobile load balancing (MLB) optimization, or energy saving (ES) optimization; at least one of the following inference functions of the NWDAF inference function: slice load analysis or user data congestion analysis.

[0013] In one possible design, the model capability includes at least one of the following: a one-sided model, a two-sided model, a network device model, or a user equipment model.

[0014] In one possible design, the training request includes at least one of the following: the name of the first training technique, the role corresponding to the first training technique, the entity to which the role corresponding to the first training technique belongs, and the model capabilities of the first training technique.

[0015] In one possible design, the method further includes sending a capability request to a first management device, the capability request being used to request training capabilities supported by the first management device and / or network element devices. This design allows the first management device to report training capabilities when needed (i.e., when requested by the second management device), improving the real-time performance of reporting training capabilities and reducing reporting overhead compared to other reporting methods such as periodic reporting.

[0016] In one possible design, the method further includes selecting a first training technique based on at least one of the following: computational resources, number of training data samples, training data dimensionality, training data sample overlap rate, data dimensionality, training data feature overlap rate, first inference function, fault tolerance requirements, or robustness requirements. This design can improve the rationality of selecting a training technique.

[0017] Secondly, this application provides a communication method, the execution subject of which may be a first management device or a chip or circuit on the first management device side. Taking the first management device as an example, the method includes: sending first capability information to a second management device, the first capability information indicating the training capabilities supported by the first management device and / or network element device; receiving a training request from the second management device, the training request being used to request the use of a first training technique to execute a first inference function.

[0018] This application proposes a unified management and control scheme for training technologies, enabling a second management device (e.g., a cross-domain management unit) to acquire training technologies supported by a first management device (e.g., a domain management entity) and / or network element devices. This allows the second management device (e.g., the cross-domain management unit) to select appropriate training technologies for specific inference functions as needed, triggering the first management device (e.g., the domain management entity) and / or network element devices to perform training. This helps reduce the training overhead of inference functions (or inference services) and improves training speed and accuracy.

[0019] In one possible design, the method further includes training the AI / ML model using a first training technique. This approach, by performing training according to the training technique selected by the second management device, can improve the rationality of the inference function.

[0020] In one possible design, the method further includes: triggering the network element device to train the AI / ML model using a first training technique based on a training request. This approach enables the network element device to perform training according to the training technique selected by the second management device, which can improve the rationality of the inference function.

[0021] In one possible design, the training technique includes at least one of the following: federated training, distributed training, reinforcement training, or generative AI.

[0022] In one possible design, the first capability information includes at least one of the following: the name of the training technology supported by the first management device and / or network element device; the role corresponding to each training technology supported by the first management device and / or network element device; the entity to which the role corresponding to each training technology supported by the first management device and / or network element device belongs; the inference type applicable to each training technology supported by the first management device and / or network element device; the inference function applicable to each training technology supported by the first management device and / or network element device; or the model capability applicable to each training technology supported by the first management device and / or network element device. This design, by refining the reported training capabilities, allows the second management device to select a more reasonable training technology, thereby further reducing the training overhead of the inference function (or inference service) and improving training speed and accuracy.

[0023] In one possible design, the first capability information includes the names of subclasses of training technologies supported by the first management device and / or network element device. This design, by refining the reported training capabilities, allows the second management device to select more appropriate training technologies, thereby further reducing the training overhead of the inference function (or inference service) and improving training speed and accuracy.

[0024] In one possible design, the first capability information includes at least one of the following: the name of at least one subclass of federated training: vertical federated training or horizontal federated training; the name of at least one subclass of distributed training: data distributed learning or model distributed learning; and the name of at least one subclass of generative AI: pre-training, fine-tuning training, incremental pre-training, cue-based fine-tuning, retrieval enhancement techniques, or generative adversarial network techniques.

[0025] In one possible design, the first capability information includes at least one of the following: at least one role in federated training: federated training server role or federated training agent role; at least one role in reinforcement training: reinforcement learning agent or reinforcement learning environment; at least one role in distributed training: distributed learning principal or distributed learning agent; at least one role in generative AI: generator or discriminator.

[0026] In one possible design, the inference type includes at least one of the following: Management Data Analysis (MDA) inference function, Access Network (RAN) intelligent inference function, or Network Data Analysis Function (NWDAF) inference function.

[0027] In one possible design, the inference function includes at least one of the following: at least one of the following inference functions of the MDA inference function: coverage analysis capability or mobility analysis capability; at least one of the following inference functions of the RAN intelligent inference function: random access channel (RACH) optimization, mobile load balancing (MLB) optimization, or energy saving (ES) optimization; at least one of the following inference functions of the NWDAF inference function: slice load analysis or user data congestion analysis.

[0028] In one possible design, the model capability includes at least one of the following: a one-sided model, a two-sided model, a network device model, or a user equipment model.

[0029] In one possible design, the training request includes at least one of the following: the name of the first training technique, the role corresponding to the first training technique, the entity to which the role corresponding to the first training technique belongs, and the model capabilities of the first training technique.

[0030] In one possible design, the method further includes: receiving a capability request from a second management device, the capability request being used to request training capabilities supported by the first management device and / or network element devices. This design allows the first management device to report training capabilities when needed (i.e., when the second management device requests them), which improves the real-time performance of reporting training capabilities and reduces reporting overhead compared to other reporting methods such as periodic reporting.

[0031] In one possible design, the method further includes: receiving second capability information from at least one network element device, wherein the second capability information of any network element device indicates the training capabilities supported by the network element device. This design enables the first management device to acquire the training capabilities of the network element devices.

[0032] In one possible design, the method further includes sending a capability request to at least one network element device, the capability request being used to request training capabilities supported by at least one network element device. This design allows network elements devices to report training capabilities when needed (i.e., when the first management device requests them), improving the real-time performance of reporting training capabilities and reducing reporting overhead compared to other reporting methods such as periodic reporting.

[0033] Thirdly, this application provides a communication method, the execution subject of which can be a network element device, or a chip or circuit on the network element device side. Taking a network element device as an example, the method includes: acquiring second capability information, the second capability information indicating the training capabilities supported by the network element device; and sending the second capability information to a first management device.

[0034] This application proposes a unified management and control scheme for training technologies, enabling a second management device (e.g., a cross-domain management unit) to acquire training technologies supported by network element devices. As a result, the second management device (e.g., the cross-domain management unit) can select appropriate training technologies for specific inference functions according to requirements to trigger network element devices to perform training, which helps to reduce the training overhead of inference functions (or inference services) and improve training speed and accuracy.

[0035] In one possible design, the method further includes receiving a capability request from a first management device, the capability request being used to request training capabilities supported by the network element device. This design allows the network element device to report training capabilities when needed (i.e., when the first management device requests them), improving the real-time performance of reporting training capabilities and reducing reporting overhead compared to other reporting methods such as periodic reporting.

[0036] In one possible design, the method further includes: receiving a training request, the training request being used to request the execution of a first inference function using a first training technique; and training an AI / ML model using the first training technique. This approach, by performing training according to the training technique selected by the second management device, can improve the rationality of the inference function.

[0037] In one possible design, the training technique includes at least one of the following: federated training, distributed training, reinforcement training, or generative AI.

[0038] In one possible design, the first capability information includes the names of subclasses of training technologies supported by the first management device and / or network element device. This design, by refining the reported training capabilities, allows the second management device to select more appropriate training technologies, thereby further reducing the training overhead of the inference function (or inference service) and improving training speed and accuracy.

[0039] In one possible design, the first capability information includes at least one of the following: the name of at least one subclass of federated training: vertical federated training or horizontal federated training; the name of at least one subclass of distributed training: data distributed learning or model distributed learning; and the name of at least one subclass of generative AI: pre-training, fine-tuning training, incremental pre-training, cue-based fine-tuning, retrieval enhancement techniques, or generative adversarial network techniques.

[0040] In one possible design, the first capability information includes at least one of the following: at least one role in federated training: federated training server role or federated training agent role; at least one role in reinforcement training: reinforcement learning agent or reinforcement learning environment; at least one role in distributed training: distributed learning principal or distributed learning agent; at least one role in generative AI: generator or discriminator.

[0041] In one possible design, the inference type includes at least one of the following: Management Data Analysis (MDA) inference function, Access Network (RAN) intelligent inference function, or Network Data Analysis Function (NWDAF) inference function.

[0042] In one possible design, the inference function includes at least one of the following: at least one of the following inference functions of the MDA inference function: coverage analysis capability or mobility analysis capability; at least one of the following inference functions of the RAN intelligent inference function: random access channel (RACH) optimization, mobile load balancing (MLB) optimization, or energy saving (ES) optimization; at least one of the following inference functions of the NWDAF inference function: slice load analysis or user data congestion analysis.

[0043] In one possible design, the model capability includes at least one of the following: a one-sided model, a two-sided model, a network device model, or a user equipment model.

[0044] In one possible design, the training request includes at least one of the following: the name of the first training technique, the role corresponding to the first training technique, the entity to which the role corresponding to the first training technique belongs, and the model capabilities of the first training technique.

[0045] Fourthly, this application provides a communication method, the execution subject of which can be a second management device, or a chip or circuit on the second management device side. Taking the second management device as an example, the method includes: acquiring requirement information for at least one training technique; and sending the requirement information for at least one training technique to a first management device.

[0046] This application proposes a unified management and control scheme for training technologies, which enables a second management device (e.g., a domain management unit) to configure at least one training technology requirement information for a first management device (e.g., a domain management entity) based on its needs (e.g., business needs, management needs, inference needs, etc.). This allows the first management device (e.g., the domain management entity) and / or network element devices to select and execute appropriate training technologies for specific inference functions based on the requirement information of at least one training technology, thereby reducing training overhead, increasing training speed, and improving the accuracy of inference functions.

