Communication method and related apparatus

Abstract

A communication method and a related apparatus. In the method, first information acquired by a first communication apparatus is used for indicating inference state information of a first model group, and the first communication apparatus can determine a second model group on the basis of the inference state information of the first model group, wherein the first model group comprises a first model deployed on the first communication apparatus and a second model deployed on a second communication apparatus, and the second model group comprises a third model deployed on the first communication apparatus and a fourth model deployed on the second communication apparatus. In other words, when AI processing is jointly performed by means of model groups deployed on two or more communication apparatuses, the first communication apparatus can update the first model group on the basis of the inference state information of the first model group to obtain the second model group. In this way, the first communication apparatus can perform model updating on a model group on the basis of inference state information of the model group, thereby implementing model management of the model group.

Description

A communication method and related apparatus

[0001] This application claims priority to Chinese Patent Application No. 202411456594.7, filed on October 17, 2024, entitled "A Communication Method and Related Device", the entire contents of which are incorporated herein by reference. Technical Field

[0002] This application relates to the field of communication technology, and in particular to a communication method and related apparatus. Background Technology

[0003] With the development of communication technology, communication equipment in communication systems can now perform not only traditional communication services but also other new types of services, such as artificial intelligence (AI) services. Generally, a communication system capable of handling AI services can also be called an AI system.

[0004] Currently, communication devices can serve as participating nodes in AI systems, applying their computing power to a specific stage of the AI ​​system. Generally, AI functions introduced into communication networks rely on models for implementation. However, in this process, there is currently no solution to address how to manage these models. Summary of the Invention

[0005] This application provides a communication method and related apparatus for implementing model management of model groups.

[0006] The first aspect of this application provides a communication method applied to a first communication device, for example, the method being executed by the first communication device. The first communication device may be a communication equipment (such as a terminal device or network device), or it may be a component of the communication equipment (such as a circuit or chip responsible for communication functions (such as a modem chip, also known as a baseband chip, or a system-on-chip (SoC) chip containing a modem core, or a system-in-package (SIP) chip), or it may be a logic module or software capable of implementing all or part of the functions of the communication equipment.

[0007] In this method, a first communication device acquires first information indicating inference state information of a first model group, the first model group including a first model deployed on the first communication device and a second model deployed on a second communication device; wherein the input of the first model includes the output of the second model, or the input of the second model includes the output of the first model; the first communication device determines a second model group related to the inference state information of the first model group, the second model group including a third model deployed on the first communication device and a fourth model deployed on the second communication device; wherein the input of the third model includes the output of the fourth model, or the input of the fourth model includes the output of the third model; the first communication device sends second information indicating the third model and / or the fourth model. For example, the first model group is different from the second model group.

[0008] Based on the above scheme, the first information acquired by the first communication device is used to indicate the inference state information of the first model group. The second model group is related to the inference state information of the first model group. For example, the first communication device can determine the second model group based on the inference state information of the first model group. The first model group includes a first model deployed on the first communication device and a second model deployed on the second communication device. The second model group includes a third model deployed on the first communication device and a fourth model deployed on the second communication device. In other words, when model groups deployed on two or more communication devices jointly perform AI processing, the first communication device can update the first model group based on its inference state information to obtain the second model group. In this way, the first communication device can update the model groups using their inference state information to achieve model management of the model groups.

[0009] Furthermore, in the above scheme, the first communication device can update the first model group based on the inference state information of the first model group, which can improve the inference performance of the updated model group.

[0010] Furthermore, in the above scheme, the first communication device can update the first model group based on the inference state information of the first model group. Therefore, the above scheme can be applied to scenarios where distributed multi-node collaborative processing of the same AI task occurs, where the multi-nodes include at least the first communication device and the second communication device. In this scenario, the first communication device can update some or all of the models contained in the model group based on the inference state information of the model group, improving the efficiency of model management in multi-node collaborative processing scenarios.

[0011] Furthermore, in the above scheme, the second information sent by the first communication device indicates the third model and / or fourth model included in the updated second model group, so that the first communication device can indicate or register the second model group to the recipient of the second information, so that the recipient can subsequently schedule the second model group.

[0012] In this application, the model may include an AI model, a mathematical model, a neural network model, an AI neural network model, a machine learning model, or an AI processing model, etc. Accordingly, the model involved in this application can be replaced with an AI model, a mathematical model, a neural network model, an AI neural network model, a machine learning model, or an AI processing model, etc.

[0013] In this application, a model is deployed on a communication device (e.g., a first model is deployed on a first communication device, a second model is deployed on a second communication device, etc.), which may include: after the communication device obtains the model parameters of the model, it obtains, generates or constructs the model based on the model parameters of the model, and subsequently the communication device can adjust the model.

[0014] Optionally, model parameters may include one or more of the following: model hyperparameters, model dataset (including model input data and corresponding label data), model structure, model weights, and model parameters.

[0015] It should be understood that a model group contains two or more models that can be used to jointly complete the same AI task. For example, these models complete the same AI task through collaboration. The processing of these models can be serial (or chained), that is, the output of any model can be part or all of the input of one or more other models, and / or, the input of any model can include the output of other models.

[0016] As an example of implementation, the output of the first model can be part or all of the input to the second model. After processing by the second model, the output of the second model is obtained. The execution result of the AI ​​task can include the output of the second model (optionally, it can also include the output of the first model).

[0017] As another implementation example, the output of the second model can be part or all of the input of the first model. After being processed by the first model, the output of the first model is obtained. The execution result of the AI ​​task can include the output of the first model (optionally, it can also include the output of the second model).

[0018] Optionally, when a model group is considered as a single model, the models contained within that model group can be understood as sub-models within that single model. For example, when the first model group is considered as a single model, the first and second models contained within the aforementioned first model group can be understood as sub-models of that single model. Similarly, the second model group, third model group, etc., involved in this application can also be considered as a single model, and correspondingly, that single model can contain two or more sub-models, as can be seen in the implementation of the first model group.

[0019] Optionally, a model group may include two or more models. For example, in addition to the first model and the second model, the first model group may also include other models, which may be deployed on other communication devices different from the first and second communication devices, without limitation here.

[0020] It should be understood that wireless communication signals (such as the transmission and reception of configuration information of communication resources, the transmission and reception of reference signals, etc.) can be transmitted between different communication devices (e.g., the first communication device and the second communication device).

[0021] Optionally, the model group involved in this application (e.g., the first model group, the second model group, the third model group, etc., hereinafter) can be used to manage the wireless communication signal (including at least one of processing, configuration, updating, and optimization). For example, the model may include a model for modulation and demodulation, a model for channel compression, a model for signal transmission and reception, etc. Alternatively, the model involved in this application may also be a model for other tasks, such as a model for image recognition, a model for natural language processing, a model for computer vision, etc.

[0022] Optionally, the inference state information of the first model group mentioned above includes at least one of the following: the performance of the first model group, the number of times the first model group is used for model inference, or the duration of model inference using the first model group, to improve the flexibility of the solution implementation. For example, the performance of the first model group includes model performance (such as one or more of inference accuracy or inference latency) and / or system performance (such as one or more of throughput, packet loss rate, latency, or fairness).

[0023] Optionally, the inference state information can be replaced with other descriptions, such as inference performance information, inference information, model state information, or model performance information.

[0024] Optionally, reasoning can be replaced with other descriptions, such as prediction, deduction, identification, or decision.

[0025] In one possible implementation of the first aspect, the second information satisfies any of the following:

[0026] The first model is different from the third model, the second model is the same as the fourth model, and the second information indicates the third model;

[0027] The first model is the same as the third model, the second model is different from the fourth model, and the second information indicates the fourth model; or,

[0028] The first model is different from the third model, the second model is different from the fourth model, and the second information indicates the third model and the fourth model.

[0029] Based on the above scheme, the inference state information of the first model group can be used to update the first model and / or the second model contained in the first model group. Correspondingly, the second information sent by the first communication device can indicate the update result corresponding to the update, so that the recipient of the second information can obtain the update result.

[0030] For example, if the inference state information of the first model group can be used to update the first model but not the second model, the second information can indicate the updated third model.

[0031] For example, if the inference state information of the first model group can be used to update the second model but not to update the first model, the second information can indicate the updated fourth model.

[0032] For example, if the inference state information of the first model group can be used to update the first model and the second model, the second information can indicate the updated third model and the updated fourth model.

[0033] In one possible implementation of the first aspect, where the second information indicates the third model or the second information indicates the fourth model, the method further includes: the first communication device acquiring third information, the third information being used to indicate inference state information of the second model group; the first communication device determining a third model group, the third model group being related to the inference state information of the second model group, the third model group including a fifth model deployed on the first communication device and a sixth model deployed on the second communication device; wherein the input of the fifth model includes the output of the sixth model, or the input of the sixth model includes the output of the fifth model; and the first communication device sending fourth information, the fourth information indicating the fifth model and the sixth model.

[0034] Optionally, the inference state information of the second model group mentioned above includes at least one of the following: the performance of the second model group, the number of times the second model group was used for model inference, or the duration of model inference using the second model group, to improve the flexibility of the solution implementation. For example, the performance of the second model group includes model performance (such as one or more of inference accuracy or inference latency) and / or system performance (such as one or more of throughput, packet loss rate, latency, or fairness).

[0035] Based on the above scheme, when the second information indicates a third or fourth model, the first communication device updates one of the first and second models based on the inference state information of the first model group. In this case, the first communication device can also update the second model group based on the inference state information of the second model group to obtain the third model group. Thus, after the first communication device can update part of the models based on the inference state information of the first model group to obtain the second model group, the first communication device can further update it based on the inference state information of the second model group to obtain the third model group. Model management can be achieved through two or more model group update processes, so as to improve model processing efficiency through the updated model groups.

[0036] Furthermore, in the above scheme, the fourth information sent by the first communication device indicates the fifth model and / or the sixth model included in the updated third model group, enabling the first communication device to indicate or register the third model group to the recipient of the fourth information, so that the recipient can subsequently schedule the third model group.

[0037] In one possible implementation of the first aspect, if the first condition is satisfied, any of the following is met:

[0038] The fifth model is obtained by switching the third model, and the fourth model is the same as the sixth model.

[0039] The sixth model is obtained by switching the fourth model, and the third model is the same as the fifth model; or

[0040] The fifth model is obtained by switching the third model, and the sixth model is obtained by switching the fourth model.

[0041] Based on the above scheme, when the first condition is met, the first communication device can update the second model group to the third model group by switching models, so as to realize the update of the model group.

[0042] Optionally, a communication device may support one or more models, meaning the communication device can deploy the one or more models and support inference based on the one or more models. Similarly, two or more communication devices may support one or more model groups, meaning the two or more communication devices can each deploy the one or more models and support joint inference based on their respective deployed models.

[0043] In the above process, the model switching process may include: when two or more communication devices support one or more model groups, one communication device switches from supporting one model to supporting another model, and / or, another communication device switches from supporting one model to supporting another model. Before the switch, the two communication devices can perform joint inference using their respective models; after the switch, the two communication devices can also perform joint inference using their respective models.

[0044] As an example, the first communication device supports one or more models including models A1 and B1; the second communication device supports one or more models including model A2; and models A1 and A2 support joint inference, as do models B1 and A2. Accordingly, models A1 and A2 can be considered different models within the same model group, and models B1 and A2 can be considered different models within the same model group. Accordingly, in the above scheme, when the fifth model is obtained by switching the third model, and the fourth and sixth models are the same, the third model can be model A1, the fifth model can be model B1, and the fourth and sixth models can be model A2.

[0045] As another example, the first communication device supports one or more models including model A1; the second communication device supports one or more models including models A2 and B2; and models A1 and A2 support joint reasoning, as do models A1 and B2. Accordingly, models A1 and A2 can be considered different models within the same model group, and models A1 and B2 can be considered different models within the same model group. Accordingly, in the above scheme, when the sixth model is obtained by switching the fourth model, and the third and fifth models are the same, the third and fifth models can be model A1, the fourth model can be model A2, and the fourth and sixth models can be model B2.

[0046] As another example, the first communication device supports one or more models including model A1 and model B1; the second communication device supports one or more models including model A2 and model B2; and model A1 and model A2 support joint reasoning, as do model B1 and model B2. Accordingly, model A1 and model A2 can be considered different models within the same model group, and model B1 and model B2 can be considered different models within the same model group. Correspondingly, in the above scheme, when the fifth model is obtained by switching the third model and the sixth model is obtained by switching the fourth model, the third model can be model A1, the fourth model can be model A2, the fifth model can be model B1, and the sixth model can be model B2.

[0047] As an example, the first condition is associated with the reasoning state information of the second model group. Accordingly, the satisfaction of the first condition can be understood as: the reasoning state information of the second model group satisfies condition A.

[0048] As an example, the first condition is associated with the inference state information and inference configuration information of the second model group. Accordingly, the satisfaction of the first condition can be understood as: the inference state information of the second model group satisfies condition A and the inference configuration information satisfies condition B.

[0049] For example, the inference state information of the second model group satisfies condition A, which includes at least one of the following: the performance of the second model group is lower than or equal to a first threshold, the number of inferences of the second model group is greater than or equal to a second threshold, or the inference duration of the second model group is greater than or equal to a third threshold.

[0050] Optionally, the thresholds involved in this application (such as the first threshold, the second threshold, the third threshold, and other thresholds that may appear later) may be pre-configured or pre-defined, or may be configured by network devices or servers, and are not limited here.

[0051] For example, the inference configuration information satisfying condition B includes: at least one model among the models supported by the first communication device, excluding the third model (or excluding the first and third models), is adapted to the inference configuration information; and / or, at least one model among the models supported by the second communication device, excluding the fourth model (or excluding the second and fourth models), is adapted to the inference configuration information. Correspondingly, the inference configuration information not satisfying condition B includes: none of the models supported by the first communication device are adapted to the inference configuration information, excluding the third model (or excluding the first and third models); and none of the models supported by the second communication device are adapted to the inference configuration information, excluding the fourth model (or excluding the second and fourth models).

[0052] Taking the first communication device as an example, when the inference configuration information satisfies condition B, at least one model among the models supported by the first communication device is adapted to the inference configuration information; in other words, one or more other models supported by the first communication device, other than the third model (or other than the first model and the third model), can be adapted to the inference configuration information, that is, the first model can be switched to any of the other one or more models, and the inference process of any model can satisfy the inference configuration information.

[0053] Similarly, if the inference configuration information does not meet condition B, none of the models supported by the first communication device are compatible with the inference configuration information; in other words, one or more models supported by the first communication device, other than the third model (or other than the first model and the third model), are compatible with the inference configuration information, meaning that the first model cannot obtain a model that meets the inference configuration information through model switching.

[0054] Optionally, the inference configuration information includes one or more model identifiers and / or inference aid information. For example, the inference configuration information includes one or more model identifiers (or identifiers of one or more model groups), and correspondingly, the inference configuration information indicates that the model with the identifier of the one or more model identifiers (or identifiers of one or more model groups) conforms to or matches the inference configuration. As another example, the inference configuration information includes inference aid information, and correspondingly, the inference configuration information indicates that the model's update, optimization, or iteration process needs to be based on the inference aid information.

[0055] For example, inference aid information can indicate the configuration involved in the model inference process. For instance, the inference aid information may include one or more of the following: scene information (e.g., inference scene is indoor, outdoor, idle, busy, movement speed, etc.), model information (e.g., model performance, model complexity, etc.), environmental information (e.g., urban, suburban, etc.), system information (e.g., antenna configuration, system bandwidth, carrier frequency, subcarrier spacing, adoption frequency, throughput, packet loss rate, latency, fairness, etc.), or other information associated with the inference configuration.

