CN-122002312-A - Communication method and device

Abstract

A communication method and device are applied to the technical field of communication, and the method comprises the steps that a first device determines first information of a first model, the first information comprises model training information and/or model reasoning information, wherein the model training information comprises at least one of data interpretability information, process interpretability information or model interpretability information, the model reasoning information comprises reasoning interpretability information, the first information is sent to a second device, and the second device is a consumer of the first model. By the method provided by the application, the consumer of the first model can better understand and trust the model based on the first information, which is beneficial to improving the usability of the model.

Inventors

• SHI XIAOLI
• ZHANG HAONAN
• XU RUIYUE
• ZOU LAN

Assignees

• 华为技术有限公司

Dates

Publication Date

20260508

Application Date

20241108

Claims (20)

1. 1. A communication method applied to a first device, the method comprising: Determining first information of a first model, wherein the first information comprises model training information and/or model reasoning information, and the model training information comprises at least one of data interpretability information, process interpretability information or model interpretability information, and the model reasoning information comprises reasoning interpretability information; the first information is sent to a second device, which is a consumer of the first model.
2. 2. The method according to claim 1, wherein the method further comprises: and sending second information to the second device, wherein the second information is used for indicating the reasoning type applicable to the first information.
3. 3. The method according to claim 1 or 2, wherein said determining first information of the first model comprises: Receiving third information from the second device, wherein the third information comprises at least one of first indication information, first index, second index, third index, fourth index or first reasoning type, wherein the first indication information is used for indicating to execute at least one of data interpretability, process interpretability, model interpretability or reasoning interpretability, the first index comprises a distribution mode of data samples and/or characteristics of the data samples, the second index comprises training duration and/or iteration times, and the third index comprises at least one of complexity, simulation environment or reasons corresponding to predicted results, and the fourth index comprises performance indexes influenced by reasoning results and/or performance indexes influenced by interference results; and determining the first information according to the third information.
4. 4. The method of claim 3, wherein the third information comprises the first indication information, the first indication information indicating at least one of data interpretability, process interpretability, or model interpretability; the determining the first information according to the third information includes: and according to the interpretability of the third information, performing model training, obtaining the first information, wherein the first information comprises the information of the model training.
5. 5. The method of claim 4, wherein the third information comprises the first inference type; the first inference type includes management data analysis, the information trained by the model includes interpretable information applicable to the management data analysis, or The first inference type includes access network intellectualization, the information trained by the model includes interpretability information applicable to the access network intellectualization, or The first inference type includes a network data analysis function, and the information trained by the model includes interpretable information applicable to the network data analysis function.
6. 6. The method according to any one of claims 3 to 5, wherein the third information comprises the first indication information for indicating that an inference interpretability is performed; the determining the first information according to the third information includes: and according to the interpretability of the third information, performing model reasoning, obtaining the first information, wherein the first information comprises the information of the model reasoning.
7. 7. The method of claim 6, wherein the third information comprises the first inference type; The first reasoning type comprises management data analysis, and the information of the model reasoning comprises at least one of a second reasoning result corresponding to the management data analysis, information of performance influenced by the second reasoning result, a second interference result related to the second reasoning result or information of performance influenced by the second interference result, or The first reasoning type comprises access network intellectualization, and the information of the model reasoning comprises at least one of a third reasoning result corresponding to the access network intellectualization, information of performance influenced by the third reasoning result, a third interference result related to the third reasoning result or information of performance influenced by the third interference result, or The first reasoning type comprises a network data analysis function, and the information of the model reasoning comprises at least one of a fourth reasoning result corresponding to the network data analysis function, information of performance influenced by the fourth reasoning result, a fourth interference result related to the fourth reasoning result or information of performance influenced by the fourth interference result.
8. 8. A communication method applied to a second device, the method comprising: Receiving first information from a first device, the first information comprising model-trained information and/or model-inferred information, wherein the model-trained information comprises at least one of data-interpretable information, process-interpretable information, or model-interpretable information, the model-inferred information comprising inferred-interpretable information; And carrying out corresponding processing according to the first information.
9. 9. The method of claim 8, wherein the method further comprises: Second information is received from the first device, the second information indicating a type of reasoning to which the first information applies.
10. 10. The method according to claim 8 or 9, characterized in that the method further comprises: and sending third information to the first device, wherein the third information comprises at least one of first indication information, first index, second index, third index, fourth index or first reasoning type, the first indication information is used for indicating to execute at least one of data interpretability, process interpretability, model interpretability or reasoning interpretability, the first index comprises a distribution mode of data samples and/or characteristics of the data samples, the second index comprises training duration and/or iteration times, the third index comprises at least one of complexity, simulation environment or reasons corresponding to a prediction result, and the fourth index comprises performance indexes influenced by the reasoning result and/or performance indexes influenced by an interference result.
11. 11. The method of claim 10, wherein the first information comprises information of the model training and the third information comprises the first inference type; the first inference type includes management data analysis, the information trained by the model includes interpretable information applicable to the management data analysis, or The first inference type includes access network intellectualization, the information trained by the model includes interpretability information applicable to the access network intellectualization, or The first inference type includes a network data analysis function, and the information trained by the model includes interpretable information applicable to the network data analysis function.
12. 12. The method of claim 10 or 11, wherein the first information comprises information of the model reasoning and the third information comprises the first reasoning type; The first reasoning type comprises management data analysis, and the information of the model reasoning comprises at least one of a second reasoning result corresponding to the management data analysis, information of performance influenced by the second reasoning result, a second interference result related to the second reasoning result or information of performance influenced by the second interference result, or The first reasoning type comprises access network intellectualization, and the information of the model reasoning comprises at least one of a third reasoning result corresponding to the access network intellectualization, information of performance influenced by the third reasoning result, a third interference result related to the third reasoning result or information of performance influenced by the third interference result, or The first reasoning type comprises a network data analysis function, and the information of the model reasoning comprises at least one of a fourth reasoning result corresponding to the network data analysis function, information of performance influenced by the fourth reasoning result, a fourth interference result related to the fourth reasoning result or information of performance influenced by the fourth interference result.
13. 13. A method according to any one of claims 3 to 7, 10 to 12, wherein the third information is carried in an information object class of a machine learning ML training request, or wherein part of the information in the third information is carried in an information object class of an ML training request and the remaining information is carried in an information object class of an ML reasoning request.
14. 14. The method according to any of claims 1 to 13, wherein the data interpretability information comprises a distribution of data samples and/or characteristics of data samples.
15. 15. The method of claim 14, wherein the characteristics of the data samples include at least one of mobility, coverage, power saving, loading, failure, or traffic experience.
16. 16. The method according to any one of claims 1 to 15, wherein the process interpretability information comprises a training duration and/or a number of iterations.
17. 17. The method according to any one of claims 1 to 16, wherein the model interpretability information includes at least one of complexity, simulation environment, or a reason for which the prediction result corresponds.
18. 18. The method of any one of claims 1 to 17, wherein the inference interpretability information includes at least one of a first inference result, information of a performance affected by the first inference result, a first interference result related to the first inference result, or information of a performance affected by the first interference result.
19. 19. The method of claim 18, wherein the first inference result affected performance comprises at least one of a cell energy consumption, a cell energy efficiency, a cell throughput rate, a cell resource utilization, a number of radio resource control connections, a number of cell handover successes, a number of cell handover failures, or a cell signal strength distribution, and/or wherein the first interference result affected performance comprises at least one of a cell energy consumption, a cell energy efficiency, a cell throughput rate, a cell resource utilization, a number of radio resource control connections, a number of cell handover successes, a number of cell handover failures, or a cell signal strength distribution.
20. 20. The method according to any one of claims 1 to 19, wherein the first information is carried in an information object class of an ML training report or the information of the model training is carried in an information object class of an ML training report and the information of the model reasoning is carried in an information object class of an ML reasoning report.