[0047] In one possible design, at least one training technique includes at least one of the following: federated training, distributed training, reinforcement training, or generative AI.

[0048] In one possible design, the requirement information includes at least one of the following: threshold information, inference type, or inference function. By refining the requirement information, the above design allows the first management device to select a more appropriate training technique, thereby further reducing the training overhead of the inference function (or inference business) and improving training speed and accuracy.

[0049] In one possible design, the threshold information for distributed training includes at least one of the following: threshold information for distributed model learning or threshold information for distributed data learning, wherein the threshold information for distributed model learning includes a computational resource threshold, and the threshold information for distributed data learning includes at least one of the following: a threshold for the number of training data samples and a threshold for the dimension of training data; the threshold information for federated training includes at least one of the following: threshold information for vertical federated training or threshold information for horizontal federated training, wherein the threshold information for vertical federated training includes at least one of the following: a threshold for the overlap rate of training data samples, a threshold for the dimension of data, or a threshold for the number of training data samples, and the threshold information for horizontal federated training includes at least one of the following: a threshold for the dimension of data, a threshold for the overlap rate of training data features, or a threshold for the overlap rate of training data samples; the threshold information for generative AI includes at least one of the following: a threshold for computational resources or a threshold for the number of training data samples; and the threshold information for reinforcement training includes at least one of the following: fault tolerance requirements or robustness requirements.

[0050] In one possible design, the inference type includes at least one of the following: MDA inference function, RAN intelligent inference function, or NWDAF inference function.

[0051] In one possible design, the inference function includes at least one of the following: at least one of the following inference functions of the MDA inference function: coverage analysis capability or mobility analysis capability; at least one of the following capabilities of the RAN intelligent inference function: RACH optimization, MLB optimization or ES optimization; at least one of the following capabilities of the NWDAF inference function: slice load analysis or user data congestion analysis.

[0052] Fifthly, this application provides a communication method, the execution subject of which may be a first management device or a chip or circuit on the first management device side. Taking the first management device as an example, the method includes: receiving demand information for at least one training technique from a second management device; training an AI / ML model based on the demand information for at least one training technique; and / or triggering a network element device to train the AI / ML model based on the demand information for at least one training technique.

[0053] This application proposes a unified management and control scheme for training technologies, which enables a second management device (e.g., a domain management unit) to configure at least one training technology requirement information for a first management device (e.g., a domain management entity) based on its needs (e.g., business needs, management needs, inference needs, etc.). This allows the first management device (e.g., the domain management entity) and / or network element devices to select and execute appropriate training technologies for specific inference functions based on the requirement information of at least one training technology, thereby reducing training overhead, increasing training speed, and improving the accuracy of inference functions.

[0054] In one possible design, at least one training technique includes at least one of the following: federated training, distributed training, reinforcement training, or generative AI.

[0055] In one possible design, the requirement information includes at least one of the following: threshold information, inference type, or inference function. By refining the requirement information, the above design allows the first management device to select a more appropriate training technique, thereby further reducing the training overhead of the inference function (or inference business) and improving training speed and accuracy.

[0056] In one possible design, the threshold information for distributed training includes at least one of the following: threshold information for distributed model learning or threshold information for distributed data learning, wherein the threshold information for distributed model learning includes a computational resource threshold, and the threshold information for distributed data learning includes at least one of the following: a threshold for the number of training data samples and a threshold for the dimension of training data; the threshold information for federated training includes at least one of the following: threshold information for vertical federated training or threshold information for horizontal federated training, wherein the threshold information for vertical federated training includes at least one of the following: a threshold for the overlap rate of training data samples, a threshold for the dimension of data, or a threshold for the number of training data samples, and the threshold information for horizontal federated training includes at least one of the following: a threshold for the dimension of data, a threshold for the overlap rate of training data features, or a threshold for the overlap rate of training data samples; the threshold information for generative AI includes at least one of the following: a threshold for computational resources or a threshold for the number of training data samples; and the threshold information for reinforcement training includes at least one of the following: fault tolerance requirements or robustness requirements.

[0057] In one possible design, the inference type includes at least one of the following: MDA inference function, RAN intelligent inference function, or NWDAF inference function.

[0058] In one possible design, the inference function includes at least one of the following: at least one of the following inference functions of the MDA inference function: coverage analysis capability or mobility analysis capability; at least one of the following capabilities of the RAN intelligent inference function: RACH optimization, MLB optimization or ES optimization; at least one of the following capabilities of the NWDAF inference function: slice load analysis or user data congestion analysis.

[0059] In one possible design, training an AI / ML model based on requirements information includes: selecting a training technique based on at least one of the following: computational resources, number of training data samples, training data dimensionality, training data sample overlap rate, data dimensionality, training data feature overlap rate, first inference function, fault tolerance requirements, or robustness requirements; and training the AI / ML model based on the training technique. This design can improve the rationality of selecting a training technique.

[0060] Sixthly, this application provides a communication method, the execution subject of which can be a network element device, or a chip or circuit on the network element device side. Taking a network element device as an example, the method includes: receiving at least one training technology requirement information from a second management device.

[0061] This application proposes a unified management and control scheme for training technologies, which enables a second management device (e.g., a domain management unit) to configure at least one training technology requirement information for network element devices according to its needs (e.g., business needs, management needs, inference needs, etc.), which helps to reduce training overhead, improve training speed, and improve the accuracy of inference functions.

[0062] In one possible design, the method further includes training an AI / ML model based on the requirements information of at least one training technique. This design allows network element devices to select and execute appropriate training techniques for specific inference functions based on the requirements information of at least one training technique, which helps reduce training overhead, improve training speed, and enhance the accuracy of inference functions.

[0063] In one possible design, at least one training technique includes at least one of the following: federated training, distributed training, reinforcement training, or generative AI.

[0064] In one possible design, the requirement information includes at least one of the following: threshold information, inference type, or inference function. By refining the requirement information, the above design allows the first management device to select a more appropriate training technique, thereby further reducing the training overhead of the inference function (or inference business) and improving training speed and accuracy.

[0065] In one possible design, the threshold information for distributed training includes at least one of the following: threshold information for distributed model learning or threshold information for distributed data learning, wherein the threshold information for distributed model learning includes a computational resource threshold, and the threshold information for distributed data learning includes at least one of the following: a threshold for the number of training data samples and a threshold for the dimension of training data; the threshold information for federated training includes at least one of the following: threshold information for vertical federated training or threshold information for horizontal federated training, wherein the threshold information for vertical federated training includes at least one of the following: a threshold for the overlap rate of training data samples, a threshold for the dimension of data, or a threshold for the number of training data samples, and the threshold information for horizontal federated training includes at least one of the following: a threshold for the dimension of data, a threshold for the overlap rate of training data features, or a threshold for the overlap rate of training data samples; the threshold information for generative AI includes at least one of the following: a threshold for computational resources or a threshold for the number of training data samples; and the threshold information for reinforcement training includes at least one of the following: fault tolerance requirements or robustness requirements.

[0066] In one possible design, the inference type includes at least one of the following: MDA inference function, RAN intelligent inference function, or NWDAF inference function.

[0067] In one possible design, the inference function includes at least one of the following: at least one of the following inference functions of the MDA inference function: coverage analysis capability or mobility analysis capability; at least one of the following capabilities of the RAN intelligent inference function: RACH optimization, MLB optimization or ES optimization; at least one of the following capabilities of the NWDAF inference function: slice load analysis or user data congestion analysis.

[0068] In one possible design, training an AI / ML model based on requirements information includes: selecting a training technique based on at least one of the following: computational resources, number of training data samples, training data dimensionality, training data sample overlap rate, data dimensionality, training data feature overlap rate, first inference function, fault tolerance requirements, or robustness requirements; and training the AI / ML model based on the training technique. This design can improve the rationality of selecting a training technique.

[0069] Seventhly, this application also provides a communication device having any of the methods provided in the first or fourth aspect above. This communication device can be implemented in hardware or by hardware executing corresponding software. The hardware or software includes one or more units or modules corresponding to the functions described above.

[0070] In one possible implementation, the communication device includes a processor configured to support the communication device in performing corresponding functions of the second management device in the method described above. The communication device may also include a memory coupled to the processor, which stores necessary program instructions and data for the communication device. Optionally, the communication device further includes interface circuitry for supporting communication between the communication device and devices such as the first management device.

[0071] In one possible implementation, the communication device includes corresponding functional modules, each used to implement the steps in the above method. The functions can be implemented in hardware or by hardware executing corresponding software. The hardware or software includes one or more modules corresponding to the functions described above.

[0072] In one possible implementation, the communication device includes a processing unit and a communication unit, which can perform the corresponding functions in the above method examples, as described in the methods provided in the first or fourth aspects, and will not be repeated here.

[0073] Eighthly, this application also provides a communication device having any of the methods provided in the second or fifth aspect above. This communication device can be implemented in hardware or by hardware executing corresponding software. The hardware or software includes one or more units or modules corresponding to the functions described above.

[0074] In one possible implementation, the communication device includes a processor configured to support the communication device in performing corresponding functions of the first management device in the method described above. The communication device may also include a memory coupled to the processor, which stores necessary program instructions and data for the communication device. Optionally, the communication device further includes an interface circuit for supporting communication between the communication device and devices such as a second management device or network element devices.

[0075] In one possible implementation, the communication device includes corresponding functional modules, each used to implement the steps in the above method. The functions can be implemented in hardware or by hardware executing corresponding software. The hardware or software includes one or more modules corresponding to the functions described above.