[0056] In one possible implementation of the first aspect, the process of the first communication device determining the second model group includes: the first communication device determining the second model group based on inference state information and inference configuration information of the first model group. Similarly, the process of the first communication device determining the third model group includes: the first communication device determining the third model group based on inference state information and inference configuration information of the second model group.

[0057] Based on the above scheme, the first communication device can determine the second model group and / or the third model group based on the inference configuration information, so as to obtain the second model group and / or the third model group that match the inference configuration information, so that the subsequent inference process of the second model group and / or the third model group can obtain the expected inference results, thereby improving the inference performance.

[0058] Optionally, the aforementioned inference configuration information may be provided by other communication devices, or the inference configuration information may be determined by the first communication device (e.g., pre-configured or pre-defined, or determined based on information it collects itself).

[0059] Optionally, the inference configuration information used in the second model group update process can be the same as or different from the inference configuration information used in the third model group update process.

[0060] In one possible implementation of the first aspect, if the second condition is satisfied, any of the following is also satisfied:

[0061] The fifth model is obtained by fine-tuning the third model, and the fourth model is the same as the sixth model.

[0062] The sixth model is obtained by fine-tuning the fourth model, and the third model is the same as the fifth model; or

[0063] The fifth model is obtained by fine-tuning the third model, and the sixth model is obtained by fine-tuning the fourth model.

[0064] Based on the above scheme, when the second condition is met, the first communication device can update the second model group to the third model group by fine-tuning the model, so as to realize the update of the model group.

[0065] Optionally, the communication device can support model fine-tuning of one or more models. That is, after deploying one or more models, the communication device can fine-tune the one or more models to obtain fine-tuned models, and perform model inference based on the fine-tuned models. Similarly, two or more communication devices can support model fine-tuning of one or more model groups. That is, after each of the two or more communication devices deploys one or more models, each of them can fine-tune the one or more models they have deployed to obtain fine-tuned models, and support joint inference based on their respective fine-tuned models.

[0066] In the above process, the model fine-tuning process may include: when two or more communication devices support one or more model groups, one communication device fine-tunes a model from one of the supported models to obtain a fine-tuned model, and / or, another communication device fine-tunes a model from one of the supported models to obtain a fine-tuned model. Before fine-tuning, the two communication devices can perform joint inference using their respective models; after fine-tuning, the two communication devices can also perform joint inference using their respective models.

[0067] For example, a communication device fine-tuning a model may include: the communication device fine-tuning, adjusting, or updating some parameters of the model, where the number of parameters in this part is small (e.g., the number of parameters in this part is below or equal to a threshold, or the ratio of the number of parameters in this part to the total number of parameters in the model is below or equal to a threshold) and / or the number of adjustment rounds is relatively small (e.g., the number of adjustment rounds is less than or equal to a threshold). Optionally, model fine-tuning can be implemented online; for example, during model fine-tuning, the communication device can fine-tune, adjust, or update some parameters of the model online.

[0068] As an example, the second condition is associated with the reasoning state information of the second model group. Accordingly, the satisfaction of the second condition can be understood as: the reasoning state information of the second model group satisfies condition C.

[0069] As an example, the first condition is related to the inference state information and inference configuration information of the second model group. Accordingly, the second condition being met can be understood as: the inference state information of the second model group satisfies condition C, and the inference configuration information does not satisfy condition B.

[0070] For example, the inference state information of the second model group satisfies condition C, including at least one of the following: the performance of the second model group is lower than or equal to the fourth threshold, the number of inferences of the second model group is greater than or equal to the fifth threshold, or the inference duration of the second model group is greater than or equal to the sixth threshold.

[0071] Optionally, the fourth threshold may be less than or equal to the first threshold.

[0072] Optionally, the fifth threshold may be greater than or equal to the second threshold.

[0073] Optionally, the sixth threshold is greater than or equal to the third threshold.

[0074] In one possible implementation of the first aspect, the method further includes: the first communication device sending first request information, the first request information being used to request first data, the first data being used for model fine-tuning of the second model group; and the first communication device receiving or sending the first data.

[0075] Based on the above scheme, when the first communication device determines that the second model group needs to be fine-tuned, the first communication device can send a first request message, so that the recipient of the first request message can receive or send first data based on the first request message, so that the recipient of the first data can achieve model fine-tuning of the second model group based on the first data.

[0076] In one possible implementation of the first aspect, if the third condition is satisfied, any of the following is also satisfied:

[0077] The fifth model is obtained by training the third model, and the fourth model is the same as the sixth model.

[0078] The sixth model is obtained by training the fourth model, and the third model is the same as the fifth model; or

[0079] The fifth model was obtained by training the third model, and the sixth model was obtained by training the fourth model.

[0080] Based on the above scheme, when the third condition is met, the first communication device can update the second model group to the third model group through model training, so as to realize the update of the model group.

[0081] Optionally, the communication device can support model training of one or more models. That is, after deploying one or more models, the communication device can train the one or more models to obtain trained one or more models, and perform model inference based on the trained one or more models. Similarly, two or more communication devices can support model training of one or more model groups. That is, after each of the two or more communication devices deploys one or more models, each of them can train the one or more models they have deployed to obtain trained one or more models, and support joint inference based on their respective trained one or more models.

[0082] In the above process, the model training process may include: when two or more communication devices support one or more model groups, one communication device trains from one of the supported models to obtain a trained model, and / or, another communication device trains from one of the supported models to obtain a trained model. Before training, the two communication devices can perform joint inference using their respective models; after training, the two communication devices can also perform joint inference using their respective models.

[0083] For example, training a model using a communication device may include: training, adjusting, or updating some or all of the model's parameters, where the number of parameters is relatively large (e.g., the number of parameters is higher than or equal to a threshold, or the ratio of the number of parameters to the total number of parameters in the model is higher than or equal to a threshold) and / or the number of adjustment rounds is relatively large (e.g., the number of adjustment rounds is greater than or equal to a threshold). Optionally, model training may be implemented offline; for example, during model training, the communication device may train, adjust, or update some or all of the model's parameters offline.

[0084] As an example, the third condition is associated with the reasoning state information of the second model group. Accordingly, the satisfaction of the third condition can be understood as: the reasoning state information of the second model group satisfies condition D.

[0085] As an example, the first condition is related to the inference state information and inference configuration information of the second model group. Accordingly, the second condition being satisfied can be understood as: the inference state information of the second model group satisfies condition D and the inference configuration information does not satisfy condition B.

[0086] For example, the inference state information of the second model group satisfies condition D, including at least one of the following: the performance of the second model group is lower than or equal to the seventh threshold, the number of inferences of the second model group is greater than or equal to the eighth threshold, or the inference duration of the second model group is greater than or equal to the ninth threshold.

[0087] Optionally, the seventh threshold may be less than or equal to the fourth threshold.

[0088] Optionally, the eighth threshold is greater than or equal to the fifth threshold.

[0089] Optionally, the ninth threshold is greater than or equal to the sixth threshold.

[0090] In one possible implementation of the first aspect, the method further includes: the first communication device sending a second request message for requesting second data for model training of the second model group; and the first communication device receiving or sending the second data.

[0091] Based on the above scheme, when the first communication device determines that model training is to be performed on the second model group, the first communication device can send a second request message, so that the recipient of the second request message can receive or send second data based on the second request message, so that the recipient of the second data can perform model training on the second model group based on the second data.

[0092] In one possible implementation of the first aspect, if the fourth condition is satisfied, then any of the following conditions must be met:

[0093] The third model is obtained by switching the first model, and the second model is the same as the fourth model.

[0094] The fourth model is obtained by switching the second model, and the first model is the same as the third model; or

[0095] The third model is obtained by switching the first model, and the fourth model is obtained by switching the second model.

[0096] Based on the above scheme, when the fourth condition is met, the first communication device can update the first model group to the second model group by switching models, so as to realize the update of the model group.

[0097] As an example, the fourth condition is associated with the reasoning state information of the first model group. Accordingly, the satisfaction of the fourth condition can be understood as: the reasoning state information of the first model group satisfies condition E.

[0098] As an example, the fourth condition is related to the inference state information and inference configuration information of the first model group. Accordingly, the satisfaction of the fourth condition can be understood as: the inference state information of the first model group satisfies condition E and the inference configuration information satisfies condition F.

[0099] For example, the inference state information of the first model group satisfies condition E including at least one of the following: the performance of the first model group is lower than or equal to the tenth threshold, the number of inferences of the first model group is greater than or equal to the eleventh threshold, or the inference duration of the first model group is greater than or equal to the twelfth threshold.

[0100] Optionally, the first threshold is the same as the tenth threshold.

[0101] Optionally, the second threshold is the same as the eleventh threshold.

[0102] Optionally, the third threshold is the same as the twelfth threshold.

[0103] For example, the inference configuration information satisfying condition F includes: at least one model among the models supported by the first communication device, excluding the first model, is adapted to the inference configuration information; and / or, at least one model among the models supported by the second communication device, excluding the second model, is adapted to the inference configuration information. Correspondingly, the inference configuration information not satisfying condition F includes: none of the models supported by the first communication device, excluding the first model, are adapted to the inference configuration information; and none of the models supported by the second communication device, excluding the second model, are adapted to the inference configuration information.

[0104] Taking the first communication device as an example, when the inference configuration information satisfies condition F, at least one model among the models supported by the first communication device is adapted to the inference configuration information; in other words, one or more other models supported by the first communication device besides the first model can be adapted to the inference configuration information, that is, the first model can be switched to any of the other one or more models, and the inference process of any model can satisfy the inference configuration information.

[0105] Similarly, if the inference configuration information does not meet condition F, none of the models supported by the first communication device are compatible with the inference configuration information; in other words, none of the other models supported by the first communication device besides the first model are compatible with the inference configuration information, that is, the first model cannot obtain a model that meets the inference configuration information through model switching.

[0106] In one possible implementation of the first aspect, if the fifth condition is satisfied, then any of the following must be met:

[0107] The third model is obtained by fine-tuning the first model, and the second model is the same as the fourth model.

[0108] The fourth model is obtained by fine-tuning the second model, and the first model is the same as the third model; or

[0109] The third model is obtained by fine-tuning the first model, and the fourth model is obtained by fine-tuning the second model.

[0110] Based on the above scheme, when the fifth condition is met, the first communication device can update the first model group to the second model group by fine-tuning the model, so as to realize the update of the model group.

[0111] As an example, the fifth condition is associated with the reasoning state information of the first model group. Accordingly, the satisfaction of the fifth condition can be understood as: the reasoning state information of the first model group satisfies condition F.

[0112] As an example, the fifth condition is related to the inference state information and inference configuration information of the first model group. Accordingly, the satisfaction of the fifth condition can be understood as: the inference state information of the first model group satisfies condition G and the inference configuration information does not satisfy condition F.

[0113] For example, the inference state information of the first model group satisfies condition G, including at least one of the following: the performance of the first model group is lower than or equal to the thirteenth threshold, the number of inferences of the first model group is greater than or equal to the fourteenth threshold, or the inference duration of the first model group is greater than or equal to the fifteenth threshold.

[0114] Optionally, the thirteenth threshold is less than or equal to the tenth threshold.

[0115] Optionally, the fourteenth threshold is greater than or equal to the eleventh threshold.

[0116] Optionally, the fifteenth threshold is greater than or equal to the twelfth threshold.

[0117] In one possible implementation of the first aspect, the method further includes: the first communication device sending third request information, the third request information being used to request third data, the third data being used for model fine-tuning of the first model group;

[0118] Receive or send the third data.

[0119] Based on the above scheme, when the first communication device determines that the model of the first model group needs to be fine-tuned, the first communication device can send a third request message, so that the recipient of the third request message can receive or send third data based on the third request message, so that the recipient of the third data can achieve model fine-tuning of the first model group based on the third data.

[0120] In one possible implementation of the first aspect, if the sixth condition is satisfied, then any of the following is also satisfied:

[0121] The third model is obtained by training the first model, and the second model is the same as the fourth model.

[0122] The fourth model is obtained by training the second model, and the first model is the same as the third model; or

[0123] The third model is obtained by training the first model, and the fourth model is obtained by training the second model.

[0124] Based on the above scheme, when the sixth condition is met, the first communication device can update the first model group to the second model group through model training to achieve the update of the model group.

[0125] As an example, the sixth condition is associated with the reasoning state information of the first model group. Accordingly, the satisfaction of the sixth condition can be understood as: the reasoning state information of the first model group satisfies condition H.

[0126] As an example, the sixth condition is associated with the inference state information and inference configuration information of the first model group. Accordingly, the satisfaction of the sixth condition can be understood as: the inference state information of the first model group satisfies condition H and the inference configuration information does not satisfy condition F.

[0127] For example, the inference state information of the first model group satisfies condition H, including at least one of the following: the performance of the first model group is lower than or equal to the sixteenth threshold, the number of inferences of the first model group is greater than or equal to the seventeenth threshold, or the inference duration of the first model group is greater than or equal to the eighteenth threshold.

[0128] Optionally, the sixteenth threshold may be less than or equal to the thirteenth threshold.

[0129] Optionally, the seventeenth threshold is greater than or equal to the fourteenth threshold.

[0130] Optionally, the eighteenth threshold is greater than or equal to the fifteenth threshold.

[0131] In one possible implementation of the first aspect, the method further includes: the first communication device sending a fourth request message for requesting fourth data for model training of the first model group;

[0132] Receive or send this fourth data.

[0133] Based on the above scheme, when the first communication device determines that model training is to be performed on the first model group, the first communication device can send a fourth request message, so that the recipient of the fourth request message can receive or send fourth data based on the fourth request message, so that the recipient of the fourth data can realize model training of the first model group based on the fourth data.

[0134] In one possible implementation of the first aspect, the method further includes: the first communication device sending fifth information indicating AI capability information of the first communication device; and / or, the first communication device receiving sixth information indicating AI capability information of the second communication device; wherein the AI ​​capability information of the first communication device and / or the AI ​​capability information of the second communication device are used to determine the second model group.

[0135] Based on the above scheme, the first communication device can also receive or send AI capability information, enabling the recipient of the AI ​​capability information to jointly execute AI tasks based on the AI ​​capabilities of the peer device. For example, any communication device can determine the models supported by the peer communication device based on its AI capability information, facilitating model scheduling for joint AI task execution. Furthermore, any communication device can determine the model update methods supported by the peer communication device based on its AI capability information, enabling the execution of one-sided or multi-sided model update processes. Additionally, any communication device can determine the model update range supported by the peer communication device based on its AI capability information, enabling the execution of single-sided model update processes or multi-sided model joint update processes.

[0136] Optionally, the AI ​​capability information of the first communication device indicates at least one of the following: supported one or more AI models, supported model update methods, whether the model deployed on the first communication device supports independent model updates, or whether the model deployed on the first communication device supports joint model updates with associated models; similarly, the AI ​​capability information of the second communication device indicates at least one of the following: supported one or more AI models, supported model update methods, whether the model deployed on the second communication device supports independent model updates, or whether the model deployed on the second communication device supports joint model updates with associated models; wherein, the supported model update methods include at least one of model switching, model fine-tuning, or model training.

[0137] Optionally, before sending the fifth message, the first communication device may also receive a capability request message, enabling the requesting party to obtain the AI ​​capability information of the first communication device and perform AI-related operations based on the AI ​​capability information (e.g., scheduling the first communication device to perform AI model inference and / or AI model updates). Similarly, before receiving the sixth message, the first communication device may also send a capability request message to the second communication device, enabling the first communication device to obtain the AI ​​capability information of the second communication device and perform AI-related operations based on the AI ​​capability information (e.g., scheduling the second communication device to perform AI model inference and / or AI model updates).