Description

Communication method and device Technical Field The present application relates to the field of communications technologies, and in particular, to a communications method and apparatus. Background Artificial intelligence/machine learning (ARTIFICIAL INTELLIGENCE/MACHINE LEARNING, AI/ML) technology and related applications are increasingly being adopted by the more widespread industry. Currently, AI/ML techniques introduce an interpretability of the AI/ML model, which is intended to understand and trust the AI/ML model. How to implement the interpretability of AI/ML models for better understanding and trust of the model's consumers (or users of the model) is a current research hotspot. Disclosure of Invention The embodiment of the application provides a communication method and a communication device, which are used for enabling a consumer of a model to better understand the model and a trust model. In a first aspect, the present application provides a communication method, the method being applicable to a first device. Alternatively, the first device may be the producer of the first model. Alternatively, the first model may be an AI/ML model, and the naming and implementation of the first model are not limited in the embodiments of the present application. In one example, the first device may be a domain management functional unit, or a device in the domain management functional unit (e.g., a module, a communication module, a circuit or a chip responsible for communication functions (such as a Modem (Modem) chip, or a system-on-a-chip (SoC) chip or a system-in-a-chip (SYSTEMIN PACKAGE) chip), a chip system or a processor), or a logic node, a logic module or software, etc. capable of implementing all or part of the domain management functional unit. In another example, the first device may also be a network element, or a device in the network element (for example, a module, a communication module, a circuit or a chip responsible for a communication function (such as a Modem chip, or an SoC chip or an SIP chip including a Modem core), a system on a chip or a processor), or a logic node, a logic module, or software that can implement all or part of the network element, or the like. The network element is not limited, and may be, for example, an access network element, a core network element, or the like. In another example, the first device may also be a cross-domain management functional unit, or be a device in the cross-domain management functional unit (for example, a module, a communication module, a circuit or a chip responsible for a communication function (such as a Modem chip, or an SoC chip or an SIP chip containing a Modem core), a system on a chip or a processor), or be a logic node, a logic module, or software, etc. capable of implementing all or part of the cross-domain management functional unit. The method may include the first device determining first information for a first model, the first information including model-trained information and/or model-inferred information, wherein the model-trained information includes at least one of data-interpretable information, process-interpretable information, or model-interpretable information, the model-inferred information including inferred-interpretable information, transmitting the first information to the second device, the second device being a consumer of the first model. Or the method may comprise the first device sending first information to the second device, the first information comprising model-trained information and/or model-inferred information, wherein the model-trained information comprises at least one of data-interpretable information, process-interpretable information, or model-interpretable information, the model-inferred information comprises inference-interpretable information, and the second device is a consumer of the first model. In the application, the first device sends the first information to the second device, the first information comprises at least one of data interpretability information, process interpretability information, model interpretability information and reasoning interpretability information, and the first model is interpreted to the second device from multiple angles of training data (or data samples), training process, training model and model reasoning, so that the second device can better understand and trust the model based on the first information, and the model usefulness is improved. In one possible implementation, the data interpretability information may include a distribution manner (e.g., uniform distribution or non-uniform distribution, etc.) of the data samples and/or characteristics of the data samples. By way of example, the characteristics of the data samples may include at least one of mobility, coverage, power saving, load, failure, or traffic experience. The data sample is the basic input of the training model, the quality of the data sample is critical to the precision of the training m