[0076] In one possible implementation, the communication device includes a processing unit and a communication unit, which can perform the corresponding functions in the above method examples, as described in the methods provided in the second or fifth aspects, and will not be repeated here.

[0077] Ninthly, this application also provides a communication device having any of the methods provided in the third or sixth aspect above. The communication device can be implemented in hardware or by hardware executing corresponding software. The hardware or software includes one or more units or modules corresponding to the functions described above.

[0078] In one possible implementation, the communication device includes a processor configured to support the communication device in performing corresponding functions of the network element device in the methods described above. The communication device may also include a memory coupled to the processor, which stores necessary program instructions and data for the communication device. Optionally, the communication device further includes interface circuitry for supporting communication between the communication device and devices such as a first management device.

[0079] In one possible implementation, the communication device includes corresponding functional modules, each used to implement the steps in the above method. The functions can be implemented in hardware or by hardware executing corresponding software. The hardware or software includes one or more modules corresponding to the functions described above.

[0080] In one possible implementation, the communication device includes a processing unit and a communication unit, which can perform the corresponding functions in the above method examples, as described in the methods provided in the third or sixth aspects, and will not be repeated here.

[0081] In a tenth aspect, a communication device is provided, including a processor and an interface circuit. The interface circuit is configured to receive signals from other communication devices outside the communication device and transmit them to the processor, or to send signals from the processor to other communication devices outside the communication device. The processor is configured to implement the methods of the first or fourth aspect and any possible design via logic circuits or execution code instructions.

[0082] Eleventhly, a communication device is provided, including a processor and an interface circuit. The interface circuit is used to receive signals from other communication devices outside the communication device and transmit them to the processor, or to send signals from the processor to other communication devices outside the communication device. The processor is used to implement the methods of the aforementioned second or fifth aspect and any possible design through logic circuits or execution code instructions.

[0083] In a twelfth aspect, a communication device is provided, including a processor and an interface circuit. The interface circuit is configured to receive signals from other communication devices outside the communication device and transmit them to the processor, or to send signals from the processor to other communication devices outside the communication device. The processor is configured to implement the methods of the aforementioned third or sixth aspect and any possible design through logic circuits or execution code instructions.

[0084] In a thirteenth aspect, a computer-readable storage medium is provided that stores a computer program or instructions which, when executed by a processor, implement the methods of any one of the first to sixth aspects and any possible design of any one aspect.

[0085] In a fourteenth aspect, a chip system is provided, comprising a processor and potentially a memory, for implementing any of the first to sixth aspects and any possible design methods of any of the aspects. The chip system may be composed of chips or may include chips and other discrete devices.

[0086] In a fifteenth aspect, a communication system is provided, the system comprising the apparatus of the first aspect (such as a second management device), the apparatus of the second aspect (such as a first management device), and the apparatus of the third aspect (such as a network element device).

[0087] In a sixteenth aspect, a communication system is provided, the system comprising the apparatus of the fourth aspect (such as a second management device), the apparatus of the fifth aspect (such as a first management device), and the apparatus of the sixth aspect (such as a network element device).

[0088] The technical effects that can be achieved by the technical solutions of any of the seventh to sixteenth aspects mentioned above can be described with reference to the technical effects that can be achieved by the technical solutions of the first aspect mentioned above, and the repeated parts will not be repeated. Attached Figure Description

[0089] Figure 1 A schematic diagram of the architecture of a service system provided in this application;

[0090] Figure 2 A schematic diagram of the architecture of another service system provided for this application;

[0091] Figure 3 A flowchart illustrating a communication method provided in this application;

[0092] Figure 4 A flowchart illustrating another communication method provided in this application;

[0093] Figure 5 A schematic diagram of the structure of a communication device provided in this application;

[0094] Figure 6 A schematic diagram of another communication device provided in this application. Detailed Implementation

[0095] The embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0096] The embodiments of this application can be applied to various mobile communication systems, such as: New Radio (NR) systems, Long Term Evolution (LTE) systems, Advanced Long Term Evolution (LTE-A) systems, Future Communication systems, and other communication systems. Specifically, no limitations are imposed here. Exemplarily, the various embodiments in this application can be used in the NR network management architecture. The NR network management architecture may include management functions (MnFs). An MnF is a management entity defined by the 3rd Generation Partnership Project (3GPP), and its externally visible behavior and interfaces are defined as management services (MnSs). In a service-providing management architecture, an MnF acts as an MnS producer or MnS consumer. The management services produced by an MnF's MnS producer may have multiple MnS consumers. An MnF can consume multiple management services from one or more management service producers. Figure 1 As shown, the MnS provided by MnF can be used to provide services to other MnFs (such as MnF#A). In this case, MnF can act as an MnS producer, and MnF#A can act as an MnS consumer. Furthermore, MnF can also obtain MnS services provided by other MnFs (denoted as MnF#B, which may be the same as or different from MnF#A). Figure 1 The MnF shown can also be a consumer of the MnS service provided by MnF#B. In other words, the same MnF can be both a consumer of the MnS service and a producer of the MnS service.

[0097] like Figure 2 The diagram illustrates a service-oriented management architecture applicable to embodiments of the method described in this application. This service-oriented management architecture includes a business support system (BSS), a cross-domain management function (CD-MnF), a domain management function (Domain-MnF), and a net element (NE). It should be understood that the cross-domain management function can be nodes such as a network management system (NMS), an MnS Producer, and an MnS Consumer, while the domain management function can be nodes such as a wireless automation engine (MBB automation engine, MAE), an element management system (EMS), an MnS Producer, and an MnS Consumer.

[0098] In this application, the cross-domain management function unit is used to manage one or more domain management function units. A domain management function unit can be used to manage one or more network elements. The following is a brief introduction to each unit.

[0099] If the management service is a management service provided by the cross-domain management functional unit, then the cross-domain management functional unit is the management service producer, and the business support system is the management service consumer.

[0100] If the management service is a management service provided by a domain management function unit, then the domain management function unit is the management service producer, and the cross-domain management function unit is the management service consumer.

[0101] When the management service is a management service provided by the network element, the network element is the management service producer, and the domain management functional unit is the management service consumer.

[0102] A business support system (BSS) is oriented towards communication services, providing functions and management services such as billing, settlement, accounting, customer service, sales, network monitoring, communication service lifecycle management, and service intent translation. The BSS can be an operator's operating system or a vertical OT system.

[0103] Cross-domain management function units, also known as network management function units (NMFs), can be network management entities such as NMSs or network function management service consumers (NFMS_C). These NMFs provide one or more of the following management functions or services: network lifecycle management, network deployment, network fault management, network performance management, network configuration management, network assurance, network optimization, and translation of network intents from communication service providers (Intent-CSPs).

[0104] The network referred to in the aforementioned management functions or services may include one or more network elements or sub-networks, or it may be a network slice. That is to say, the network management function unit may be a network slice management function (NSMF), a management data analytical function (MDAF), a self-organization network function (SON Function), or an intent-driven management service (Intent Driven MnS).

[0105] In various embodiments of this application, a network element is an entity that provides network services. A network element may include core network elements, radio access network elements, or transport network elements, etc. For example, Figure 2 In the architecture shown, the domain management function unit may include a radio access network domain management function unit, a core network element domain management function unit, or a transport network domain management function unit. The radio access network domain management function unit can be used to manage radio access network elements, the core network element domain management function unit can be used to manage core network elements, and the transport network domain management function unit can be used to manage transport network elements.

[0106] For example, core network elements may include, but are not limited to, access and mobility management function (AMF) entities, session management function (SMF) entities, policy control function (PCF) entities, network data analysis function (NWDAF) entities, network repository function (NRF) entities, gateways, etc.

[0107] Wireless access network elements may include, but are not limited to: various base stations (e.g., next-generation node B (gNB), evolved Node B (eNB), central unit control panel (CUCP), central unit (CU), distributed unit (DU), central unit user panel (CUUP), etc. In this application, network function NF is also referred to as network element NE.

[0108] Optionally, in certain deployment scenarios, the cross-domain management function unit can also provide sub-network lifecycle management, sub-network deployment, sub-network fault management, sub-network performance management, sub-network configuration management, sub-network assurance, sub-network optimization functions, and translation of network intents from communication service consumers (Intent-CSC) for sub-network service producers or service consumers. Here, a sub-network consists of multiple smaller sub-networks, which can be network slice sub-networks.

[0109] Domain management function unit (NMF) can also be called network element management function unit (NFMS). For example, a domain management function unit can be a network element management entity such as MAE, EMS, or network function management service provider (NFMS_P).

[0110] The domain management function unit provides one or more of the following functions or management services: lifecycle management of subnetworks or network elements, deployment of subnetworks or network elements, fault management of subnetworks or network elements, performance management of subnetworks or network elements, assurance of subnetworks or network elements, optimization functions of subnetworks or network elements, and translation of intents from network operators (Intent-NOPs) of subnetworks or network elements. Here, a subnetwork includes one or more network elements. A subnetwork can also include other subnetworks, meaning one or more subnetworks can form a larger subnetwork.

[0111] It should be understood that in this application, "intent" can refer to the expectations of the intent producer (such as a network element) and the system in which the intent producer resides (such as a network or sub-network), and may include requirements, goals, or constraints. The translation of intent refers to the process of determining the strategy for the intent. For example, a strategy can be used to indicate conditions that do not meet the intent. For instance, when the intent is energy saving, strategy A could be: when energy consumption exceeds a first threshold, energy consumption is abnormal (i.e., not energy-efficient); strategy B could be: when energy consumption exceeds a second threshold, energy consumption is abnormal (i.e., not energy-efficient). It is understandable that even for the same intent, different strategies may determine different solutions that satisfy the intent.