[0138] The second aspect of this application provides a communication method applied to a second communication device, such as being executed by the second communication device, which may be a communication device (e.g., a terminal device or a network device), or the second communication device may be a component of the communication device (e.g., a circuit or chip responsible for communication functions (e.g., a modem chip, also known as a baseband chip, or a SoC chip or SIP chip containing a modem core, etc.), or the second communication device may also be a logic module or software capable of implementing all or part of the functions of the communication device.

[0139] In this method, a second communication device receives second information indicating a third model and / or a fourth model included in a second model group, wherein the third model is deployed on a first communication device and the fourth model is deployed on a second communication device; wherein the input of the third model includes the output of the fourth model, or the input of the fourth model includes the output of the third model; wherein the second model group is determined based on inference state information of a first model group; the first model group includes a first model deployed on the first communication device and a second model deployed on the second communication device; wherein the input of the first model includes the output of the second model, or the input of the second model includes the output of the first model.

[0140] Based on the above scheme, the second communication device receives second information from the first communication device, indicating the third and / or fourth models included in the second model group. This second model group is determined based on the inference state information of the first model group. Specifically, the first model group includes a first model deployed on the first communication device and a second model deployed on the second communication device; the second model group includes a third model deployed on the first communication device and a fourth model deployed on the second communication device. In other words, when model groups deployed on two or more communication devices jointly perform AI processing, the first communication device can update the first model group based on its inference state information to obtain the second model group. In this way, the first communication device can update the model groups using their inference state information, thereby achieving model management of the model groups.

[0141] Furthermore, in the above scheme, the first communication device can update the first model group based on the inference state information of the first model group. Therefore, the above scheme can be applied to scenarios where distributed multi-node collaborative processing of the same AI task occurs, where the multi-nodes include at least the first communication device and the second communication device. In this scenario, the first communication device can update some or all of the models contained in the model group based on the inference state information of the model group, improving the efficiency of model management in multi-node collaborative processing scenarios.

[0142] Furthermore, in the above scheme, the second information received by the second communication device indicates the third model and / or fourth model included in the updated second model group, enabling the first communication device to indicate or register the second model group to the second communication device so that the second communication device can subsequently schedule the third model group.

[0143] Optionally, the inference state information of the first model group includes at least one of the following: the performance of the first model group, the number of times the first model group is used for model inference, or the duration of model inference using the first model group.

[0144] In one possible implementation of the second aspect, the second model group is determined based on the inference state information and inference configuration information of the first model group. Similarly, the third model group is determined based on the inference state information and inference configuration information of the second model group.

[0145] Based on the above scheme, the first communication device can determine the second model group and / or the third model group based on the inference configuration information, so as to obtain the second model group and / or the third model group that match the inference configuration information, so that the subsequent inference process of the second model group and / or the third model group can obtain the expected inference results, thereby improving the inference performance.

[0146] Optionally, the aforementioned inference configuration information may be provided by other communication devices, or the inference configuration information may be determined by the first communication device (e.g., pre-configured or pre-defined, or determined based on information it collects itself).

[0147] Optionally, the inference configuration information used in the second model group update process can be the same as or different from the inference configuration information used in the third model group update process.

[0148] In one possible implementation of the second aspect, the second information satisfies any of the following:

[0149] The first model is different from the third model, the second model is the same as the fourth model, and the second information indicates the third model;

[0150] The first model is the same as the third model, the second model is different from the fourth model, and the second information indicates the fourth model; or,

[0151] The first model is different from the third model, the second model is different from the fourth model, and the second information indicates the third model and the fourth model.

[0152] Based on the above scheme, the inference state information of the first model group can be used to update the first model and / or the second model contained in the first model group. Correspondingly, the second information received by the second communication device can indicate the update result corresponding to the update, so that the second communication device can obtain the update result.

[0153] In one possible implementation of the second aspect, where the second information indicates the third model or the second information indicates the fourth model, the method further includes: the second communication device receiving fourth information indicating a fifth model and / or a sixth model included in the third model group, the fifth model being deployed on the first communication device, and the sixth model being deployed on the second communication device; wherein the input of the fifth model includes the output of the sixth model, or the input of the sixth model includes the output of the fifth model; wherein the third model is determined based on the inference state information of the second model group.

[0154] Optionally, the inference state information of the second model group mentioned above includes at least one of the following: the performance of the second model group, the number of times the second model group was used for model inference, or the duration of model inference using the second model group, to improve the flexibility of the solution implementation. For example, the performance of the second model group includes model performance (such as one or more of inference accuracy or inference latency) and / or system performance (such as one or more of throughput, packet loss rate, latency, or fairness).

[0155] Based on the above scheme, when the second information indicates a third or fourth model, the first communication device updates one of the first and second models based on the inference state information of the first model group. In this case, the first communication device can also update the second model group based on the inference state information of the second model group to obtain the third model group. Thus, after the first communication device can update part of the models based on the inference state information of the first model group to obtain the second model group, the first communication device can further update it based on the inference state information of the second model group to obtain the third model group. Model management can be achieved through two or more model group update processes, so as to improve model processing efficiency through the updated model groups.

[0156] Furthermore, in the above scheme, the fourth information received by the second communication device indicates the fifth model and / or the sixth model included in the updated third model group, enabling the first communication device to indicate or register the third model group to the second communication device so that the second communication device can subsequently schedule the third model group.

[0157] In one possible implementation of the second aspect, if the first condition is satisfied, any of the following is also satisfied:

[0158] The fifth model is obtained by switching the third model, and the fourth model is the same as the sixth model.

[0159] The sixth model is obtained by switching the fourth model, and the third model is the same as the fifth model; or

[0160] The fifth model is obtained by switching the third model, and the sixth model is obtained by switching the fourth model.

[0161] Based on the above scheme, when the first condition is met, the first communication device can update the second model group to the third model group by switching models, so as to realize the update of the model group.

[0162] As an example, the first condition is associated with the reasoning state information of the second model group. Accordingly, the satisfaction of the first condition can be understood as: the reasoning state information of the second model group satisfies condition A.

[0163] As an example, the first condition is associated with the inference state information and inference configuration information of the second model group. Accordingly, the satisfaction of the first condition can be understood as: the inference state information of the second model group satisfies condition A and the inference configuration information satisfies condition B.

[0164] For example, the inference state information of the second model group satisfies condition A, which includes at least one of the following: the performance of the second model group is lower than or equal to a first threshold, the number of inferences of the second model group is greater than or equal to a second threshold, or the inference duration of the second model group is greater than or equal to a third threshold.

[0165] In one possible implementation of the second aspect, if the second condition is satisfied, then any of the following is true:

[0166] The fifth model is obtained by fine-tuning the third model, and the fourth model is the same as the sixth model.

[0167] The sixth model is obtained by fine-tuning the fourth model, and the third model is the same as the fifth model; or

[0168] The fifth model is obtained by fine-tuning the third model, and the sixth model is obtained by fine-tuning the fourth model.

[0169] Based on the above scheme, when the second condition is met, the first communication device can update the second model group to the third model group by fine-tuning the model, so as to realize the update of the model group.

[0170] As an example, the second condition is associated with the reasoning state information of the second model group. Accordingly, the satisfaction of the second condition can be understood as: the reasoning state information of the second model group satisfies condition C.

[0171] As an example, the first condition is related to the inference state information and inference configuration information of the second model group. Accordingly, the second condition being met can be understood as: the inference state information of the second model group satisfies condition C, and the inference configuration information does not satisfy condition B.

[0172] For example, the inference state information of the second model group satisfies condition C, including at least one of the following: the performance of the second model group is lower than or equal to the fourth threshold, the number of inferences of the second model group is greater than or equal to the fifth threshold, or the inference duration of the second model group is greater than or equal to the sixth threshold.

[0173] Optionally, the fourth threshold may be less than or equal to the first threshold.

[0174] Optionally, the fifth threshold may be greater than or equal to the second threshold.

[0175] Optionally, the sixth threshold is greater than or equal to the third threshold.

[0176] In one possible implementation of the second aspect, the method further includes: the second communication device receiving first request information, the first request information being used to request first data, the first data being used for model fine-tuning of the second model group; and the second communication device receiving or sending the first data.

[0177] Based on the above scheme, when the first communication device determines that the second model group needs to be fine-tuned, the first communication device can send a first request message to the second communication device, so that the second communication device can receive or send first data based on the first request message, so that the recipient of the first data can perform model fine-tuning of the second model group based on the first data.

[0178] In one possible implementation of the second aspect, if the third condition is satisfied, any of the following must be met:

[0179] The fifth model is obtained by training the third model, and the fourth model is the same as the sixth model.

[0180] The sixth model is obtained by training the fourth model, and the third model is the same as the fifth model; or

[0181] The fifth model was obtained by training the third model, and the sixth model was obtained by training the fourth model.

[0182] Based on the above scheme, when the third condition is met, the first communication device can update the second model group to the third model group through model training, so as to realize the update of the model group.

[0183] As an example, the third condition is associated with the reasoning state information of the second model group. Accordingly, the satisfaction of the third condition can be understood as: the reasoning state information of the second model group satisfies condition D.

[0184] As an example, the first condition is related to the inference state information and inference configuration information of the second model group. Accordingly, the second condition being satisfied can be understood as: the inference state information of the second model group satisfies condition D and the inference configuration information does not satisfy condition B.

[0185] For example, the inference state information of the second model group satisfies condition D, including at least one of the following: the performance of the second model group is lower than or equal to the seventh threshold, the number of inferences of the second model group is greater than or equal to the eighth threshold, or the inference duration of the second model group is greater than or equal to the ninth threshold.

[0186] Optionally, the seventh threshold may be less than or equal to the fourth threshold.

[0187] Optionally, the eighth threshold is greater than or equal to the fifth threshold.

[0188] Optionally, the ninth threshold is greater than or equal to the sixth threshold.

[0189] In one possible implementation of the second aspect, the method further includes: the second communication device receiving second request information, the second request information being used to request configuration of second data, the second data being used for model training of the second model group; and the second communication device receiving or sending the second data.

[0190] Based on the above scheme, when the first communication device determines that model training is to be performed on the second model group, the first communication device can send a second request message to the second communication device, so that the second communication device can receive or send second data based on the second request message, so that the recipient of the second data can perform model training on the second model group based on the second data.

[0191] In one possible implementation of the second aspect, if the fourth condition is satisfied, then any of the following conditions must be met:

[0192] The third model is obtained by switching the first model, and the second model is the same as the fourth model.

[0193] The fourth model is obtained by switching the second model, and the first model is the same as the third model; or

[0194] The third model is obtained by switching the first model, and the fourth model is obtained by switching the second model.

[0195] Based on the above scheme, when the fourth condition is met, the first communication device can update the first model group to the second model group by switching models, so as to realize the update of the model group.

[0196] As an example, the fourth condition is associated with the reasoning state information of the first model group. Accordingly, the satisfaction of the fourth condition can be understood as: the reasoning state information of the first model group satisfies condition E.

[0197] As an example, the fourth condition is related to the inference state information and inference configuration information of the first model group. Accordingly, the satisfaction of the fourth condition can be understood as: the inference state information of the first model group satisfies condition E and the inference configuration information satisfies condition F.

[0198] For example, the inference state information of the first model group satisfies condition E including at least one of the following: the performance of the first model group is lower than or equal to the tenth threshold, the number of inferences of the first model group is greater than or equal to the eleventh threshold, or the inference duration of the first model group is greater than or equal to the twelfth threshold.

[0199] For example, the inference configuration information satisfying condition F includes: at least one model among the models supported by the first communication device, excluding the first model, is adapted to the inference configuration information; and / or, at least one model among the models supported by the second communication device, excluding the second model, is adapted to the inference configuration information. Correspondingly, the inference configuration information not satisfying condition F includes: none of the models supported by the first communication device, excluding the first model, are adapted to the inference configuration information; and none of the models supported by the second communication device, excluding the second model, are adapted to the inference configuration information.

[0200] Taking the first communication device as an example, when the inference configuration information satisfies condition F, at least one model among the models supported by the first communication device is adapted to the inference configuration information; in other words, one or more other models supported by the first communication device besides the first model can be adapted to the inference configuration information, that is, the first model can be switched to any of the other one or more models, and the inference process of any model can satisfy the inference configuration information.

[0201] Similarly, if the inference configuration information does not meet condition F, none of the models supported by the first communication device are compatible with the inference configuration information; in other words, none of the other models supported by the first communication device besides the first model are compatible with the inference configuration information, that is, the first model cannot obtain a model that meets the inference configuration information through model switching.

[0202] In one possible implementation of the second aspect, if the fifth condition is satisfied, then any of the following must be satisfied:

[0203] The third model is obtained by fine-tuning the first model, and the second model is the same as the fourth model.

[0204] The fourth model is obtained by fine-tuning the second model, and the first model is the same as the third model; or

[0205] The third model is obtained by fine-tuning the first model, and the fourth model is obtained by fine-tuning the second model.

[0206] Based on the above scheme, when the fifth condition is met, the first communication device can update the first model group to the second model group by fine-tuning the model, so as to realize the update of the model group.

[0207] As an example, the fifth condition is associated with the reasoning state information of the first model group. Accordingly, the satisfaction of the fifth condition can be understood as: the reasoning state information of the first model group satisfies condition G.

[0208] As an example, the fifth condition is related to the inference state information and inference configuration information of the first model group. Accordingly, the satisfaction of the fifth condition can be understood as: the inference state information of the first model group satisfies condition G and the inference configuration information does not satisfy condition F.

[0209] For example, the inference state information of the first model group satisfies condition G, including at least one of the following: the performance of the first model group is lower than or equal to the thirteenth threshold, the number of inferences of the first model group is greater than or equal to the fourteenth threshold, or the inference duration of the first model group is greater than or equal to the fifteenth threshold.

[0210] Optionally, the thirteenth threshold is less than or equal to the tenth threshold.

[0211] Optionally, the fourteenth threshold is greater than or equal to the eleventh threshold.

[0212] Optionally, the fifteenth threshold is greater than or equal to the twelfth threshold.

[0213] In one possible implementation of the second aspect, the method further includes: the second communication device sending a third request message, the third request message being used to request third data, the third data being used for model fine-tuning of the first model group;

[0214] Receive or send the third data.

[0215] Based on the above scheme, when the first communication device determines that the model of the first model group needs to be fine-tuned, the first communication device can send a third request information to the second communication device, so that the second communication device can receive or send third data based on the third request information, so that the recipient of the third data can achieve model fine-tuning of the first model group based on the third data.

[0216] In one possible implementation of the second aspect, if the sixth condition is satisfied, then any of the following must be satisfied:

[0217] The third model is obtained by training the first model, and the second model is the same as the fourth model.

[0218] The fourth model is obtained by training the second model, and the first model is the same as the third model; or

[0219] The third model is obtained by training the first model, and the fourth model is obtained by training the second model.

[0220] Based on the above scheme, when the sixth condition is met, the first communication device can update the first model group to the second model group through model training to achieve the update of the model group.

[0221] As an example, the sixth condition is associated with the reasoning state information of the first model group. Accordingly, the satisfaction of the sixth condition can be understood as: the reasoning state information of the first model group satisfies condition H.

[0222] As an example, the sixth condition is associated with the inference state information and inference configuration information of the first model group. Accordingly, the satisfaction of the sixth condition can be understood as: the inference state information of the first model group satisfies condition H and the inference configuration information does not satisfy condition F.

[0223] For example, the inference state information of the first model group satisfies condition H, including at least one of the following: the performance of the first model group is lower than or equal to the sixteenth threshold, the number of inferences of the first model group is greater than or equal to the seventeenth threshold, or the inference duration of the first model group is greater than or equal to the eighteenth threshold.

[0224] Optionally, the sixteenth threshold may be less than or equal to the thirteenth threshold.

[0225] Optionally, the seventeenth threshold is greater than or equal to the fourteenth threshold.

[0226] Optionally, the eighteenth threshold is greater than or equal to the fifteenth threshold.

[0227] In one possible implementation of the second aspect, the method further includes: the second communication device sending a fourth request message for requesting fourth data for model training of the first model group; and the second communication device receiving or sending the fourth data.

[0228] Based on the above scheme, when the first communication device determines that model training is to be performed on the first model group, the first communication device can send a fourth request message to the second communication device, so that the second communication device can receive or send fourth data based on the fourth request message, so that the recipient of the fourth data can perform model training on the first model group based on the fourth data.

[0229] In one possible implementation of the second aspect, the method further includes: the second communication device receiving fifth information indicating AI capability information of the first communication device; and / or, the second communication device sending sixth information indicating AI capability information of the second communication device; wherein the AI ​​capability information of the first communication device and / or the AI ​​capability information of the second communication device are used to determine the second model group.

[0230] Based on the above scheme, the second communication device can also receive or send AI capability information, enabling the recipient of the AI ​​capability information to jointly execute AI tasks based on the AI ​​capabilities of the peer device. For example, any communication device can determine the models supported by the peer communication device based on its AI capability information, facilitating model scheduling for joint AI task execution. Furthermore, any communication device can determine the model update methods supported by the peer communication device based on its AI capability information, enabling the execution of one-sided or multi-sided model update processes. Additionally, any communication device can determine the model update range supported by the peer communication device based on its AI capability information, enabling the execution of single-sided model update processes or multi-sided model joint update processes.

[0231] Optionally, the AI ​​capability information of the first communication device indicates at least one of the following: supported one or more AI models, supported model update methods, whether the model deployed on the first communication device supports independent model updates, or whether the model deployed on the first communication device supports joint model updates with associated models; similarly, the AI ​​capability information of the second communication device indicates at least one of the following: supported one or more AI models, supported model update methods, whether the model deployed on the second communication device supports independent model updates, or whether the model deployed on the second communication device supports joint model updates with associated models; wherein, the supported model update methods include at least one of model switching, model fine-tuning, or model training.

[0232] Optionally, before receiving the fifth message, the second communication device may also send a capability request message, enabling the second communication device to obtain the AI ​​capability information of the first communication device and perform AI-related operations based on the AI ​​capability information of the first communication device (e.g., scheduling the first communication device to perform AI model inference and / or AI model updates). Similarly, before sending the sixth message, the second communication device may also receive a capability request message from the first communication device, enabling the requesting party to obtain the AI ​​capability information of the second communication device and perform AI-related operations based on the AI ​​capability information of the second communication device (e.g., scheduling the second communication device to perform AI model inference and / or AI model updates).

[0233] A third aspect of this application provides a communication device, comprising a processing unit and a transceiver unit; the processing unit is configured to acquire first information, the first information being used to indicate inference state information of a first model group, the first model group including a first model deployed on a first communication device and a second model deployed on a second communication device; wherein the input of the first model includes the output of the second model, or the input of the second model includes the output of the first model; the first communication device determines a second model group, the second model group being related to the inference state information of the first model group, the second model group including a third model deployed on the first communication device and a fourth model deployed on the second communication device; wherein the input of the third model includes the output of the fourth model, or the input of the fourth model includes the output of the third model; the transceiver unit is configured to transmit second information, the second information indicating the third model and / or the fourth model.

[0234] In the third aspect of this application, the constituent modules of the communication device can also be used to execute the steps performed in various possible implementations of the first aspect and achieve the corresponding technical effects. For details, please refer to the first aspect, which will not be repeated here.

[0235] A fourth aspect of this application provides a communication device including a transceiver unit; the transceiver unit is configured to receive second information indicating a third model and / or a fourth model included in a second model group, the third model being deployed in a first communication device and the fourth model being deployed in a second communication device; wherein the input of the third model includes the output of the fourth model, or the input of the fourth model includes the output of the third model; wherein the second model group is determined based on inference state information of a first model group; the first model group includes a first model deployed in the first communication device and a second model deployed in the second communication device; wherein the input of the first model includes the output of the second model, or the input of the second model includes the output of the first model.

[0236] In the fourth aspect of this application, the constituent modules of the communication device can also be used to perform the steps executed in various possible implementations of the second aspect and achieve the corresponding technical effects. For details, please refer to the second aspect, which will not be repeated here.

[0237] The fifth aspect of this application provides a communication device including at least one processor for executing computer programs or instructions to enable the device to implement the method described in any one of the first to second aspects and any possible implementation thereof.

[0238] Optionally, the at least one memory is coupled to a memory used to store computer programs or instructions.

[0239] Optionally, the communication device includes the memory.

[0240] The sixth aspect of this application provides a communication device including at least one logic circuit and an input / output interface; the logic circuit is used to perform the method described in any of the possible implementations of the first to second aspects described above.

[0241] The seventh aspect of this application provides a communication system, which includes the first communication device and the second communication device described above.

[0242] An eighth aspect of this application provides a computer-readable storage medium for storing one or more computer-executable instructions, which, when executed by a processor, perform the method as described in any possible implementation of any of the first to second aspects described above.

[0243] The ninth aspect of this application provides a computer program product (or computer program) that, when executed by a processor, performs the method described in any possible implementation of any of the first to second aspects described above.

[0244] The tenth aspect of this application provides a chip or chip system including at least one processor for supporting a communication device in implementing the methods described in any possible implementation of any of the first to second aspects. For example, the chip may be a baseband chip, a modem chip, a system-on-a-chip (SoC) chip containing a modem core, a system-in-package (SIP) chip, or a communication module, etc.

[0245] In one possible design, the chip or chip system may further include a memory for storing program instructions and data necessary for the communication device. The chip system may be composed of chips or may include chips and other discrete devices. Optionally, the chip system may also include interface circuitry that provides program instructions and / or data to the at least one processor.

[0246] The technical effects of any of the design methods in aspects three through ten can be found in the technical effects of the different design methods in aspects one through two above, and will not be repeated here. Attached Figure Description

[0247] Figures 1a to 1c are schematic diagrams of the communication system provided in this application;

[0248] Figures 2a to 2i are schematic diagrams of the AI ​​processing involved in this application;

[0249] Figure 3 is an interactive schematic diagram of the communication method provided in this application;

[0250] Figures 4a to 4i are some schematic diagrams showing the application of the communication method provided in this application;

[0251] Figures 5 to 9 are schematic diagrams of the communication device provided in this application. Detailed Implementation

[0252] First, some terms used in the embodiments of this application will be explained to facilitate understanding by those skilled in the art.

[0253] (1) Terminal device: can be a wireless terminal device that can receive network device scheduling and instruction information. The wireless terminal device can be a device that provides voice and / or data connectivity to the user, or a handheld device with wireless connection function, or other processing device connected to a wireless modem.

[0254] Terminal devices can communicate with one or more core networks or the Internet via a radio access network (RAN). Terminal devices can be mobile terminal devices, such as mobile phones (or "cellular" phones), computers, and data cards. For example, they can be portable, pocket-sized, handheld, computer-embedded, or vehicle-mounted mobile devices that exchange voice and / or data with the RAN. Examples include personal communication service (PCS) phones, cordless phones, session initiation protocol (SIP) phones, wireless local loop (WLL) stations, personal digital assistants (PDAs), tablets, and computers with wireless transceiver capabilities. Wireless terminal equipment can also be referred to as a system, subscriber unit, subscriber station, mobile station (MS), remote station, access point (AP), remote terminal, access terminal, user terminal, user agent, subscriber station (SS), customer premises equipment (CPE), terminal, user equipment (UE), mobile terminal (MT), etc.

[0255] By way of example and not limitation, in this embodiment, the terminal device can also be a wearable device. Wearable devices, also known as wearable smart devices or smart wearable devices, are a general term for devices that utilize wearable technology to intelligently design and develop everyday wearables, such as glasses, gloves, watches, clothing, and shoes. Wearable devices are portable devices that are worn directly on the body or integrated into the user's clothing or accessories. Wearable devices are not merely hardware devices, but also achieve powerful functions through software support, data interaction, and cloud interaction. Broadly speaking, wearable smart devices include those that are feature-rich, large in size, and can achieve complete or partial functions without relying on a smartphone, such as smartwatches or smart glasses, as well as those that focus on a specific type of application function and require the use of other devices such as smartphones, such as various smart bracelets, smart helmets, and smart jewelry for vital sign monitoring.

[0256] Terminals can also be drones, robots, devices in device-to-device (D2D) communication, vehicles to everything (V2X) communication, virtual reality (VR) terminal devices, augmented reality (AR) terminal devices, wireless terminals in industrial control, wireless terminals in self-driving, wireless terminals in telemedicine or telehealth services, wireless terminals in smart grids, wireless terminals in transportation safety, wireless terminals in smart cities, wireless terminals in smart homes, etc.

[0257] Furthermore, the terminal device can also be a terminal device for a communication system evolved from the fifth generation (5G) communication system (such as 5G Advanced or future communication systems). For example, the form and function of the communication terminal can be further expanded, including but not limited to vehicles, cellular network terminals (integrating satellite terminal functions), drones, and Internet of Things (IoT) devices.

[0258] In this embodiment, the terminal device can also obtain artificial intelligence (AI) services provided by the network device. Optionally, the terminal device can also have AI processing capabilities.

[0259] (2) Network equipment: This can be equipment within a wireless network, such as RAN nodes (or devices) that connect terminal devices to the wireless network. Examples of RAN equipment currently include: base stations, evolved NodeBs (eNodeBs), gNBs (gNodeBs) in 5G communication systems, transmission reception points (TRPs), evolved Node Bs (eNBs), radio network controllers (RNCs), Node Bs (NBs), home base stations (e.g., home evolved Node Bs, or home Node Bs (HNBs), base band units (BBUs), or wireless fidelity (Wi-Fi) access points (APs). Additionally, in a network architecture, network equipment may include central unit (CU) nodes, distributed unit (DU) nodes, or RAN equipment comprising both CU and DU nodes.

[0260] Optionally, RAN nodes can also be macro base stations, micro base stations, indoor stations, relay nodes, donor nodes, or radio controllers in cloud radio access network (CRAN) scenarios. RAN nodes can also be servers, wearable devices, vehicles, or in-vehicle equipment. For example, the access network equipment in V2X technology can be a roadside unit (RSU).

[0261] In another possible scenario, multiple RAN nodes collaborate to assist the terminal in achieving wireless access, with different RAN nodes each implementing some of the base station's functions. For example, RAN nodes can be CUs, DUs, CUs (control plane, CP), CUs (user plane, UP), or radio units (RUs). CUs and DUs can be configured separately or included in the same network element, such as a baseband unit (BBU). RUs can be included in radio equipment or radio units, such as remote radio units (RRUs), active antenna units (AAUs), radio heads (RHs), or remote radio heads (RRHs).

[0262] In different systems, CU (or CU-CP and CU-UP), DU, or RU may have different names, but those skilled in the art will understand their meaning. For example, in an open access network (open RAN, O-RAN, or ORAN) system, CU can also be called O-CU (open CU), DU can also be called O-DU, CU-CP can also be called O-CU-CP, CU-UP can also be called O-CU-UP, and RU can also be called O-RU. For ease of description, this application uses CU, CU-CP, CU-UP, DU, and RU as examples. Any of the units among CU (or CU-CP, CU-UP), DU, and RU in this application can be implemented through software modules, hardware modules, or a combination of software modules and hardware modules.

[0263] Communication between access network devices and terminal devices follows a specific protocol layer structure. This protocol layer may include a control plane protocol layer and a user plane protocol layer. The control plane protocol layer may include at least one of the following: radio resource control (RRC) layer, packet data convergence protocol (PDCP) layer, radio link control (RLC) layer, media access control (MAC) layer, or physical (PHY) layer, etc. The user plane protocol layer may include at least one of the following: service data adaptation protocol (SDAP) layer, PDCP layer, RLC layer, MAC layer, or physical layer, etc.

[0264] The correspondence between network elements and their achievable protocol layer functions in the ORAN system can be found in Table 1 below.

[0265] Table 1

[0266] Network devices can be other devices that provide wireless communication functions for terminal devices. The embodiments of this application do not limit the specific technology or form of the network device. For ease of description, the embodiments of this application are not limited.

[0267] Network equipment may also include core network equipment, such as the Mobility Management Entity (MME), Home Subscriber Server (HSS), Serving Gateway (S-GW), Policy and Charging Rules Function (PCRF), and Public Data Network Gateway (PDN gateway or P-GW) in 4th generation (4G) networks; and access and mobility management function (AMF), user plane function (UPF), or session management function (SMF) in 5G networks. Furthermore, this core network equipment may also include other core network equipment in 5G networks and future networks.

[0268] In this embodiment of the application, the network device may also have network nodes with AI capabilities, which can provide AI services to terminals or other network devices. For example, it may be an AI node, computing node, RAN node with AI capabilities, or core network element with AI capabilities on the network side (access network or core network).

[0269] In this application embodiment, the device for implementing the function of the network device can be the network device itself, or it can be a device capable of supporting the network device in implementing the function, such as a chip system. This device can be disposed within the network device. In the technical solutions provided in this application embodiment, the example of a network device being used to implement the function of the network device is used to describe the technical solutions provided in this application embodiment.

[0270] (3) Configuration and Pre-configuration: In this application, both configuration and pre-configuration are used. Configuration refers to the network device / server sending configuration information or parameter values ​​to the terminal via messages or signaling, so that the terminal can determine communication parameters or resources for transmission based on these values ​​or information. Pre-configuration is similar to configuration; it can be parameter information or parameter values ​​pre-negotiated between the network device / server and the terminal device, parameter information or parameter values ​​specified by standard protocols for use by the base station / network device or terminal device, or parameter information or parameter values ​​pre-stored in the base station / server or terminal device. This application does not limit this.

[0271] Furthermore, these values ​​and parameters can be changed or updated.

[0272] (4) The terms "system" and "network" in the embodiments of this application can be used interchangeably. "Multiple" refers to two or more. "And / or" describes the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist simultaneously, or B exists alone, where A and B can be singular or plural. The character " / " generally indicates that the related objects before and after are in an "or" relationship. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, "at least one of A, B and C" includes A, B, C, AB, AC, BC or ABC. And, unless otherwise specified, the ordinal numbers such as "first" and "second" mentioned in the embodiments of this application are used to distinguish multiple objects and are not used to limit the order, sequence, priority or importance of multiple objects.

[0273] (5) In the embodiments of this application, "send" and "receive" indicate the direction of signal transmission. For example, "send information to XX" can be understood as the destination of the information being XX, which may include sending directly through the air interface or sending indirectly through the air interface by other units or modules. "Receive information from YY" can be understood as the source of the information being YY, which may include receiving directly from YY through the air interface or receiving indirectly from YY through the air interface by other units or modules. "Send" can also be understood as the "output" of the chip interface, and "receive" can also be understood as the "input" of the chip interface.

[0274] In the embodiments of this application, sending and receiving can be performed between devices, such as between a network device and a terminal device, or they can be performed within a device, such as between components, modules, chips, software modules, or hardware modules within the device via a bus, wiring, or interface.