[0112] Optionally, the subnet here can also be a network slice subnet. The domain management system can be a network slice subnet management function (NSSMF), a domain management data analytical function (domain MDAF), a domain self-organization network function (SON Function), a domain intent management function, etc.

[0113] The domain management functional units can be classified as follows:

[0114] Based on network type, functions can be categorized as follows: Radio Access Network (RAN) Domain Management Function (RAN Domain MnF), Core Network Domain Management Function (CN Domain MnF), and Transport Network Domain Management Function (TN Domain MnF). It's important to note that a Domain Management Function can also be a domain network management system, capable of managing one or more of the access network, core network, or transport network.

[0115] According to administrative regions, they can be divided into: regional management functional units of a certain region, such as regional management functional units of city A, regional management functional units of city B, etc.

[0116] Network elements are entities that provide network services, including core network elements and access network elements. Core network elements include: access and mobility management function (AMF), session management function (SMF), policy control function (PCF), network data analytical function (NWDAF), network repository function (NRF), and gateways. Access network elements include: base stations (such as gNB, eNB), central unit control plane (CUCP), central unit (CU), distribution unit (DU), and central unit user plane (CUUP).

[0117] Among them, network elements can provide one or more of the following management functions or services: network element lifecycle management, network element deployment, network element fault management, network element performance management, network element assurance, network element optimization functions, and network element intent translation, etc.

[0118] AI / ML technologies and related applications are being adopted by a wider range of industries, and AI / ML capabilities are also being used in various areas of 5GS, including intelligent optimization use cases for RAN base stations, such as mobility load balancing (MLB), mobility robustness optimization (MRO), and energy saving (ES); management data analytics services for network management, such as management data analytics (MDA); and network data analytics services for the core network, such as network data analytics function (NWDAF).

[0119] Currently, an AI / ML management workflow has been defined, including the training phase, testing phase, simulation phase, deployment phase, and inference phase.

[0120] The training phase involves training one or more ML models, including initial training and retraining. It also includes the validation of ML entities to evaluate their performance on both training and validation data.

[0121] During the testing phase, if the validation results do not meet expectations, such as unacceptable variance, the ML model associated with that ML entity needs to be retrained. The training phase is the initial stage of the AI / ML management workflow.

[0122] During the simulation phase, ML entities used for inference are run in a simulation environment. The purpose is to evaluate the inference performance of the ML entities in the simulation environment before applying them to the target network or system.

[0123] The deployment phase is the process of enabling trained ML entities to be used in the target AI / ML inference function.

[0124] The reasoning phase is the process of using ML entities to perform reasoning through AI / ML reasoning functions.

[0125] Currently, with the commencement of research in 3GPP Release 19, TR 28.858 defines new training techniques, such as federated learning, generative AI learning, reinforcement learning, and distributed learning. How operators can uniformly manage and control these training techniques remains an unresolved issue.

[0126] In view of this, embodiments of this application provide a communication method and apparatus. In this method, a cross-domain management function unit can obtain training techniques supported by a domain management function unit, thereby selecting a suitable training technique to trigger the domain management function unit to create a suitable AI / ML model. Alternatively, the cross-domain management function unit can send a training technique request to the domain management function unit, enabling the domain management function unit to select a suitable training technique and create a suitable AI / ML model. The method and apparatus are based on the same technical concept. Since the methods and apparatus solve problems based on similar principles, their implementations can be mutually referenced, and repeated details will not be elaborated further.

[0127] It should be noted that the term "AI / ML" in this application can be replaced with "AI", "ML", or "AIML", etc. AI / ML can be understood as AI and ML, or AI or ML, or AI and / or ML.

[0128] The following describes a communication method provided by an embodiment of this application. This embodiment is applicable to scenarios where both training and inference functions are deployed on a domain management function unit (MDU), such as MDA; it is also applicable to scenarios where the training function is deployed on a domain management function unit and the inference function is deployed on a base station, such as SON or RAN intelligence; it is also applicable to scenarios where both training and inference functions are deployed on a base station, such as SON or RAN intelligence; and it is also applicable to scenarios where training is performed on a cross-domain management function unit and the inference function is performed on a management function unit or a base station. Refer to the function management scenarios defined in 3GPP TS28.105V18.2.0 4a.2, which will not be elaborated upon here. In the above scenarios, the cross-domain management function unit can act as a service consumer of the domain management function unit, managing and controlling the functions within the domain management function unit.

[0129] In this embodiment, the first management device can be an AI / ML MnS producer, or a component (such as a chip or chip system) within an AI / ML MnS producer. The second management device can be an AI / ML MnS consumer, or a component (such as a chip or chip system) within an AI / ML MnS consumer. Network element devices can be network devices, core network elements, or other network elements, or components (such as chips or chip systems) within network elements. The first and second management devices can be deployed in different entities, or they can be deployed in the same entity, such as... Figure 1As shown. For ease of understanding of the embodiments of this application, the following text uses the example of the first management device and the second management device being deployed in different entities. It can be understood that in the embodiments of this application, the role of the management service (MnS) that provides AI / ML capabilities is referred to as the producer, and the role of the caller of the AI / ML capability management service is referred to as the consumer.

[0130] In one exemplary description, the first management device can be Figure 2 The domain management function unit in the middle, the second management device can be Figure 2 The cross-domain management function unit in the network element device can be... Figure 2 Network elements within the network. For descriptions of cross-domain management functional units, domain management functional units, and network elements, please refer to [reference needed]. Figure 2 The relevant content will not be repeated here.

[0131] See Figure 3 Here is an exemplary flowchart of a communication method provided in an embodiment of this application, the method comprising:

[0132] 301, The first management device sends first capability information to the second management device. Correspondingly, the second management device receives the first capability information from the first management device.

[0133] The first capability information indicates the training capabilities supported by the first management device and / or network element device. The aforementioned network element device may be a network element device managed by the first management device, and the number of such network element devices may be one or more, without specific limitation here.

[0134] Optionally, the first capability information may include at least one of the following: the name of the training technology supported by the first management device and / or network element device, the role corresponding to the training technology supported by the first management device and / or network element device, the entity to which the role corresponding to the training technology supported by the first management device and / or network element device belongs, the inference type applicable to the training technology supported by the first management device and / or network element device, the inference function applicable to the training technology supported by the first management device and / or network element device, or the model capability applicable to the training technology supported by the first management device and / or network element device.

[0135] The information mentioned above will be introduced below.

[0136] 1) The name of the training technology supported by the first management device and / or network element device.

[0137] For example, the first capability information may include the name of at least one of the following training techniques: federated learning (FL), distributed learning (DL), reinforcement learning (RL), or generative AI.

[0138] Furthermore, the first capability information may include the name of a subclass of supported training techniques. For example, taking federated training as an example, the first capability information may include the name of at least one of the following subclasses of federated training: Vertical Federated Training (VFL) or Horizontal Federated Training (HFL). Taking distributed training as an example, the first capability information may include the name of at least one of the following subclasses of distributed training: Data-distributed learning or Model-distributed learning. Taking generative AI as an example, the first capability information may include the name of at least one of the following subclasses of generative AI: Pre-Training, Fine-Tuning, Incremental Pre-training, Prompt-tuning, Retrievable Augmented Generation (RAG), or Generative Adversarial Networks (GANs).

[0139] 2) The roles corresponding to the training technologies supported by the first management device and / or network element device.

[0140] For example, in federated training, the roles corresponding to the training technology include the federated training server (FLserver) and the federated training agent (FL client). In reinforcement training, the roles include the reinforcement learning agent (RL agent) and the reinforcement learning environment. In distributed training, the roles include the distributed learning delegator (DL delegator) and the distributed learning agent (DL client). In generative AI's GAN technology, the roles include the generator and the discriminator.

[0141] For example, the first capability information may include the name of the federated training and the role corresponding to the federated learning.

[0142] The first ability information may include the name of the reinforcement training and the role corresponding to the reinforcement training.

[0143] The first capability information may include the name of the distributed training and the role corresponding to the distributed training.

[0144] The first capability information can include the name of the generative AI and the role it corresponds to.

[0145] 3) The entities to which the roles of the training technologies supported by the first management device and / or network element devices are located.

[0146] For example, the entity can be a network element (such as a base station, core network element, etc.) or a domain management function unit; this application does not limit this. It can be identified by the identifier of the network element (such as a network device identifier, cell identifier, core network element identifier, etc.) or the identifier of the domain management function unit (such as a vendor identifier, etc.).

[0147] 4) The inference types applicable to the training techniques supported by the first management device and / or network element devices.

[0148] For example, the inference type includes at least one of the following: inference function supported by the domain management function unit, inference function supported by the radio network element, or inference function supported by the core network element. Wherein, the inference function supported by the domain management function unit can be an MDA inference function, the inference function supported by the radio network element can be a RAN intelligent inference function, and the inference function supported by the core network element can be an NWDAF inference function.

[0149] For example, the first capability information may include the name of the federated training and the inference type corresponding to the federated learning. The inference type corresponding to the federated learning can be understood as the inference type to which federated learning is applicable. For example, federated learning is applicable to the NWDAF inference function.

[0150] The first capability information may include the name of the reinforcement training and the corresponding inference type. The inference type of the reinforcement training can be understood as the type of inference that the reinforcement training is applicable to. For example, reinforcement training is applicable to MDA inference functionality.

[0151] The first capability information may include the name of the distributed training and the corresponding inference type. The inference type of distributed training can be understood as the type of inference that distributed training is applicable to. For example, distributed training is applicable to the RAN intelligent inference function.