[0275] Optionally, the information may undergo necessary processing, such as encoding or modulation, between the source and destination, but the destination can still understand the valid information from the source. Similar statements in this application can be understood in a similar way and will not be elaborated further.

[0276] (6) In the embodiments of this application, "instruction" may include direct instruction and indirect instruction, as well as explicit instruction and implicit instruction. The information indicated by a certain piece of information (as described below, the instruction information) is called the information to be instructed. In the specific implementation process, there are many ways to indicate the information to be instructed, such as, but not limited to, directly indicating the information to be instructed, such as the information to be instructed itself or its index. It can also indirectly indicate the information to be instructed by indicating other information, where there is an association between the other information and the information to be instructed; or it can only indicate a part of the information to be instructed, while the other parts of the information to be instructed are known or pre-agreed upon. For example, the instruction of specific information can be achieved by using a pre-agreed (e.g., protocol predefined) arrangement order of various information, thereby reducing the instruction overhead to a certain extent. This application does not limit the specific method of instruction. Optionally, for the sender of the instruction information, the instruction information can be used to indicate the information to be instructed; for the receiver of the instruction information, the instruction information can be used to determine the information to be instructed.

[0277] In this application, unless otherwise specified, the same or similar parts between the various embodiments can be referred to each other. In the various embodiments of this application, and the various methods / designs / implementations within each embodiment, unless otherwise specified or logically conflicting, the terminology and / or descriptions between different embodiments and between the various methods / designs / implementations within each embodiment are consistent and can be mutually referenced. The technical features in different embodiments and the various methods / designs / implementations within each embodiment can be combined to form new embodiments, methods, or implementations based on their inherent logical relationships. The following descriptions of the embodiments of this application do not constitute a limitation on the scope of protection of this application.

[0278] This application can be applied to long-term evolution (LTE) systems, new radio (NR) systems, or future communication systems beyond 5G. These communication systems include at least one network device and / or at least one terminal device.

[0279] Please refer to Figure 1a, which is a schematic diagram of the architecture of the communication system 1000 used in the embodiments of this application. As shown in Figure 1a, the communication system may include a Radio Access Network (RAN) 100. Optionally, the communication system 1000 may also include a core network 200 and an Internet 300. The RAN 100 includes at least one RAN node (110a and 110b in Figure 1a, collectively referred to as 110) and at least one terminal (120a-120j in Figure 1a, collectively referred to as 120). The RAN 100 may also include other RAN nodes, such as wireless relay devices and / or wireless backhaul devices (not shown in Figure 1a). The terminal 120 is wirelessly connected to the RAN node 110, and the RAN node 110 is wirelessly or wired connected to the core network 200. The core network equipment in the core network 200 and the RAN node 110 in the RAN 100 may be independent and different physical devices, or they may be the same physical device integrating the logical functions of the core network equipment and the logical functions of the RAN node. Terminals can be connected to each other, as can RAN nodes, via wired or wireless means.

[0280] Taking the communication system shown in Figure 1a as an example, in addition to performing communication-related services, different devices (including network devices and network devices, network devices and terminal devices, and / or terminal devices and terminal devices) may also perform AI-related services.

[0281] As shown in Figure 1b, taking a network device as a base station as an example, the base station can perform communication-related services and AI-related services with one or more terminal devices, and different terminal devices can also perform communication-related services and AI-related services.

[0282] As shown in Figure 1c, taking terminal devices including televisions and mobile phones as an example, televisions and mobile phones can also perform communication-related services and AI-related services.

[0283] The technical solutions provided in this application can be applied to wireless communication systems (such as the systems shown in Figures 1a, 1b, or 1c). For example, AI network elements can be introduced into the communication system provided in this application to realize some or all AI-related operations. AI network elements can also be called AI nodes, AI devices, AI entities, AI modules, AI models, or AI units, etc. The AI ​​network element can be built into a network element within the communication system. For example, the AI ​​network element can be an AI module built into: access network equipment, core network equipment, cloud server, or operation, administration, and maintenance (OAM) to realize AI-related functions. The OAM can act as the network management system for the core network equipment and / or the access network equipment. Alternatively, the AI ​​network element can also be an independently set network element in the communication system. Optionally, the terminal or its built-in chip can also include an AI entity to realize AI-related functions.

[0284] Optionally, in communication systems, AI application cases may include, but are not limited to: channel state information (CSI) feedback enhancement, beam management enhancement, positioning accuracy enhancement, network energy saving, load balancing, and mobility optimization. These will be explained below.

[0285] 1. Enhanced CSI feedback

[0286] Channel quality information (CSI) is the channel attribute of a communication link, reported by the terminal device to the network device. By reporting this information, the terminal device can select an appropriate modulation and coding scheme (MCS) to adapt to changing wireless channels. For example, the terminal device might perform channel estimation based on the received channel state information-reference signal (CSI-RS) and then feed back the CSI-RS to the network device. This feedback CSI-RS information serves as input to the network device's model, enabling AI model training. Applying AI to CSI feedback enhancement can reduce overhead, improve accuracy, and enhance predictive capabilities.

[0287] CSI-RS feedback enhancement may include at least one sub-function, such as: CSI compression, CSI prediction, and CSI-RS configuration signaling reduction. CSI compression may further include CSI compression in at least one domain: spatial, time, and frequency.

[0288] 2. Enhanced Beam Management

[0289] Enhanced beam management primarily aims to discover the strongest transmit / receive beam pairs. AI-based sparse beam prediction can improve accuracy. This can be achieved through both network-side and terminal-side AI sparse beam prediction, based on AI training and inference. Taking terminal-side AI sparse beam prediction as an example, the pre-trained AI model on the terminal device can be provided by the network or pre-stored on the terminal device. During training, the network device scans all possible beams and then provides the transmit beam pattern to the terminal device. Once training is complete, the network device only needs to scan a small subset of beams, and the terminal device then feeds back the inference results to the network device. AI-based beam management can achieve beam prediction in, for example, the temporal and / or spatial domains, reducing overhead and latency and improving beam selection accuracy.

[0290] Beam management enhancements may include at least one sub-function, such as beam scan matrix prediction and / or optimal beam prediction.

[0291] 3. Enhanced positioning accuracy

[0292] In line-of-sight (LOS) or non-line-of-sight (NLOS) scenarios, AI-based positioning can improve positioning accuracy with a smaller number of TRP antennas. Positioning enhancement can include at least one sub-function, such as: positioning enhancement based on access network devices, positioning enhancement based on positioning management function network elements, and positioning enhancement based on terminal devices.

[0293] 4. Network energy saving

[0294] Network energy conservation can be achieved through cell activation / deactivation, load reduction, coverage improvement, or other RAN setting adjustments. AI technology can be used to optimize energy-saving decisions by leveraging data collected in the RAN. AI algorithms can predict energy efficiency and load status for the next cycle, which can be used to assist in cell activation / deactivation decisions to save energy. Based on the predicted load, the system can dynamically configure energy-saving strategies to maintain a balance between system performance and energy efficiency, and reduce energy consumption.

[0295] 5. Load balancing

[0296] Load balancing can distribute the load evenly between cells and across different areas within a cell, or transfer some traffic from congested cells, or offload users across a single cell, carrier, or access standard, thereby improving network performance. Using AI models to enhance load balancing performance—such as inputting various measurements and feedback from terminal devices and network nodes, as well as historical data—can provide a higher quality user experience and increase system capacity.

[0297] 6. Mobility Management

[0298] Mobility management is a solution that ensures service continuity during terminal device mobility by minimizing dropped calls, radio link failures (RLFs), unnecessary handovers, and ping-pong effects. AI can enhance mobility management by, for example, reducing the probability of unexpected events, predicting at least one of the terminal device's location, mobility, or performance, and traffic redirection.

[0299] Optionally, the definitions of the above technical terms are merely illustrative. For example, as technology continues to develop, the scope of the above definitions may also change, and the embodiments of this application are not limited thereto.

[0300] For example, an AI function may include multiple AI sub-functions.

[0301] Optionally, AI application cases are also called AI application scenarios or AI functions.

[0302] As described above regarding AI application examples, AI can be widely used to improve network performance in areas such as CSI feedback enhancement, beam management, positioning accuracy enhancement, energy saving, mobility enhancement, and load balancing. AI models can typically be deployed on the network side and / or the terminal device side. The training of AI models relies on the collection of training data, which can come from measurements and feedback from the terminal devices.

[0303] The following is a brief introduction to the concepts that may be involved in this application.

[0304] AI can endow machines with human-like intelligence, for example, allowing them to use computer hardware and software to simulate certain intelligent human behaviors. To achieve artificial intelligence, machine learning methods can be employed. In machine learning, machines learn (or train) a model using training data. This model represents the mapping between inputs and outputs. The learned model can be used for reasoning (or prediction), that is, it can be used to predict the output corresponding to a given input. This output can also be called the reasoning result (or prediction result).

[0305] Machine learning can include supervised learning, unsupervised learning, and reinforcement learning. Unsupervised learning can also be called learning without supervision.

[0306] Supervised learning, based on collected sample values ​​and labels, uses machine learning algorithms to learn the mapping relationship between sample values ​​and labels, and then expresses this learned mapping relationship using an AI model. The process of training the machine learning model is the process of learning this mapping relationship. During training, sample values ​​are input into the model to obtain the model's predicted values, and the model parameters are optimized by calculating the error between the model's predicted values ​​and the sample labels (ideal values). After the mapping relationship is learned, it can be used to predict new sample labels. The mapping relationship learned in supervised learning can include linear or non-linear mappings. Based on the type of label, the learning task can be divided into classification tasks and regression tasks.

[0307] Unsupervised learning relies on collected sample values ​​to discover inherent patterns within the samples themselves. One type of unsupervised learning algorithm uses the samples themselves as supervisory signals, meaning the model learns the mapping relationship from sample to sample; this is called self-supervised learning. During training, model parameters are optimized by calculating the error between the model's predictions and the samples themselves. Self-supervised learning can be used for signal compression and decompression recovery applications; common algorithms include autoencoders and generative adversarial networks.

[0308] Reinforcement learning, unlike supervised learning, is a type of algorithm that learns problem-solving strategies through interaction with the environment. Unlike supervised and unsupervised learning, reinforcement learning problems do not have explicit "correct" action labels. The algorithm needs to interact with the environment to obtain reward signals from the environment, and then adjust its decision actions to obtain a larger reward signal value. For example, in downlink power control, the reinforcement learning model adjusts the downlink transmission power of each user based on the total system throughput feedback from the wireless network, aiming to achieve a higher system throughput. The goal of reinforcement learning is also to learn the mapping relationship between the environment state and a better (e.g., optimal) decision action. However, because the label of the "correct action" cannot be obtained in advance, the network cannot be optimized by calculating the error between the action and the "correct action." Reinforcement learning training is achieved through iterative interaction with the environment.

[0309] Neural networks (NNs) are a specific model in machine learning techniques. According to the general approximation theorem, neural networks can theoretically approximate any continuous function, thus enabling them to learn arbitrary mappings. Traditional communication systems rely on extensive expert knowledge to design communication modules, while deep learning communication systems based on neural networks can automatically discover hidden pattern structures from large datasets, establish mapping relationships between data, and achieve performance superior to traditional modeling methods.

[0310] The idea behind neural networks comes from the neuronal structure of the brain. For example, each neuron performs a weighted summation of its input values ​​and outputs the result through an activation function.

[0311] Figure 2a shows a schematic diagram of a neuron structure. Assume the input to the neuron is x = [x0, x1, ..., x...]. n The weights corresponding to each input are w = [w0, w1, ..., w], respectively. n ], where n is a positive integer, w i and x i It can be any possible type, such as a decimal, an integer (e.g., 0, a positive integer, or a negative integer), or a complex number. i As x i The weights are used to assign weights to x. i Weighting is applied. The bias for the weighted sum of the input values ​​is, for example, b. Activation functions can take many forms. Suppose the activation function of a neuron is: y = f(z) = max(0, z), then the output of that neuron is: For example, if the activation function of a neuron is y = f(z) = z, then the output of that neuron is: Here, b can be any possible type, such as a decimal, an integer (e.g., 0, a positive integer, or a negative integer), or a complex number. The activation functions of different neurons in a neural network can be the same or different.

[0312] Furthermore, neural networks generally consist of multiple layers, each of which may include one or more neurons. Increasing the depth and / or width of a neural network can improve its expressive power, providing more powerful information extraction and abstract modeling capabilities for complex systems. The depth of a neural network can refer to the number of layers it includes, and the number of neurons in each layer can be called the width of that layer. In one implementation, a neural network includes an input layer and an output layer. The input layer processes the received input information through neurons and passes the processing result to the output layer, which then obtains the output of the neural network. In another implementation, a neural network includes an input layer, hidden layers, and an output layer. The input layer processes the received input information through neurons and passes the processing result to the hidden layer. The hidden layer calculates the received processing result and passes the calculation result to the output layer or the next adjacent hidden layer, ultimately obtaining the output of the neural network. A neural network may include one hidden layer or multiple sequentially connected hidden layers, without limitation.

[0313] Neural networks, for example, are deep neural networks (DNNs). Depending on how the network is constructed, DNNs can include feedforward neural networks (FNNs), convolutional neural networks (CNNs), and recurrent neural networks (RNNs).

[0314] Figure 2b is a schematic diagram of an FNN network. A characteristic of FNN networks is that neurons in adjacent layers are completely connected pairwise. This characteristic makes FNNs typically require a large amount of storage space, leading to high computational complexity.

[0315] CNNs are neural networks specifically designed to process data with a grid-like structure. For example, time-series data (e.g., discrete sampling along a time axis) and image data (e.g., two-dimensional discrete sampling) can both be considered grid-like data. CNNs do not use all the input information at once for computation; instead, they use a fixed-size window to extract a portion of the information for convolution operations, which significantly reduces the computational cost of model parameters. Furthermore, depending on the type of information extracted by the window (e.g., people and objects in an image represent different types of information), each window can use different convolution kernels, allowing CNNs to better extract features from the input data.

[0316] Recurrent Neural Networks (RNNs) are a type of neural network that utilizes feedback time-series information. The input to an RNN includes the current input value and its own output value from the previous time step. RNNs are suitable for acquiring temporally correlated sequence features, and are applicable to applications such as speech recognition and channel coding / decoding.

[0317] In the model training process described above, a loss function can be defined. The loss function describes the difference between the model's output value and the ideal target value. The loss function can be expressed in various forms, and there are no restrictions on its specific form. The model training process can be viewed as follows: by adjusting some or all of the model's parameters, the value of the loss function is made to be less than a threshold or to meet the target requirement.

[0318] A model can also be called an AI model, a rule, or other names. An AI model can be considered a specific method for implementing AI functions. An AI model represents the mapping relationship or function between the model's input and output. AI functions can include one or more of the following: data collection, model training (or model learning), model information dissemination, model inference (or model reasoning, inference, or prediction, etc.), model monitoring or model validation, or inference result publication, etc. AI functions can also be called AI (related) operations or AI-related functions.

[0319] The implementation process of the neural network will be described below with reference to the accompanying drawings.

[0320] 1. Fully connected neural network, also known as multilayer perceptron (MLP).

[0321] As shown in Figure 2c, an MLP consists of an input layer (left side), an output layer (right side), and multiple hidden layers (middle). Each layer of an MLP contains several nodes, called neurons. Neurons in adjacent layers are connected pairwise.

[0322] Optionally, considering neurons in two adjacent layers, the output h of the next layer's neurons is the weighted sum of all neurons x in the previous layer connected to it, processed by an activation function, and can be expressed as: h = f(wx + b).

[0323] Where w is the weight matrix, b is the bias vector, and f is the activation function.

[0324] Alternatively, the output of the neural network can be recursively expressed as: y = f z (w z f z-1 (…)+b z ).

[0325] Where z is the index of the neural network layer, z is greater than or equal to 1 and z is less than or equal to Z, where Z is the total number of layers in the neural network.

[0326] In other words, a neural network can be understood as a mapping from an input data set to an output data set. Neural networks are typically initialized randomly; the process of obtaining this mapping from random values ​​w and b using existing data is called training the neural network.

[0327] Optionally, the training method involves using a loss function to evaluate the output of the neural network.

[0328] As shown in Figure 2d, the error can be backpropagated, and the neural network parameters (including w and b) can be iteratively optimized using gradient descent until the output of the loss function reaches its minimum value, which is the "better point (e.g., the optimal point)" in Figure 2d. Optionally, the neural network parameters corresponding to the "better point (e.g., the optimal point)" in Figure 2d can be used as the neural network parameters in the trained AI model information.

[0329] Alternatively, the gradient descent process can be represented as:

[0330] Where θ represents the parameters to be optimized (including w and b), L is the loss function, and η is the learning rate, controlling the step size of gradient descent. This represents the differentiation operation. This indicates taking the derivative of θ with respect to L.

[0331] Alternatively, the backpropagation process can utilize the chain rule for partial derivatives.

[0332] As shown in Figure 2e, the gradient of the parameters in the previous layer can be recursively calculated from the gradient of the parameters in the next layer, and can be expressed as:

[0333] Among them, w ij Let s be the weight of the connection between node j and node i. i The weighted sum of the inputs at node i.

[0334] 2. Federated learning (FL).

[0335] The concept of federated learning effectively addresses the current challenges in the development of artificial intelligence. While fully protecting user data privacy and security, it enables various edge devices and central servers to collaborate efficiently to complete the model's learning task.

[0336] As shown in Figure 2f, the FL architecture is the most widely used training architecture in the current FL field, and the FedAvg algorithm is the basic algorithm of FL. The FedAvg algorithm flow is roughly as follows:

[0337] (1) Initialize the model to be trained at the center end. And broadcast it to all clients.

[0338] (2) In the t∈[1,T] round, the client k∈[1,K] is based on the local dataset. For the received global model Perform E epochs of training to obtain the local training results. This is then reported to the central node. In the example shown in Figure 2f, the local training results sent by distributed nodes n, k, and m are denoted as G, respectively. n G k G m .

[0339] (3) The central node collects local training results from all (or some) clients. Assume the set of clients uploading local models in round t is... The central server will use the number of samples from the corresponding client as weights to calculate the new global model. The specific update rule is as follows: Then the central end will send the latest version of the global model. The broadcast is sent to all clients for a new round of training.

[0340] (4) Repeat steps (2) and (3) until the model finally converges or the number of training rounds reaches the upper limit.

[0341] Optionally, in addition to reporting the local model, the client can also... It can also train local gradients The central node averages the local gradients reported by all clients and updates the global model based on this average gradient.

[0342] As can be seen in the FL framework, the dataset resides on distributed nodes (such as clients). These distributed nodes collect their local datasets, perform local training, and report the local results (model or gradients) to the central node. The central node itself may not have a dataset; it can be responsible for fusing the training results from the distributed nodes to obtain a global model, which is then distributed back to the distributed nodes.

[0343] 3. Decentralized learning.

[0344] Figure 2g illustrates a fully distributed system without a central node. The design goal f(x) of a decentralized learning system is generally the goal f of each node. i The mean of (x), i.e. Where n is the number of distributed nodes, and x is the parameter to be optimized; in machine learning, x is the parameter of the machine learning model (such as a neural network). Each node utilizes local data and its local target f. i (x) Calculate the local gradient Then it is sent to its communicatively reachable neighboring nodes. Upon receiving the gradient information from its neighbor, any node can update the parameters x of its local model according to the following formula:

[0345] in, This represents the parameters of the local model after the (k+1)th update (k is a natural number) in the i-th node. This represents the parameters of the local model for the i-th node after the k-th update (if k is 0, then it represents...). (where α is the parameter of the local model of the i-th node that is not involved in the update) k N represents the tuning coefficient. i It is the set of neighboring nodes of node i, |N i | represents the number of elements in the set of neighboring nodes of node i, that is, the number of neighboring nodes of node i. Through information interaction between nodes, the decentralized learning system will eventually learn a unified model.

[0346] The technical solutions provided in this application can be applied to wireless communication systems (such as the systems shown in Figure 1a, 1b, or 1c). In wireless communication systems, communication nodes generally possess signal transmission and reception capabilities as well as computing capabilities. To address the vision of future intelligent and inclusive accessibility, intelligence will further evolve at the wireless network architecture level, taking AI and / or machine learning (ML) as examples. In future networks, AI and / or ML will be further deeply integrated with wireless networks to achieve network-inherent intelligence, which also includes the intelligence of terminals.

[0347] For example, neural network-based AI and / or ML can be applied to wireless communication systems. Through data-driven training, accurate modeling and prediction of wireless data can be achieved. For instance, AI and / or ML can be used for various wireless tasks, such as time-frequency domain channel estimation, prediction, and beam prediction related to wireless channels. AI and / or ML models for specific tasks require data collection, model training, selection, switching, and retraining based on applicable conditions.

[0348] As an example, Figure 2h illustrates how AI models can be deployed on both sides of a wireless communication system. For instance, the AI ​​model can be deployed on both the UE (User Equipment) side and the NW (Network Wireless) side. In this case, the neural network models deployed on both sides form an autoencoder structure, with one side being the encoder and the other the decoder. For example, a modulation / demodulation AI model can be deployed on both the transmitter and receiver sides. Specifically, during downlink transmission, the encoder model (for modulation) is deployed on the NW side, and the decoder model (for demodulation) is deployed on the UE side; during uplink transmission, the encoder model (for modulation) is deployed on the UE side, and the decoder model (for demodulation) is deployed on the NW side. Similarly, a CSI (Compression Feedback) model can be deployed on both the UE and NW sides, with the encoder model (for CSI compression) deployed on the UE side and the decoder model (for CSI reconstruction) deployed on the NW side.

[0349] As an example, as shown in Figure 2i, the current 3GPP discussion provides a functional framework for using AI and / or ML to improve air interface performance, including data collection, model training, management, inference, and model storage. Optionally, these functions may interact with each other. For example, the data collection module can provide data to the model training module, management module, and inference module; the model training and management modules can provide models to the model storage module; and the model storage module can provide models for inference.

[0350] However, the above process only provides a framework of functional description. For communication devices, there is still no specific solution on how to manage models (especially models deployed on both sides or multiple sides) (including model usage and / or updates).

[0351] To address the aforementioned problems, this application provides a communication method and related apparatus, which will be described in detail below with reference to the accompanying drawings.

[0352] Please refer to Figure 3, which is a schematic diagram of an implementation of the communication method provided in this application. The method includes the following steps.

[0353] Optionally, in the following text, Figure 3 illustrates the method using a first communication device and other communication devices (such as a second communication device) as examples of the execution subjects of this interaction illustration, but this application does not limit the execution subjects of this interaction illustration. For example, the communication device can be a communication device (such as a terminal device or a network device), or a chip, baseband chip, modem chip, SoC chip (such as an SoC chip containing a modem core), SIP chip, communication module, chip system, processor, logic module, or software in the communication device.

[0354] As an example, the first communication device can be a terminal device and the second communication device can be a network device.

[0355] As another example, the first communication device can be a network device, and the second communication device can be a terminal device.

[0356] As another example, both the first and second communication devices are network devices.

[0357] Optionally, the aforementioned network equipment may be access network equipment or ORAN equipment (including at least one of O-CU, O-DU, and O-RU).

[0358] As another example, both the first and second communication devices are terminal devices, meaning that the scheme shown in Figure 3 can be applied to side link communication scenarios.

[0359] S301. A first communication device acquires first information, which is used to indicate inference state information of a first model group, the first model group including a first model deployed on the first communication device and a second model deployed on a second communication device; wherein the input of the first model includes the output of the second model, or the input of the second model includes the output of the first model; the first communication device determines a second model group, the second model group being related to the inference state information of the first model group, the second model group including a third model deployed on the first communication device and a fourth model deployed on the second communication device; wherein the input of the third model includes the output of the fourth model, or the input of the fourth model includes the output of the third model. For example, the first model group is different from the second model group.

[0360] As an example, in step S301, the first communication device can locally acquire first information. For instance, the first communication device acquires the output of the first model and / or the second model, and determines the inference state information of the first model group based on the tag data and the output. That is, the inference state information of the first model group may include some or all of the model's performance (e.g., accuracy, processing latency, etc.). Alternatively, the first communication device performs communication based on the output of the first model and / or the second model, and determines the inference state information of the first model group based on the communication performance information of the communication process. That is, the inference state information of the first model group may include some or all of the communication performance of the model output in the communication application (e.g., bit error rate, transmission rate, block error rate, communication latency, etc.). Or, the first communication device determines the inference state information of the first model group based on the number of times the first model group is used for model inference, and / or the duration of model inference using the first model group.

[0361] As another example, in step S301, the first communication device can obtain the first information through step A. For example, after the second communication device obtains the inference state information of the first model group in the above manner, the second communication device sends the first information indicating the inference state information of the first model group to the first communication device through step A.

[0362] S302. The first communication device sends second information, and correspondingly, the second communication device receives the second information. The second information indicates the third model and / or the fourth model.

[0363] In this application, the model may include an AI model, a mathematical model, a neural network model, an AI neural network model, a machine learning model, or an AI processing model, etc. Accordingly, the model involved in this application can be replaced with an AI model, a mathematical model, a neural network model, an AI neural network model, a machine learning model, or an AI processing model, etc.

[0364] In this application, a model is deployed on a communication device (e.g., a first model is deployed on a first communication device, a second model is deployed on a second communication device, etc.), which may include: after the communication device obtains the model parameters of the model, it obtains, generates or constructs the model based on the model parameters of the model, and subsequently the communication device can adjust the model.

[0365] Optionally, model parameters may include one or more of the following: model hyperparameters, model dataset (including model input data and corresponding label data), model structure, model weights, and model parameters.

[0366] Optionally, a model group may contain two or more models that can be used to jointly complete the same AI task. For example, these models can collaborate to complete the same AI task. The processing of these models may be serial (or chained), that is, the output of any model may be part or all of the input of one or more other models, and / or, the input of any model may include the output of other models.

[0367] As an example of implementation, the output of the first model can be part or all of the input to the second model. After processing by the second model, the output of the second model is obtained. The execution result of the AI ​​task can include the output of the second model (optionally, it can also include the output of the first model).

[0368] As another implementation example, the output of the second model can be part or all of the input of the first model. After being processed by the first model, the output of the first model is obtained. The execution result of the AI ​​task can include the output of the first model (optionally, it can also include the output of the second model).

[0369] Optionally, when a model group is considered as a single model, the models contained within that model group can be understood as sub-models within that model. For example, when the first model group is considered as a single model, the first and second models contained within the aforementioned first model group can be understood as sub-models of that model. Similarly, the second model group, third model group, etc., involved in this application can also be considered as a single model, and correspondingly, this model can contain two or more sub-models, as detailed in the implementation of the first model group.

[0370] Optionally, a model group may include two or more models. For example, in addition to the first model and the second model, the first model group may also include other models, which may be deployed on other communication devices different from the first and second communication devices, without limitation here.

[0371] Optionally, different communication devices (e.g., the first communication device and the second communication device) may transmit wireless communication signals (e.g., the transmission and reception of configuration information of communication resources, the transmission and reception of reference signals, etc.).

[0372] Optionally, the model group involved in this application (e.g., the first model group, the second model group, the third model group, etc., hereinafter) can be used to manage the wireless communication signal (including at least one of processing, configuration, updating, and optimization). For example, the model may include a model for modulation and demodulation, a model for channel compression, a model for signal transmission and reception, etc. Alternatively, the model involved in this application may also be a model for other tasks, such as a model for image recognition, a model for natural language processing, a model for computer vision, etc.

[0373] Optionally, the inference state information of the first model group mentioned above includes at least one of the following: the performance of the first model group, the number of times the first model group is used for model inference, or the duration of model inference using the first model group, to improve the flexibility of the solution implementation. For example, the performance of the first model group includes model performance (such as one or more of inference accuracy or inference latency) and / or system performance (such as one or more of throughput, packet loss rate, latency, or fairness).

[0374] Optionally, the inference state information can be replaced with other descriptions, such as inference performance information, inference information, model state information, or model performance information.

[0375] Optionally, reasoning can be replaced with other descriptions, such as prediction, deduction, identification, or decision.

[0376] Based on the scheme shown in Figure 3, the first information obtained by the first communication device in step S301 is used to indicate the inference state information of the first model group. Furthermore, in step S302, the first communication device can determine the second model group, which is related to the inference state information of the first model group. The first model group includes a first model deployed on the first communication device and a second model deployed on the second communication device. The second model group includes a third model deployed on the first communication device and a fourth model deployed on the second communication device. Optionally, when model groups deployed on two or more communication devices jointly perform AI processing, the first communication device can update the first model group based on its inference state information to obtain the second model group. In this way, the first communication device can update the model groups using their inference state information to achieve model management of the model groups.

[0377] Furthermore, in the above scheme, the first communication device can update the first model group based on the inference state information of the first model group, which can improve the inference performance of the updated model group.

[0378] Furthermore, in the above scheme, the first communication device can update the first model group based on the inference state information of the first model group. Therefore, the above scheme can be applied to scenarios where distributed multi-node collaborative processing of the same AI task occurs, where the multi-nodes include at least the first communication device and the second communication device. In this scenario, the first communication device can update some or all of the models contained in the model group based on the inference state information of the model group, improving the efficiency of model management in multi-node collaborative processing scenarios.

[0379] Furthermore, in the above scheme, the second information sent by the first communication device indicates the third model and / or fourth model included in the updated second model group, so that the first communication device can indicate or register the second model group to the recipient of the second information, so that the recipient can subsequently schedule the second model group.

[0380] In one possible implementation, the second information sent by the first communication device in step S302 satisfies any of the following:

[0381] The first model is different from the third model, the second model is the same as the fourth model, and the second information indicates the third model;

[0382] The first model is the same as the third model, the second model is different from the fourth model, and the second information indicates the fourth model; or,

[0383] The first model is different from the third model, the second model is different from the fourth model, and the second information indicates the third model and the fourth model.

[0384] Specifically, the inference state information of the first model group can be used to update the first model and / or the second model contained in the first model group. Correspondingly, the second information sent by the first communication device can indicate the update result corresponding to the update, so that the recipient of the second information can obtain the update result.

[0385] For example, if the inference state information of the first model group can be used to update the first model but not the second model, the second information can indicate the updated third model.

[0386] For example, if the inference state information of the first model group can be used to update the second model but not to update the first model, the second information can indicate the updated fourth model.

[0387] For example, if the inference state information of the first model group can be used to update the first model and the second model, the second information can indicate the updated third model and the updated fourth model.

[0388] In one possible implementation, the second information sent by the first communication device in step S302 indicates the third model or the second information indicates the fourth model. The method shown in FIG3 further includes:

[0389] S303. The first communication device acquires third information, which is used to indicate the reasoning state information of the second model group.