[0152] The first capability information may include the name of the generative AI and the corresponding reasoning type. The reasoning type of the generative AI can be understood as the type of reasoning that the generative AI is suitable for. The generative AI is suitable for MDA reasoning functions.

[0153] 5) The inference functions applicable to the training techniques supported by the first management device and / or network element device.

[0154] For example, the inference function includes at least one of the following: at least one capability under the MDA inference function, at least one capability under the RAN intelligent inference function, or at least one capability under the NWDAF inference function.

[0155] For example, MDA inference capabilities may include at least one of the following capabilities: coverage analysis capability, or mobility analysis capability, etc. For details, please refer to the types of MDA defined in 3GPP TS28.104.

[0156] The RAN intelligent reasoning function may include at least one of the following capabilities: random access channel (RACH) optimization, mobility load balancing (MLB) optimization, or energy saving (ES) optimization, etc. For details, please refer to the use cases defined in 3GPP TS28.313 or TS28.310.

[0157] NWDAF inference capabilities may include at least one of the following: slice load analysis or user data congestion analysis, etc. For details, please refer to the use cases defined in 3GPP TS23.288.

[0158] For example, the first capability information may include the name of the federated training and the inference function corresponding to the federated learning. The inference function corresponding to the federated learning can be understood as the inference function applicable to federated learning.

[0159] The primary capability information may include the name of the reinforcement training and the corresponding reasoning function. The reasoning function corresponding to the reinforcement training can be understood as the reasoning function applicable to the reinforcement training.

[0160] The first capability information may include the name of the distributed training and the corresponding inference function. The inference function corresponding to the distributed training can be understood as the inference function applicable to distributed training.

[0161] The primary capability information may include the name of the generative AI and its corresponding reasoning function. The reasoning function of the generative AI can be understood as the reasoning function applicable to the generative AI.

[0162] 6) The model capabilities applicable to the training techniques supported by the first management device and / or network element device.

[0163] For example, model capabilities may include at least one of the following: a one-sided model, a two-sided model, a network device model, or a user device model. A one-sided model refers to a model trained on only one side, such as training only on the network device side or only on the terminal device side; a two-sided model refers to a model that needs to be trained and coordinated simultaneously on both the network device and terminal device sides; a network device model refers to a model that only needs to be trained on the network device; and a user device model refers to a model that only needs to be trained on the terminal device.

[0164] For example, the first capability information may include the name of the federated training and the model capabilities corresponding to federated learning.

[0165] The first capability information may include the name of the reinforcement training and the corresponding model capability.

[0166] The first capability information may include the name of the distributed training and the corresponding model capabilities.

[0167] The first capability information may include the name of the generative AI and the corresponding model capabilities.

[0168] Optionally, the first capability information may also include the name or identifier of the device corresponding to the training capability. For example, the first capability information includes the name of the first management device and the training capabilities supported by the first management device, the name of network element device 1 and the training capabilities supported by network element device 1, the name of network element device 2 and the training capabilities supported by network element device 2, and so on. The training capabilities supported by each device may include one or more of the above six items.

[0169] Optionally, the first capability information may also include the computing resources corresponding to the first management device and / or network element device, such as the resources used by the central processing unit (CPU) for network device training execution.

[0170] In one possible implementation, the first management device can send first capability information to the second management device in the following way: the first management device can create an object trained by an AI / ML model and configure the first capability information to the object trained by the AI / ML model.

[0171] The first management device can send first capability information to the second management device via the `notifyMOICreation` operation, and the second management device can receive the first capability information. If the training capabilities supported by the first management device and / or the network element device are modified or altered, updated capability information can also be sent to the second management device via the `notifyMOIAttributeValueChanges` operation. Of course, the first management device can also send capability information via other operations, which are not limited herein.

[0172] In one possible implementation, the first capability information may be sent by the first management device at the request of the second management device. For example, prior to S501, the second management device may send a capability request to the first management device, which requests training capabilities supported by the first management device and / or network element devices.

[0173] Optionally, the first management device can obtain the training capabilities supported by the network element devices. For example, the first management device can send a capability request to at least one network element device, which requests the training capabilities supported by at least one network element device. The at least one network element device sends second capability information to the first management device, wherein the second capability information of any network element device indicates the training capabilities supported by that network element device. The information structure of the second capability information is similar to that of the first capability information, and will not be elaborated further here.

[0174] As an optional approach, the first management device can acquire the training capabilities of the network element devices and forward them to the second management device. That is, the first capability information includes the second capability information of at least one network element device.

[0175] As an alternative approach, after acquiring the training capabilities of the network element devices, the first management device can process these capabilities, such as by summarizing them. That is, the first capability information is determined based on the second capability information of the at least one network element device.

[0176] S302, the second management device sends a training request to the first management device. Correspondingly, the first management device receives the training request from the second management device.

[0177] The Training Request is used to request the execution of the first inference function using the first training technique.

[0178] Optionally, a training request may include at least one of the following: the name of the first training technique, the role corresponding to the first training technique, the entity to which the role corresponding to the first training technique belongs, and the model capabilities of the first training technique. The name, role, entity to which the role belongs, and model capabilities of the first training technique can be found in the preceding description and will not be repeated here.

[0179] In one possible implementation, the second management device can select the first training technique based on at least one of the following information: computing resources, number of training data samples, training data dimensionality, training data sample overlap rate, training data feature overlap rate, first inference function, fault tolerance requirements, or robustness requirements. It should be noted that the number of training data samples refers to the number of data points used for training; each sample is a data point in the data and has one or more data features. The training data dimensionality refers to the number of training data features; it can be understood that each training data sample consists of one or more features, which can be any numerical, categorical, or other type of data. The training data sample overlap rate is the probability of repeated data points in the training data samples. The training data feature overlap rate is the probability of repeated data features in the training data samples.

[0180] In another possible implementation, when the training function is on a network element, the second management device can select a training technique for the network element's model training based on the applicable region of the inference function. For example, if two or more network elements fall within the same optimization region, the same training technique can be selected for these network elements. For instance, the second management device can determine the training technique suitable for the first inference function based on first capability information. It is understood that after receiving the first capability information supported by the first management device, the second management device selects the training technique suitable for the first inference function based on the capability information supported by the first management device.

[0181] If there are multiple training techniques applicable to the first inference function, the first training technique can be further determined based on factors such as computing resources, number of training data samples, training data dimension, training data sample overlap rate, data dimension, training data feature overlap rate, fault tolerance requirements, or robustness requirements.

[0182] For example, when computing resources are greater than or equal to a computing resource threshold, distributed learning and training of the model will be triggered first.

[0183] If the number of training data samples is greater than or equal to the threshold for the number of training data samples, distributed learning training will be triggered first.

[0184] Vertical federated training can be triggered preferentially if the overlap rate of training data samples is greater than or equal to the training data sample overlap rate threshold and the data dimension is greater than or equal to the data dimension threshold.

[0185] If the overlap rate of training data features is greater than or equal to the threshold for overlap rate of training data features, and the overlap rate of training data samples is less than or equal to the threshold for overlap rate of training data samples, then horizontal federated training can be triggered first.

[0186] If the requirements for fault tolerance and robustness are met, reinforcement training can be triggered first.

[0187] Understandably, in the examples above, triggering the corresponding training technique requires that the training technique applicable to the first inference function includes that technique. For example, triggering distributed learning training of the model requires that the training technique applicable to the first inference function includes distributed learning training of the model, and so on.

[0188] Optionally, the second management device may send a training request to the first management device in the following manner: the second management device requests the first management device to create an object for ML model training. The first management device creates the object for ML model training and configures the training technique selected by the second management device into the object for ML model training.

[0189] In one possible implementation, after receiving a training request, the first management device can use a first training technique to train the AI / ML model.

[0190] In another implementation, after receiving a training request, the first management device triggers a network element to train an AI / ML model using a first training technique based on the training request. For example, the first management device can forward the training request to a network element. Understandably, this network element satisfies the training request. This network element can be indicated by the second management device; for example, the training request may include the name of the network element training the AI / ML model. Alternatively, it can be determined by the first management device based on the training request.

[0191] The two implementation methods described above can be implemented individually or in combination; this application does not impose any specific limitations.

[0192] Optionally, the first management device may send a training report to the second management device. In one possible approach, the training report may reuse the content of the existing MLTrainingReport defined in 3GPP TS28.105, which is not limited herein.

[0193] The aforementioned training reports include training reports for training AI / ML models on the first management device and / or training reports for training AI / ML models on network element devices.

[0194] In one possible implementation, the first management device can send a training report to the second management device in the following way: the first management device can create an object for AI / ML model training and configure the training report to the object for AI / ML model training.

[0195] The first management device can send the training report to the second management device via the `notifyMOICreation` operation. If the training report of the AI / ML model trained by the first management device and / or the training report of the AI / ML model trained by the network element device are modified or changed, the updated training report can also be sent to the second management device via the `notifyMOIAttributeValueChanges` operation. Of course, the first management device can also send the training report via other operations, which are not limited here.

[0196] The above describes the methods for selecting training techniques and training AI / ML models. The following describes the implementation methods for information interaction, such as reporting primary capability information and sending training requests.

[0197] In one example Figure 3 The reporting of the aforementioned first capability information can be achieved using an existing information object class (IOC), such as by including a new, non-writable attribute `supportedLearningTechnology` in the object class `MLTrainingFunction` (IOC) that enhances ML training functionality. It should be noted that a new object class can also be used; this application does not limit this approach.