[0390] Optionally, the inference state information of the second model group mentioned above includes at least one of the following: the performance of the second model group, the number of times the second model group was used for model inference, or the duration of model inference using the second model group, to improve the flexibility of the solution implementation. For example, the performance of the second model group includes model performance (such as one or more of inference accuracy or inference latency) and / or system performance (such as one or more of throughput, packet loss rate, latency, or fairness).

[0391] As an example, in step S303, the first communication device can locally acquire third information. For instance, the first communication device acquires the output of the third model and / or the fourth model, and determines the inference state information of the second model group based on the tag data and the output. That is, the inference state information of the second model group may include the model's performance (e.g., accuracy, processing latency, etc.). Alternatively, the first communication device performs communication based on the output of the third model and / or the fourth model, and determines the inference state information of the second model group based on the communication performance information of the communication process. That is, the inference state information of the second model group may include part or all of the communication performance of the model output in the communication application (e.g., bit error rate, transmission rate, block error rate, communication latency, etc.). Or, the first communication device determines the inference state information of the second model group based on the number of times the second model group is used for model inference, and / or the duration of model inference using the second model group.

[0392] As another example, in step S303, the first communication device can obtain the third information through step B. For instance, after the second communication device obtains the inference state information of the second model group in the above manner, the second communication device sends the third information indicating the inference state information of the second model group to the first communication device through step B.

[0393] After step S303, the first communication device can determine a third model group, which is related to the inference state information of the second model group. The third model group includes a fifth model deployed on the first communication device and a sixth model deployed on the second communication device. The input of the fifth model includes the output of the sixth model, or the input of the sixth model includes the output of the fifth model.

[0394] S304. The first communication device sends a fourth message, and correspondingly, the second communication device receives the fourth message. The fourth message indicates the fifth model and the sixth model.

[0395] Specifically, when the second information indicates a third or fourth model, the first communication device updates one of the first and second models based on the inference state information of the first model group. In this case, the first communication device can also update the second model group based on the inference state information of the second model group to obtain the third model group. Thus, after the first communication device can update part of the models based on the inference state information of the first model group to obtain the second model group, the first communication device can further update it based on the inference state information of the second model group to obtain the third model group. Model management can be achieved through two or more model group update processes, so as to improve model processing efficiency through the updated model groups.

[0396] Furthermore, in the above scheme, the fourth information sent by the first communication device indicates the fifth model and / or the sixth model included in the updated third model group, enabling the first communication device to indicate or register the third model group to the recipient of the fourth information, so that the recipient can subsequently schedule the third model group.

[0397] Optionally, the third model group can be determined based on the second model group. The third model group can be determined in a variety of ways, which will be described below with some examples.

[0398] In Example 1, the third model group is determined through model switching.

[0399] In one possible implementation of Example 1, if the first condition is satisfied, then any of the following conditions must be met:

[0400] The fifth model is obtained by switching the third model, and the fourth model is the same as the sixth model.

[0401] The sixth model is obtained by switching the fourth model, and the third model is the same as the fifth model; or

[0402] The fifth model is obtained by switching the third model, and the sixth model is obtained by switching the fourth model.

[0403] Therefore, when the first condition is met, the first communication device can update the second model group to the third model group by switching models, thereby updating the model group.

[0404] Optionally, a communication device may support one or more models, meaning the communication device can deploy the one or more models and support inference based on the one or more models. Similarly, two or more communication devices may support one or more model groups, meaning the two or more communication devices can each deploy the one or more models and support joint inference based on their respective deployed models.

[0405] In the above process, the model switching process may include: when two or more communication devices support one or more model groups, one communication device switches from supporting one model to supporting another model, and / or, another communication device switches from supporting one model to supporting another model. Before the switch, the two communication devices can perform joint inference using their respective models; after the switch, the two communication devices can also perform joint inference using their respective models.

[0406] As an example, the first communication device supports one or more models including models A1 and B1; the second communication device supports one or more models including model A2; and models A1 and A2 support joint inference, as do models B1 and A2. Accordingly, models A1 and A2 can be considered different models within the same model group, and models B1 and A2 can be considered different models within the same model group. Accordingly, in the above scheme, when the fifth model is obtained by switching the third model, and the fourth and sixth models are the same, the third model can be model A1, the fifth model can be model B1, and the fourth and sixth models can be model A2.

[0407] As another example, the first communication device supports one or more models including model A1; the second communication device supports one or more models including models A2 and B2; and models A1 and A2 support joint reasoning, as do models A1 and B2. Accordingly, models A1 and A2 can be considered different models within the same model group, and models A1 and B2 can be considered different models within the same model group. Accordingly, in the above scheme, when the sixth model is obtained by switching the fourth model, and the third and fifth models are the same, the third and fifth models can be model A1, the fourth model can be model A2, and the fourth and sixth models can be model B2.

[0408] As another example, the first communication device supports one or more models including model A1 and model B1; the second communication device supports one or more models including model A2 and model B2; and model A1 and model A2 support joint reasoning, as do model B1 and model B2. Accordingly, model A1 and model A2 can be considered different models within the same model group, and model B1 and model B2 can be considered different models within the same model group. Correspondingly, in the above scheme, when the fifth model is obtained by switching the third model and the sixth model is obtained by switching the fourth model, the third model can be model A1, the fourth model can be model A2, the fifth model can be model B1, and the sixth model can be model B2.

[0409] As an example, in Implementation Example 1, the first condition is associated with the inference state information of the second model group. Accordingly, the satisfaction of the first condition can be understood as: the inference state information of the second model group satisfies condition A.

[0410] As an example, in Implementation Example 1, the first condition is associated with the inference state information and inference configuration information of the second model group. Accordingly, the satisfaction of the first condition can be understood as: the inference state information of the second model group satisfies condition A and the inference configuration information satisfies condition B.

[0411] Optionally, the inference configuration information includes one or more model identifiers and / or inference aid information. For example, the inference configuration information includes one or more model identifiers (or identifiers of one or more model groups), and correspondingly, the inference configuration information indicates that the model with the identifier of the one or more model identifiers (or identifiers of one or more model groups) conforms to or matches the inference configuration. As another example, the inference configuration information includes inference aid information, and correspondingly, the inference configuration information indicates that the model's update, optimization, or iteration process needs to be based on the inference aid information.

[0412] For example, inference aid information can indicate the configuration involved in the model inference process. For instance, the inference aid information may include some or all of the following: scene information (e.g., inference scene is indoor, outdoor, idle, busy, movement speed, etc.), model information (e.g., model performance, model complexity, etc.), environmental information (e.g., urban, suburban, etc.), system information (e.g., antenna configuration, system bandwidth, carrier frequency, subcarrier spacing, adoption frequency, throughput, packet loss rate, latency, fairness, etc.), or one or more other information related to the inference configuration.

[0413] For example, the inference state information of the second model group satisfies condition A, which includes at least one of the following: the performance of the second model group is lower than or equal to a first threshold, the number of inferences of the second model group is greater than or equal to a second threshold, or the inference duration of the second model group is greater than or equal to a third threshold.

[0414] Optionally, the thresholds involved in this application (such as the first threshold, the second threshold, the third threshold, and other thresholds that may appear later) may be pre-configured or pre-defined, or may be configured by network devices or servers, and are not limited here.

[0415] For example, the inference configuration information satisfying condition B includes: at least one model among the models supported by the first communication device, excluding the third model (or excluding the first and third models), is adapted to the inference configuration information; and / or, at least one model among the models supported by the second communication device, excluding the fourth model (or excluding the second and fourth models), is adapted to the inference configuration information. Correspondingly, the inference configuration information not satisfying condition B includes: none of the models supported by the first communication device are adapted to the inference configuration information, excluding the third model (or excluding the first and third models); and none of the models supported by the second communication device are adapted to the inference configuration information, excluding the fourth model (or excluding the second and fourth models).

[0416] Taking the first communication device as an example, when the inference configuration information satisfies condition B, at least one model among the models supported by the first communication device is adapted to the inference configuration information; in other words, one or more other models supported by the first communication device, other than the third model (or other than the first model and the third model), can be adapted to the inference configuration information, that is, the first model can be switched to any of the other one or more models, and the inference process of any model can satisfy the inference configuration information.

[0417] Similarly, if the inference configuration information does not meet condition B, none of the models supported by the first communication device are compatible with the inference configuration information; in other words, one or more models supported by the first communication device, other than the third model (or other than the first model and the third model), are compatible with the inference configuration information, meaning that the first model cannot obtain a model that meets the inference configuration information through model switching.

[0418] In Example 2, the third model group was determined through model fine-tuning.

[0419] In one possible implementation of Example 2, if the second condition is satisfied, any of the following conditions must be met:

[0420] The fifth model is obtained by fine-tuning the third model, and the fourth model is the same as the sixth model.

[0421] The sixth model is obtained by fine-tuning the fourth model, and the third model is the same as the fifth model; or

[0422] The fifth model is obtained by fine-tuning the third model, and the sixth model is obtained by fine-tuning the fourth model.

[0423] Therefore, when the second condition is met, the first communication device can update the second model group to the third model group by fine-tuning the model, thereby realizing the update of the model group.

[0424] Optionally, the communication device can support model fine-tuning of one or more models. That is, after deploying one or more models, the communication device can fine-tune the one or more models to obtain fine-tuned models, and perform model inference based on the fine-tuned models. Similarly, two or more communication devices can support model fine-tuning of one or more model groups. That is, after each of the two or more communication devices deploys one or more models, each of them can fine-tune the one or more models they have deployed to obtain fine-tuned models, and support joint inference based on their respective fine-tuned models.

[0425] In the above process, the model fine-tuning process may include: when two or more communication devices support one or more model groups, one communication device fine-tunes a model from one of the supported models to obtain a fine-tuned model, and / or, another communication device fine-tunes a model from one of the supported models to obtain a fine-tuned model. Before fine-tuning, the two communication devices can perform joint inference using their respective models; after fine-tuning, the two communication devices can also perform joint inference using their respective models.

[0426] For example, a communication device fine-tuning a model may include: the communication device fine-tuning, adjusting, or updating some parameters of the model, where the number of parameters in this part is small (e.g., the number of parameters in this part is below or equal to a threshold, or the ratio of the number of parameters in this part to the total number of parameters in the model is below or equal to a threshold) and / or the number of adjustment rounds is relatively small (e.g., the number of adjustment rounds is less than or equal to a threshold). Optionally, model fine-tuning can be implemented online; for example, during model fine-tuning, the communication device can fine-tune, adjust, or update some parameters of the model online.

[0427] As an example, in implementing Example 2, the second condition is associated with the inference state information of the second model group. Accordingly, the second condition being satisfied can be understood as: the inference state information of the second model group satisfies condition C.

[0428] As an example, in Implementation Example 2, the first condition is associated with the inference state information and inference configuration information of the second model group. Correspondingly, the second condition being satisfied can be understood as: the inference state information of the second model group satisfies condition C, and the inference configuration information does not satisfy condition B.

[0429] For example, the inference state information of the second model group satisfies condition C, including at least one of the following: the performance of the second model group is lower than or equal to the fourth threshold, the number of inferences of the second model group is greater than or equal to the fifth threshold, or the inference duration of the second model group is greater than or equal to the sixth threshold.

[0430] Optionally, the fourth threshold may be less than or equal to the first threshold.

[0431] Optionally, the fifth threshold may be greater than or equal to the second threshold.

[0432] Optionally, the sixth threshold is greater than or equal to the third threshold.

[0433] In one possible implementation of Example 2, the method shown in Figure 3 further includes: the first communication device sending a first request message, the first request message being used to request first data, the first data being used for model fine-tuning of the second model group; and the first communication device receiving or sending the first data. Thus, when the first communication device determines that model fine-tuning of the second model group is required, the first communication device can send the first request message, enabling the recipient of the first request message to receive or send the first data based on the first request message, so that the recipient of the first data can perform model fine-tuning of the second model group based on the first data.

[0434] Example 3: The third model group is determined through model training.

[0435] In one possible implementation of Example 3, if the third condition is satisfied, then any of the following conditions must be met:

[0436] The fifth model is obtained by training the third model, and the fourth model is the same as the sixth model.

[0437] The sixth model is obtained by training the fourth model, and the third model is the same as the fifth model; or

[0438] The fifth model was obtained by training the third model, and the sixth model was obtained by training the fourth model.

[0439] Therefore, when the third condition is met, the first communication device can update the second model group to the third model group through model training, thereby realizing the update of the model group.

[0440] Optionally, the communication device can support model training of one or more models. That is, after deploying one or more models, the communication device can train the one or more models to obtain trained one or more models, and perform model inference based on the trained one or more models. Similarly, two or more communication devices can support model training of one or more model groups. That is, after each of the two or more communication devices deploys one or more models, each of them can train the one or more models they have deployed to obtain trained one or more models, and support joint inference based on their respective trained one or more models.

[0441] In the above process, the model training process may include: when two or more communication devices support one or more model groups, one communication device trains from one of the supported models to obtain a trained model, and / or, another communication device trains from one of the supported models to obtain a trained model. Before training, the two communication devices can perform joint inference using their respective models; after training, the two communication devices can also perform joint inference using their respective models.

[0442] For example, training a model using a communication device may include: training, adjusting, or updating some or all of the model's parameters, where the number of parameters is relatively large (e.g., the number of parameters is higher than or equal to a threshold, or the ratio of the number of parameters to the total number of parameters in the model is higher than or equal to a threshold) and / or the number of adjustment rounds is relatively large (e.g., the number of adjustment rounds is greater than or equal to a threshold). Optionally, model training may be implemented offline; for example, during model training, the communication device may train, adjust, or update some or all of the model's parameters offline.

[0443] As one example of implementing Example 3, the third condition is associated with the inference state information of the second model group. Accordingly, the satisfaction of the third condition can be understood as: the inference state information of the second model group satisfies condition D.

[0444] As another example of implementing Example 3, the first condition is associated with the inference state information and inference configuration information of the second model group. Accordingly, the second condition is satisfied, which can be understood as: the inference state information of the second model group satisfies condition D and the inference configuration information does not satisfy condition B.

[0445] For example, the inference state information of the second model group satisfies condition D, including at least one of the following: the performance of the second model group is lower than or equal to the seventh threshold, the number of inferences of the second model group is greater than or equal to the eighth threshold, or the inference duration of the second model group is greater than or equal to the ninth threshold.

[0446] Optionally, the seventh threshold may be less than or equal to the fourth threshold.

[0447] Optionally, the eighth threshold is greater than or equal to the fifth threshold.

[0448] Optionally, the ninth threshold is greater than or equal to the sixth threshold.

[0449] In one possible implementation of Example 3, the method shown in Figure 3 further includes: the first communication device sending a second request message, the second request message being used to request second data, the second data being used for model training of the second model group; and the first communication device receiving or sending the second data. Thus, when the first communication device determines that model training is to be performed on the second model group, the first communication device can send the second request message, enabling the recipient of the second request message to receive or send the second data based on the second request message, so that the recipient of the second data can perform model training of the second model group based on the second data.

[0450] Optionally, the second model group can be determined based on the first model group. The second model group can be determined in various ways, which will be described below with some examples.

[0451] In Example A, the second model group is determined by model switching.

[0452] In one possible implementation of Example A, if the fourth condition is satisfied, then any of the following conditions must be met:

[0453] The third model is obtained by switching the first model, and the second model is the same as the fourth model.

[0454] The fourth model is obtained by switching the second model, and the first model is the same as the third model; or

[0455] The third model is obtained by switching the first model, and the fourth model is obtained by switching the second model.

[0456] Therefore, when the fourth condition is met, the first communication device can update the first model group to the second model group by switching models, thereby updating the model group.