[0198] The MLTrainingFunction (IOC) represents the ML model functionality created by a producer (such as the first management device). Consumers (such as the second management device) can use this IOC to obtain the training capabilities supported by the producer (such as the first management device) and / or network element devices. For example, the supported Learning Technology indicates the training capabilities supported by the vendor's network management system (such as the first management device) and / or network element devices. The training technology can be represented by a data type. The values ​​for "Support Qualifier" include mandatory (M), optional (O), conditional optional (CO), and conditional mandatory (CM). M indicates the attribute is mandatory, O indicates it is optional, CM indicates it is conditionally mandatory (meaning the attribute is mandatory only if a certain condition is met), and CO indicates it is optional only if a certain condition is met. The values ​​for "Readability" include TRUE (T) and FALSE (F). T indicates the attribute is readable by the second management device, and F indicates it is not readable. The values ​​for "Writability" also include TRUE and FALSE. T indicates the attribute is writable by the second management device, and F indicates it is not writable. This explanation also applies to the tables below and will not be repeated hereafter.

[0199] For example, the enhancement of MLTrainingFunction (IOC) can be shown in Table 1.

[0200] Table 1

[0201]

[0202] Among them, SupportedlearningTechnology is a data type used to indicate the first capability information. Its corresponding isWritable is F, which means that this parameter is not writable by the second management device, that is, this parameter is configured by the first management device. Taking the content of the first capability information mentioned above as an example, SupportedlearningTechnology specifically includes the following 5 fields, as shown in Table 2.

[0203] Table 2

[0204]

[0205] Among them, `supportedlearningTechnologyName` indicates the name of the supported training technology; `supportedRole` indicates the role corresponding to each training technology; `supportedInferenceTypeList` indicates the inference type corresponding to each training technology; `supportedInferenceNameList` indicates the name of the inference function corresponding to each training technology; and `supportedInferenceNameList` indicates the model capabilities corresponding to each training technology.

[0206] In one example Figure 3 The sending of the training request shown can be implemented using an existing IOC, such as by enhancing the ML training request object class (MLTrainingRequest (IOC)) to include a new writable attribute, LearningTechnology. It should be noted that a new object class can also be used; this application does not limit this approach.

[0207] MLTrainingRequest (IOC) represents an operation triggered by a consumer (such as a second management device) to create an ML model training object by a producer (such as a first management device). The consumer (such as a second management device) can use this IOC to indicate a training request to the producer (such as a first management device) (e.g., to identify the training request by the LearningTechnology attribute).

[0208] For example, an enhancement to the MLTrainingRequest (IOC) can be shown in Table 3.

[0209] Table 3

[0210]

[0211]

[0212] Here, LearningTechnology indicates the first training technology selected by the second management device, and its corresponding isWritable is T, which means that this parameter is writable by the second management device and is configured by the second management device.

[0213] Optionally, LearningTechnology can also be represented by a data type, which may include one or more of the following: the name of the training technology, the role corresponding to the training technology, the entity to which the role corresponding to the training technology belongs, and the model capabilities of the training technology.

[0214] This application proposes a unified management and control scheme for training technologies, enabling a second management device (e.g., a cross-domain management unit) to acquire training technologies supported by a first management device (e.g., a domain management entity). Thus, the second management device (e.g., the cross-domain management unit) can select appropriate training technologies for specific inference functions according to requirements to trigger the first management device (e.g., the domain management entity) and / or network element devices to perform training, which helps to reduce training overhead, improve training speed, and improve the accuracy of inference functions.

[0215] See Figure 4 This is an exemplary flowchart of another communication method provided in an embodiment of this application. The method includes:

[0216] S401, the second management device sends a request for at least one training technique to the first management device. Correspondingly, the first management device receives the request for at least one training technique from the second management device.

[0217] For example, at least one of the above training techniques includes at least one of the following: federated training, distributed training, reinforcement training, or generative AI.

[0218] Optionally, the required information may include at least one of the following: the name of the training technique, threshold information, inference type, or inference function.

[0219] The following examples illustrate the required information in conjunction with the training techniques described above.

[0220] 1) Name of training technique

[0221] For example, the requirements for federated learning include the name of the federated training method. The requirements for reinforcement training include the name of the reinforcement training method. The requirements for distributed training include the name of the distributed training method. The requirements for generative AI include the name of the generative AI method.

[0222] 2) Threshold information

[0223] For example, taking distributed training as an example, the threshold information for distributed training may include at least one of the following: threshold information for distributed model learning, or threshold information for distributed data learning.

[0224] The threshold information for distributed model learning includes a computing resource threshold. Based on this approach, one possible implementation is that the first management device can preferentially trigger distributed model learning training if the computing resources are greater than or equal to the computing resource threshold.

[0225] The threshold information for distributed data learning includes at least one of the following: a threshold for the number of training data samples, or a threshold for the dimension of the training data. Based on this approach, one possible implementation is that the first management device can preferentially trigger distributed data learning training when the number of training data samples is greater than or equal to the threshold for the number of training data samples.

[0226] Taking federated training as an example, the threshold information for federated training may include at least one of the following: threshold information for vertical federated training or threshold information for horizontal federated training.

[0227] The threshold information for vertical federated training includes at least one of the following: a training data sample overlap rate threshold, a data dimension threshold, or the number of training data samples. Based on this approach, one possible implementation is that the first management device can preferentially trigger vertical federated training when the training data sample overlap rate is greater than or equal to the training data sample overlap rate threshold, and the data dimension is greater than or equal to the data dimension threshold.

[0228] The threshold information for horizontal federated training includes at least one of the following: a data dimension threshold, a training data feature overlap rate threshold, or a training data sample overlap rate threshold. Based on this approach, one possible implementation is that the first management device can preferentially trigger the use of horizontal federated training when the training data feature overlap rate is greater than or equal to the training data feature overlap rate threshold, and the training data sample overlap rate is less than or equal to the training data sample overlap rate threshold.

[0229] Taking generative AI as an example, the threshold information for generative AI includes at least one of the following: a computing resource threshold or a training data sample number threshold. Based on this approach, one possible implementation is that the first management device can preferentially trigger the use of generative AI when the computing resources are greater than or equal to the computing resource threshold and the number of training data samples is greater than or equal to the training data sample number threshold.

[0230] Taking reinforcement training as an example, the threshold information for reinforcement training includes at least one of the following: fault tolerance requirements or robustness requirements. Based on this approach, one possible implementation is that the first management device can preferentially trigger the use of reinforcement training if both fault tolerance and robustness requirements are met.

[0231] 3) Reasoning Types

[0232] For example, the inference type includes at least one of the following: MDA inference function, RAN intelligent inference function, or NWDAF inference function.

[0233] The inference types included in the requirements information for a training technique can be understood as the inference types applicable to that training technique. For example, the inference types included in the requirements information for federated learning can be understood as the inference types applicable to federated learning. The inference types included in the requirements information for reinforcement training can be understood as the inference types applicable to reinforcement training. The inference types included in the requirements information for distributed training can be understood as the inference types applicable to distributed training. The inference types included in the requirements information for generative AI can be understood as the inference types applicable to generative AI.

[0234] 4) Reasoning function

[0235] The inference function includes at least one of the following: at least one capability under the MDA inference function, at least one capability under the RAN intelligent inference function, or at least one capability under the NWDAF inference function.

[0236] For example, MDA inference capabilities may include at least one of the following capabilities: coverage analysis capability, or mobility analysis capability, etc. For details, please refer to the types of MDA defined in 3GPP TS28.104.

[0237] The RAN intelligent reasoning function may include at least one of the following capabilities: random access channel (RACH) optimization, mobility load balancing (MLB) optimization, or energy saving (ES) optimization, etc. For details, please refer to the use cases defined in 3GPP TS28.313 or TS28.310.

[0238] NWDAF inference capabilities may include at least one of the following: slice load analysis or user data congestion analysis, etc. For details, please refer to the use cases defined in 3GPP TS23.288.

[0239] The inference functions included in the requirements information for a training technique can be understood as the inference functions applicable to that training technique. For example, the inference functions included in the requirements information for federated learning can be understood as the inference functions applicable to federated learning. The inference functions included in the requirements information for reinforcement training can be understood as the inference functions applicable to reinforcement training. The inference functions included in the requirements information for distributed training can be understood as the inference functions applicable to distributed training. The inference functions included in the requirements information for generative AI can be understood as the inference functions applicable to generative AI.

[0240] Optionally, the second management device may also send the name or identifier of the first inference function to the first management device.

[0241] In one possible implementation, the second management device can send at least one training technique requirement to the first management device in the following manner: The second management device can request the first management device to create an object for AI / ML model training. The first management device creates the AI / ML model training object. The second management device configures the requirement information of at least one training technique into the AI / ML model training object.

[0242] Optionally, prior to S401, the second management device may acquire requirement information for at least one training technique. This requirement information may be acquired by the second management device from other devices or determined by the second management device itself; no specific limitation is made here.

[0243] As an alternative, after receiving the requirement information of at least one training technology, the second management device can train an AI / ML model based on the requirement information of the at least one training technology.

[0244] In one implementation, the first management device can select a first training technique based on the requirements of at least one training technique and at least one of the following: computing resources, number of training data samples, training data dimension, training data sample overlap rate, data dimension, training data feature overlap rate, first inference function, fault tolerance requirements, or robustness requirements. The first training technique is then used to train the AI / ML model.

[0245] For example, the first management device can determine the training technology suitable for the first inference function based on the requirements information of at least one training technology. If there are multiple training technologies suitable for the first inference function, the first training technology can be further determined based on factors such as computing resources, the number of training data samples, the dimensionality of the training data, the overlap rate of the training data samples, the data dimension, the overlap rate of the training data features, fault tolerance requirements, or robustness requirements.

[0246] For example, when computing resources are greater than or equal to a computing resource threshold, distributed learning and training of the model will be triggered first.

[0247] If the number of training data samples is greater than or equal to the threshold for the number of training data samples, distributed learning training will be triggered first.

[0248] Vertical federated training can be triggered preferentially if the overlap rate of training data samples is greater than or equal to the training data sample overlap rate threshold and the data dimension is greater than or equal to the data dimension threshold.

[0249] If the overlap rate of training data features is greater than or equal to the threshold for overlap rate of training data features, and the overlap rate of training data samples is less than or equal to the threshold for overlap rate of training data samples, then horizontal federated training can be triggered first.

[0250] If the requirements for fault tolerance and robustness are met, reinforcement training can be triggered first.

[0251] Understandably, in the examples above, triggering the corresponding training technique requires that the training technique applicable to the first inference function includes that technique. For example, triggering distributed learning training of the model requires that the training technique applicable to the first inference function includes distributed learning training of the model, and so on.

[0252] As an alternative, after receiving the requirement information of at least one training technology, the second management device can trigger the network element device to train the AI / ML model based on the requirement information of the at least one training technology.

[0253] In one implementation, a first management device can determine a first training technology and a network element device for training the AI / ML model based on the requirement information of at least one training technology. It is understood that the network element device supports the first training technology. The first management device can send a training request to the network element device, which requests the use of the first training technology to train the AI / ML model. The training request can be found in [reference needed]. Figure 3 The description of the training request in the method will not be repeated here.

[0254] The two optional methods mentioned above can be implemented individually or in combination; this application does not impose any specific limitations.

[0255] Optionally, the first management device may send a training report to the second management device. In one possible approach, the training report may reuse the content of the existing MLTrainingReport defined in 3GPP TS28.105, which is not limited herein.

[0256] The aforementioned training reports include training reports for training AI / ML models on the first management device and / or training reports for training AI / ML models on network element devices.

[0257] In one possible implementation, the first management device can send a training report to the second management device in the following way: the first management device can create an object of AI / ML model training report and configure the training results to the object of AI / ML model training report.

[0258] The first management device can send the training report to the second management device via the `notifyMOICreation` operation. If the training report of the AI / ML model trained by the first management device and / or the training report of the AI / ML model trained by the network element device are modified or changed, the updated training report can also be sent to the second management device via the `notifyMOIAttributeValueChanges` operation. Of course, the first management device can also send the training report via other operations, which are not limited here.

[0259] Optionally, the first management device may also send first capability information to the second management device. The first capability information indicates the training capabilities supported by the first management device and / or the network element device. For details regarding the description of the first capability information, the acquisition of training capabilities supported by the network element device, and the methods of sending and updating it, please refer to [link to relevant documentation]. Figure 3 The relevant descriptions of the method will not be repeated here.

[0260] The above describes how to select training techniques and train AI / ML models. The following describes how to send request information.

[0261] In one example Figure 4 The sending of the requirement information shown can be achieved through an existing IOC, such as by adding a writable attribute `LearningTechnologyRequirement` to the enhanced ML training request object class (MLTrainingRequest [IOC]). It should be noted that it can also be achieved through a new object; this application does not limit this approach.

[0262] The MLTrainingRequest (IOC) indicates an operation triggered by a consumer (such as a second management device) to train an ML model object created by a producer (such as a first management device). The consumer (such as the second management device) can use this IOC to indicate the aforementioned requirement information to the producer (such as the first management device) (e.g., identifying the requirement information through the LearningTechnologyRequirement attribute). For example, an enhanced MLTrainingRequest (IOC) can be shown in Table 4.

[0263] Table 4

[0264]

[0265]

[0266] The LearningTechnologyRequirement specifies the aforementioned requirement information. For example, SupportedlearningTechnology can specifically include the following five fields, as shown in Table 5.

[0267] Table 5

[0268]

[0269] Here, `learningTechnologyName` indicates the name of the training technique; `learningTechnologyMetrics` indicates threshold information, such as one or more thresholds for computational resources, number of training data samples, training data dimensionality, training data sample overlap rate, training data feature overlap rate, fault tolerance, or robustness; `learningTechnologyValue` indicates the values ​​of computational resources, number of training data samples, training data dimensionality, training data sample overlap rate, training data feature overlap rate, fault tolerance, or robustness; `learningTechnologyRole` indicates the role of the training technique; and `modelCapabilities` indicates the model capabilities of the training technique.

[0270] This application proposes a unified management and control scheme for training technologies, which enables a second management device (e.g., a cross-domain management unit) to configure at least one training technology requirement information for a first management device (e.g., a domain management entity) based on its needs (e.g., business needs, management needs, inference needs, etc.). This allows the first management device (e.g., the domain management entity) and / or network element devices to select and execute appropriate training technologies for specific inference functions based on the requirement information of at least one training technology, thereby reducing training overhead, increasing training speed, and improving the accuracy of inference functions.

[0271] Based on the same inventive concept as the method embodiments, this application provides a communication device, the structure of which can be as follows: Figure 5 As shown, it includes a communication unit 501 and a processing unit 502.

[0272] In one embodiment, the communication device can specifically be used to implement Figure 3In the embodiments, the method executed by the second management device may be the second management device itself, or a chip or chipset within the second management device, or a portion of a chip for performing related method functions. Specifically, the processing unit 502 is configured to receive first capability information from the first management device via the communication unit 501, the first capability information indicating the training capabilities supported by the first management device and / or network element devices. Furthermore, based on the first capability information, it sends a training request to the first management device via the communication unit 501, the training request requesting the execution of a first inference function using a first training technique.

[0273] Optionally, the processing unit 502 is further configured to send a capability request to the first management device via the communication unit 501, the capability request being used to request training capabilities supported by the first management device and / or network element devices.

[0274] Optionally, the processing unit 502 is further configured to select the first training technique based on at least one of the following information: computing resources, number of training data samples, training data dimension, training data sample overlap rate, data dimension, training data feature overlap rate, the first inference function, fault tolerance requirements, or robustness requirements.

[0275] In one embodiment, the communication device can specifically be used to implement Figure 3 In the embodiments, the method executed by the first management device may be the first management device itself, or a chip or chipset within the first management device, or a portion of the chip that performs the relevant method function. Specifically, the processing unit 502 is configured to send first capability information to the second management device via the communication unit 501, the first capability information indicating the training capabilities supported by the first management device and / or network element devices; and to receive a training request from the second management device via the communication unit 501, the training request being for requesting the execution of a first inference function using a first training technique.

[0276] Optionally, the processing unit 502 is also used to train an artificial intelligence (AI) / machine learning (ML) model using the first training technique.

[0277] Optionally, the processing unit 502 is further configured to trigger the network element device to train the AI / ML model using the first training technique according to the training request.

[0278] Optionally, the processing unit 502 is further configured to receive a capability request from the second management device via the communication unit 501, the capability request being used to request training capabilities supported by the first management device and / or the network element device.

[0279] Optionally, the processing unit 502 is further configured to receive second capability information from the at least one network element device via the communication unit 501, wherein the second capability information of any one of the network element devices indicates the training capabilities supported by the network element device.

[0280] Optionally, the processing unit 502 is further configured to send a capability request to the at least one network element device via the communication unit 501, the capability request being used to request training capabilities supported by the at least one network element device.

[0281] In one embodiment, the communication device can specifically be used to implement Figure 4 In the embodiments, the method executed by the second management device may be the second management device itself, or a chip or chipset within the second management device, or a part of the chip that performs the relevant method function. Specifically, the processing unit 502 is used to acquire requirement information for at least one training technique; the communication unit 501 is used to send the requirement information for the at least one training technique to the first management device.

[0282] In one embodiment, the communication device can specifically be used to implement Figure 4 In the embodiments, the method executed by the first management device or network element device may be the first management device or network element device itself, or a chip or chipset in the first management device or network element device, or a part of the chip for performing the relevant method function. Specifically, the communication unit 501 is used to receive requirement information for at least one training technique from the second management device; the processing unit 502 trains an AI / ML model based on the requirement information for the at least one training technique, and / or triggers the network element device to train the AI / ML model based on the requirement information for the at least one training technique.

[0283] For example, the processing unit 502 is specifically used to: select a training technique based on at least one of the following information: computing resources, number of training data samples, training data dimension, training data sample overlap rate, data dimension, training data feature overlap rate, first inference function, fault tolerance requirement, or robustness requirement; and train an AI / ML model based on the training technique.

[0284] The module division in this application embodiment is illustrative and represents only one logical functional division. In actual implementation, other division methods may be used. Furthermore, the functional modules in the various embodiments of this application can be integrated into a single processor, exist as separate physical entities, or be integrated into a single module. The integrated modules described above can be implemented in hardware or as software functional modules. It is understood that the functions or implementations of the modules in the embodiments of this application can be further described in the relevant descriptions of the method embodiments.

[0285] In one possible approach, the communication device can be as follows: Figure 6 As shown, the device can be a communication device or a chip within a communication device, wherein the communication device can be either the first device or the second device in the above embodiments. The device includes a processor 601 and a communication interface 602, and may also include a memory 603. The processing unit 502 can be the processor 601. The communication unit 501 can be the communication interface 602. Optionally, the processor 601 and the memory 603 can also be integrated together.

[0286] The processor 601 can be a CPU, a digital processing unit, or something similar. The communication interface 602 can be a transceiver, an interface circuit such as a transceiver circuit, or a transceiver chip, etc. The device also includes a memory 603 for storing the program executed by the processor 601. The memory 603 can be non-volatile memory, such as a hard disk drive (HDD) or a solid-state drive (SSD), or it can be volatile memory, such as random-access memory (RAM). The memory 603 can be any other medium capable of carrying or storing desired program code in the form of instructions or data structures that can be accessed by a computer, but is not limited to this.

[0287] The processor 601 is used to execute the program code stored in the memory 603, specifically to perform the actions of the processing unit 502 described above, which will not be described in detail here. The communication interface 602 is specifically used to perform the actions of the communication unit 501 described above, which will not be described in detail here.

[0288] This application embodiment does not limit the specific connection medium between the communication interface 602, processor 601, and memory 603. This application embodiment... Figure 6 The memory 603, processor 601, and communication interface 602 are connected via a bus 604. Figure 6 The connections between other components are shown in bold lines only and are not intended to be limiting. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, Figure 6 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0289] This application also provides a computer-readable storage medium for storing computer software instructions required to execute the processor, including a program required to execute the processor.

[0290] This application also provides a communication system, including methods for implementing... Figure 3 In the embodiments, a communication device for implementing the first management device function is used. Figure 3 In the embodiments, the communication device for implementing the second management device function and the communication device for implementing the second management device function are described. Figure 3 The embodiment is a communication device that functions as a network element.

[0291] This application also provides a communication system, including methods for implementing... Figure 4 In the embodiments, a communication device for implementing the first management device function is used. Figure 4 In the embodiments, the communication device for implementing the second management device function and the communication device for implementing the second management device function are described. Figure 4 The embodiment is a communication device that functions as a network element.

[0292] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0293] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to this application. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0294] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0295] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0296] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A communication method characterized by comprising: The method comprises: receiving first capability information from a first management device, the first capability information indicating training capabilities supported by the first management device and / or a network element device; sending a training request to the first management device according to the first capability information, the training request being used to request a first inference function to be performed using a first training technique.

2. The method of claim 1, wherein, The method further comprises: sending a capability request to the first management device, the capability request being used to request training capabilities supported by the first management device and / or the network element device.

3. The method of claim 1 or 2, wherein, The method further comprises: selecting the first training technique based on at least one of the following: computing resources, number of training data samples, training data dimension, training data sample overlap rate, data dimension, training data feature overlap rate, the first inference function, fault tolerance requirement, or robustness requirement.

4. A communication method characterized by comprising: The method comprises: sending first capability information to a second management device, the first capability information indicating training capabilities supported by the first management device and / or a network element device; receiving a training request from the second management device, the training request being used to request a first inference function to be performed using a first training technique.

5. The method of claim 4, wherein, The method further comprises: training an artificial intelligence (AI) / machine learning (ML) model using the first training technique.

6. The method of claim 4 or 5, wherein, The method further comprises: triggering the network element device to train an AI / ML model using the first training technique according to the training request.

7. The method according to any one of claims 4 to 6, wherein, The method further comprises: receiving a capability request from the second management device, the capability request being used to request training capabilities supported by the first management device and / or the network element device.

8. The method according to any one of claims 4 to 7, wherein, The method further comprises: receiving second capability information from the at least one network element device, the second capability information of any of the network element devices indicating training capabilities supported by the network element device.

9. The method of claim 8, wherein, The method further comprises: sending a capability request to the at least one network element device, the capability request being used to request training capabilities supported by the at least one network element device.

10. The method of any one of claims 1-9, wherein, The training technique comprises at least one of the following: federated training, distributed training, reinforcement training, or generative AI.

11. The method of any one of claims 1-10, wherein, The first capability information comprises at least one of the following: names of training techniques supported by the first management device and / or the network element device, respective roles of training techniques supported by the first management device and / or the network element device, respective entities in which the roles of training techniques supported by the first management device and / or the network element device are located, respective inference types to which the training techniques supported by the first management device and / or the network element device are applicable, respective inference functions to which the training techniques supported by the first management device and / or the network element device are applicable, or respective model capabilities to which the training techniques supported by the first management device and / or the network element device are applicable.

12. The method of claim 11, wherein, The first capability information comprises at least one of the following: names of at least one of the following sub-classes of federated training: vertical federated training, or horizontal federated training; names of at least one of the following sub-classes of distributed training: data distributed learning, or model distributed learning; a name of at least one of the following sub-classes of generative AI: pre-training, fine-tuning training, incremental pre-training, prompt fine-tuning, retrieval-augmentation technique, or generative adversarial network technique.

13. The method of claim 11 or 12, wherein, the first capability information comprises at least one of the following: at least one of the following roles of federated training: federated training server role, or federated training agent role; at least one of the following roles of reinforcement training: reinforcement learning agent, or reinforcement learning environment; at least one of the following roles of distributed training: distributed learning commissioner, or distributed learning agent; at least one of the following roles of generative AI: generator, or discriminator.

14. The method according to any one of claims 11 to 13, wherein, the inference type comprises at least one of the following: management data analytics (MDA) inference function, radio access network (RAN) intelligent inference function, or network data analytics function (NWDAF) inference function.

15. The method according to any one of claims 11 to 14, wherein, the inference function comprises at least one of the following: at least one of the following inference functions of the MDA inference function: coverage analysis capability, or mobility analysis capability; at least one of the following inference functions of the RAN intelligent inference function: random access channel (RACH) optimization, mobile load balancing (MLB) optimization, or energy saving (ES) optimization; at least one of the following inference functions of the NWDAF inference function: slice load analysis, or user data congestion analysis.

16. The method of any one of claims 11-15, wherein, the model capability comprises at least one of the following: one-sided model, two-sided model, network device model, or user equipment model.

17. The method of any one of claims 1-16, wherein, the training request comprises at least one of the following: name of the first training technique, role corresponding to the first training technique, entity in which the role corresponding to the first training technique is located, model capability of the first training technique.

18. A method of communication, comprising: comprises: obtaining requirement information of at least one training technique; sending the requirement information of the at least one training technique to a first management device.

19. A method of communication, comprising: comprises: receiving requirement information of at least one training technique from a second management device; training an artificial intelligence / machine learning (AI / ML) model based on the requirement information of the at least one training technique; and / or, triggering a network element device to train an AI / ML model based on the requirement information of the at least one training technique.

20. The method of claim 19, wherein, the training of the artificial intelligence / machine learning (AI / ML) model based on the requirement information comprises: selecting a training technique based on at least one of the following: computing resource, number of training data samples, training data dimension, training data sample overlap rate, data dimension, training data feature overlap rate, first inference function, fault tolerance requirement, or robustness requirement; training an AI / ML model based on the training technique.

21. The method of any one of claims 18-20, wherein, the at least one training technique comprises at least one of the following: federated training, distributed training, reinforcement training, or generative AI.

22. The method of any one of claims 18-21, wherein, the requirement information comprises at least one of the following: threshold information, inference type, or inference function.

23. The method of claim 22, wherein, threshold information of the distributed training comprises at least one of the following: model distributed learning threshold information, or data distributed learning threshold information, wherein the model distributed learning threshold information comprises a computing resource threshold, and the data distributed learning threshold information comprises at least one of the following: training data sample number threshold and training data dimension threshold; The threshold information of federated training comprises at least one of: threshold information of vertical federated training, or threshold information of horizontal federated training, wherein the threshold information of vertical federated training comprises at least one of: a training data sample overlap rate threshold, a data dimension threshold, or a training data sample number threshold, and the threshold information of horizontal federated training comprises at least one of: a data dimension threshold, a training data feature overlap rate threshold, or a training data sample overlap rate threshold; The threshold information of generative AI comprises at least one of: a computing resource threshold, or a training data sample number threshold; The threshold information of reinforcement training comprises at least one of: a fault tolerance requirement, or a robustness requirement.

24. The method of claim 22 or 23, wherein, The inference type comprises at least one of: a management data analytics (MDA) inference function, a radio access network (RAN) intelligent inference function, or a network data analytics function (NWDAF) inference function.

25. The method of any one of claims 22-24, wherein, The inference function comprises at least one of: The MDA inference function comprises at least one of: a coverage analysis capability, or a mobility analysis capability; The RAN intelligent inference function comprises at least one of: a random access channel (RACH) optimization, a mobile load balancing (MLB) optimization, or an energy saving (ES) optimization; The NWDAF inference function comprises at least one of: a slice load analysis, or a user data congestion analysis.

26. A communications device, characterized by A unit or module for performing the method of any one of claims 1-3, 10-17, or a unit or module for performing the method of any one of claims 18, 21-25.

27. A communications device, characterized by A unit or module for performing the method of any one of claims 4-17, or a unit or module for performing the method of any one of claims 19-25.

28. A communications device, characterized by A processor and a memory, the memory being configured to store program instructions, the processor being configured to execute the program instructions to cause the method of any one of claims 1-3, 10-17, or the method of any one of claims 18, 21-25 to be performed.

29. A communications device, characterized by A processor and a memory, the memory being configured to store program instructions, the processor being configured to execute the program instructions to cause the method of any one of claims 4-17, or the method of any one of claims 19-25 to be performed.

30. A computer-readable storage medium, characterized in that, The computer readable storage medium has stored therein computer readable instructions which, when executed on a communication device, cause the method of any one of claims 1-3, 10-17, or the method of any one of claims 4-17, or the method of any one of claims 18, 21-25, or the method of any one of claims 19-25 to be performed.

31. A computer program product, characterised in that, The computer program product, when executed on a device, causes the device to perform the method of any one of claims 1-3, 10-17, or the method of any one of claims 4-17, or the method of any one of claims 18, 21-25, or the method of any one of claims 19-25.

Images