[0457] As an example of implementing Example A, the fourth condition is associated with the inference state information of the first model group. Accordingly, the satisfaction of the fourth condition can be understood as: the inference state information of the first model group satisfies condition E.

[0458] As another example of implementing Example A, the fourth condition is associated with the inference state information and inference configuration information of the first model group. Accordingly, the fourth condition being satisfied can be understood as: the inference state information of the first model group satisfies condition E and the inference configuration information satisfies condition F.

[0459] For example, the inference state information of the first model group satisfies condition E including at least one of the following: the performance of the first model group is lower than or equal to the tenth threshold, the number of inferences of the first model group is greater than or equal to the eleventh threshold, or the inference duration of the first model group is greater than or equal to the twelfth threshold.

[0460] Optionally, the first threshold is the same as the tenth threshold.

[0461] Optionally, the second threshold is the same as the eleventh threshold.

[0462] Optionally, the third threshold is the same as the twelfth threshold.

[0463] For example, the inference configuration information satisfying condition F includes: at least one model among the models supported by the first communication device, excluding the first model, is adapted to the inference configuration information; and / or, at least one model among the models supported by the second communication device, excluding the second model, is adapted to the inference configuration information. Correspondingly, the inference configuration information not satisfying condition F includes: none of the models supported by the first communication device, excluding the first model, are adapted to the inference configuration information; and none of the models supported by the second communication device, excluding the second model, are adapted to the inference configuration information.

[0464] Taking the first communication device as an example, when the inference configuration information satisfies condition F, at least one model among the models supported by the first communication device is adapted to the inference configuration information; in other words, one or more other models supported by the first communication device besides the first model can be adapted to the inference configuration information, that is, the first model can be switched to any of the other one or more models, and the inference process of any model can satisfy the inference configuration information.

[0465] Similarly, if the inference configuration information does not meet condition F, none of the models supported by the first communication device are compatible with the inference configuration information; in other words, none of the other models supported by the first communication device besides the first model are compatible with the inference configuration information, that is, the first model cannot obtain a model that meets the inference configuration information through model switching.

[0466] In Example B, the second model group is determined through model fine-tuning.

[0467] In one possible implementation of Example B, if the fifth condition is satisfied, then any of the following must be met:

[0468] The third model is obtained by fine-tuning the first model, and the second model is the same as the fourth model.

[0469] The fourth model is obtained by fine-tuning the second model, and the first model is the same as the third model; or

[0470] The third model is obtained by fine-tuning the first model, and the fourth model is obtained by fine-tuning the second model.

[0471] Therefore, when the fifth condition is met, the first communication device can update the first model group to the second model group by fine-tuning the model, thereby updating the model group.

[0472] As an example of implementing Example B, the fifth condition is associated with the inference state information of the first model group. Accordingly, the satisfaction of the fifth condition can be understood as: the inference state information of the first model group satisfies condition F.

[0473] As another example of implementing Example B, the fifth condition is associated with the inference state information and inference configuration information of the first model group. Accordingly, the satisfaction of the fifth condition can be understood as: the inference state information of the first model group satisfies condition G and the inference configuration information does not satisfy condition F.

[0474] For example, the inference state information of the first model group satisfies condition G, including at least one of the following: the performance of the first model group is lower than or equal to the thirteenth threshold, the number of inferences of the first model group is greater than or equal to the fourteenth threshold, or the inference duration of the first model group is greater than or equal to the fifteenth threshold.

[0475] Optionally, the thirteenth threshold is less than or equal to the tenth threshold.

[0476] Optionally, the fourteenth threshold is greater than or equal to the eleventh threshold.

[0477] Optionally, the fifteenth threshold is greater than or equal to the twelfth threshold.

[0478] In one possible implementation of Example B, the method shown in Figure 3 further includes: the first communication device sending a third request message, the third request message being used to request third data, the third data being used for model fine-tuning of the first model group; and the first communication device receiving or sending the third data. Thus, when the first communication device determines that model fine-tuning of the first model group is required, the first communication device can send the third request message, enabling the recipient of the third request message to receive or send the third data based on the third request message, so that the recipient of the third data can perform model fine-tuning of the first model group based on the third data.

[0479] In Example C, the second model group is determined through model training.

[0480] In one possible implementation of example C, if the sixth condition is satisfied, then any of the following conditions must be met:

[0481] The third model is obtained by training the first model, and the second model is the same as the fourth model.

[0482] The fourth model is obtained by training the second model, and the first model is the same as the third model; or

[0483] The third model is obtained by training the first model, and the fourth model is obtained by training the second model.

[0484] Therefore, when the sixth condition is met, the first communication device can update the first model group to the second model group through model training, thereby updating the model group.

[0485] As an example of implementation example C, the sixth condition is associated with the inference state information of the first model group. Accordingly, the satisfaction of the sixth condition can be understood as: the inference state information of the first model group satisfies condition H.

[0486] As another example of implementation example C, the sixth condition is associated with the inference state information and inference configuration information of the first model group. Accordingly, the sixth condition being satisfied can be understood as: the inference state information of the first model group satisfies condition H and the inference configuration information does not satisfy condition F.

[0487] For example, the inference state information of the first model group satisfies condition H, including at least one of the following: the performance of the first model group is lower than or equal to the sixteenth threshold, the number of inferences of the first model group is greater than or equal to the seventeenth threshold, or the inference duration of the first model group is greater than or equal to the eighteenth threshold.

[0488] Optionally, the sixteenth threshold may be less than or equal to the thirteenth threshold.

[0489] Optionally, the seventeenth threshold is greater than or equal to the fourteenth threshold.

[0490] Optionally, the eighteenth threshold is greater than or equal to the fifteenth threshold.

[0491] In one possible implementation of Example C, the method further includes: the first communication device sending a fourth request message, the fourth request message being used to request fourth data, the fourth data being used for model training of the first model group; and the first communication device receiving or sending the fourth data. Thus, when the first communication device determines that model training is to be performed on the first model group, the first communication device can send the fourth request message, enabling the recipient of the fourth request message to receive or send the fourth data based on the fourth request message, so that the recipient of the fourth data can perform model training of the first model group based on the fourth data.

[0492] In one possible implementation, the method shown in Figure 3 further includes:

[0493] Step C. The first communication device sends a fifth message, and correspondingly, the second communication device receives the fifth message. The fifth message indicates the AI ​​capability information of the first communication device; and / or,

[0494] Step D. The second communication device sends a sixth message, and correspondingly, the first communication device receives the sixth message, which indicates the AI ​​capability information of the second communication device; wherein the AI ​​capability information of the first communication device and / or the AI ​​capability information of the second communication device are used to determine the second model group.

[0495] Based on the above scheme, the first and second communication devices can also receive or send AI capability information, enabling the recipient of the AI ​​capability information to jointly execute AI tasks based on the AI ​​capabilities of the peer device. For example, either communication device can determine the models supported by the peer communication device based on its AI capability information, facilitating model scheduling for joint AI task execution. Furthermore, either communication device can determine the model update methods supported by the peer communication device based on its AI capability information, enabling the execution of one-sided or multi-sided model update processes. Additionally, either communication device can determine the model update range supported by the peer communication device based on its AI capability information, enabling the execution of single-sided model update processes or multi-sided model joint update processes.

[0496] Optionally, the AI ​​capability information of the first communication device indicates at least one of the following: supported one or more AI models, supported model update methods, whether the model deployed on the first communication device supports independent model updates, or whether the model deployed on the first communication device supports joint model updates with associated models; similarly, the AI ​​capability information of the second communication device indicates at least one of the following: supported one or more AI models, supported model update methods, whether the model deployed on the second communication device supports independent model updates, or whether the model deployed on the second communication device supports joint model updates with associated models; wherein, the supported model update methods include at least one of model switching, model fine-tuning, or model training.

[0497] Optionally, before sending the fifth message, the first communication device may also receive a capability request message, enabling the requesting party to obtain the AI ​​capability information of the first communication device and perform AI-related operations based on the AI ​​capability information (e.g., scheduling the first communication device to perform AI model inference and / or AI model updates). Similarly, before receiving the sixth message, the first communication device may also send a capability request message to the second communication device, enabling the first communication device to obtain the AI ​​capability information of the second communication device and perform AI-related operations based on the AI ​​capability information (e.g., scheduling the second communication device to perform AI model inference and / or AI model updates).

[0498] As can be seen from the above implementation process, the scheme involved in this application can be applied to a model group deployed by two or more communication devices for joint inference. Furthermore, any two communication devices can interact to achieve the process of using and updating the model group. The following description will use an example where the model group contains a dual-sided model deployed by two communication devices, and the dual-sided model's encoding and decoding model (denoted as the Codec model) consists of an encoder (Enc) model and a decoder (Dec) model, with further examples.

[0499] As shown in Figure 4a, the process of using the two-sided model includes the following steps.

[0500] Step 1. Capability Interaction.

[0501] As an example, as shown in Figure 4b, the communication devices deploying the model are the terminal equipment and the network equipment. The network equipment can query some or all of the models supported by the UE, the supported model update methods, and the supported model update range in the User Equipment Capability Enquiry. Among them, the supported models include supported functions, supported models, or models under a certain supported function; the supported model update methods may include model switching, model fine-tuning, and model training; the supported model update range includes updating only the decoder model (e.g., Dec-only mode, or Dec-only model update) and updating the complete model (e.g., Codec mode, or Codec model update).

[0502] For example, add the fields shown in Table 2 below to UECapabilityEnquiry.

[0503] Table 2

[0504] In addition, terminal devices can report the models they support, the supported model update methods, and the supported model update range in the User Equipment Capability Information (UECapabilityInformation). Example fields as shown in Table 3 are added to UECapabilityInformation.

[0505] Table 3

[0506] Step 2. Model pairing. For example, one side (e.g., network device) determines its own model based on scene information (indoor / outdoor, idle / busy time, etc.), model information (model performance / model complexity), system information (antenna configuration), etc., and completes model pairing with the other side (usually terminal device) through indicators such as model ID.

[0507] Step 3. Model Inference. For example, the terminal device and network device complete the model inference based on the pairing model completed in Step 2.

[0508] Step 4. Performance Monitoring. For example, monitor model performance (such as inference accuracy, inference latency, etc.) and system performance (one or more of throughput, packet loss rate, latency, or fairness) when using a one-sided model. Similarly, monitor model performance (such as inference accuracy, inference latency, etc.) and system performance (such as throughput, packet loss rate, latency, or fairness, etc.) when using a two-sided model.

[0509] Step 5. Model Update. For example, the network device initiates a model update when the model update conditions (performance degradation, counter or timer expiration, etc.) are met.

[0510] Step 6. Model Registration. For example, after the terminal device model has been fine-tuned or trained, the new model obtained from the fine-tuning or training is registered with the network device to obtain new model pairing information (such as model ID).

[0511] As can be seen from the above implementation examples 1 to 3, and implementation examples A to C, there may be three types of model update methods involved in this application, including model switching, model fine-tuning, and model training. These will be described below with more examples.

[0512] ① Model switching. For example, network devices infer configuration information based on scenario information, environmental information, system information, etc., and determine whether the network devices and terminal devices have stored a model that is suitable for the current conditions. If so, the model switching is performed directly.

[0513] ② Model fine-tuning. For example, if there is no model that matches the inference configuration information for network devices and terminal devices, and the model inference performance meets the fine-tuning conditions (such as model performance loss being less than a threshold, the counter using the model being less than a threshold, or the timer using the model being less than a threshold, etc.), then model fine-tuning is initiated.

[0514] ③ Model training. If no model exists on the network device and terminal device that is compatible with the inference configuration information, and the model inference performance meets the training conditions (such as model performance loss being greater than a threshold, the counter using the model being greater than a threshold, or the timer using the model being greater than a threshold, etc.), then model training will be started.

[0515] Optionally, the model update range indication can be done in a single-level or two-level manner.

[0516] As an example, in the implementation of the single-level approach, the network device can issue an instruction to the terminal device via RRC or DCI, instructing the terminal device to switch, fine-tune, or train the decoder model (denoted as Dec-only or Dec-only mode, hereinafter referred to as Dec-only) or to switch, fine-tune, or train the encoder and decoder models (denoted as Codec or Codec mode, hereinafter referred to as Codec).

[0517] As another example, in the implementation of the two-level approach, the Dec-only model update can be performed first, followed by the Codec model update. The following description, using the example of deploying a model group (including an encoder model and a decoder model) on both the terminal device and network device sides, and referring to the example shown in Figure 4c, includes the following steps.

[0518] Step 1. Capability Interaction / Model Pairing / Model Inference / Performance Monitoring. Refer to Figure 4a and related descriptions above.

[0519] Step 2. After determining that a Dec-only model update should be performed, the network device sends a Dec-only model update instruction to the terminal device.

[0520] Step 3. The terminal device and network device perform a Dec-only model update to obtain the updated decoder model.

[0521] Step 4. The terminal device and network device perform model inference based on the updated decoder and conduct performance monitoring to determine the performance information of the updated decoder.

[0522] Step 5. If the network device determines that the performance of the updated decoder is poor based on the updated decoder's performance information, the network device will determine to perform a codec model update and send a codec model update instruction to the terminal device.

[0523] Step 6. The terminal device and network device perform codec model updates to obtain the updated encoder and decoder models.

[0524] Optionally, if the terminal device model is fine-tuned or trained during the above model update process, resulting in a new terminal device model, the terminal device can register it with the network device to obtain model pairing information (such as model ID).

[0525] To facilitate understanding of the above solution, more examples will be used to describe it below.

[0526] Please refer to the example shown in Figure 4d, which takes the encoder deployed on the terminal device and the decoder model deployed on the network device as an example. In this example, model fine-tuning with the model update method of Dec-only is taken as an example.

[0527] Step 1. The network device sends a Dec-only tuning instruction to the terminal device.

[0528] Step 2. The terminal device performs Enc inference.

[0529] Step 3. The terminal device sends inference information (such as intermediate inference results and / or tags) to the network device.

[0530] Step 4. The network device performs Dec inference and / or updates.

[0531] Optionally, steps 1 to 4 can be executed in one or more rounds. The specific number of rounds can be pre-configured or predefined, or determined by the network device based on one or more parameters such as performance, power consumption, or latency.

[0532] As can be seen from the above process, for Dec-only model fine-tuning, when the decoder model is deployed on a network device, the network device can instruct the terminal device to perform Dec-only model fine-tuning via RRC or downlink control information (DCI). Specifically, the terminal device can collect data, perform encoder model inference, and send inference information (such as intermediate inference results and / or labels) to the network device. The network device will then continue inference with the received inference information, calculate the loss value based on the labels, and then calculate the gradient. The decoder model will be updated through backpropagation of the gradient.

[0533] Please refer to the example shown in Figure 4e, which takes the encoder deployed on a network device and the decoder model deployed on a terminal device as an example. In this example, model fine-tuning with the model update method of Dec-only is taken as an example.

[0534] Step 1. Analyze the network device scenario and / or data to determine the fine-tuning to be performed on Dec-only devices.

[0535] Step 2. The network device sends a Dec-only fine-tuning instruction to the terminal device.

[0536] Step 3. The network device performs Enc inference.

[0537] Step 4. The network device sends inference information (such as intermediate inference results and / or tags) to the terminal device.

[0538] Step 5. The terminal device performs Dec inference and / or update.

[0539] Optionally, steps 1 to 5 can be executed in one or more rounds. The specific number of rounds can be pre-configured or predefined, or determined by the network device based on one or more parameters such as performance, power consumption, or latency.

[0540] Step 6. After performing one or more rounds of model fine-tuning, the terminal device performs model registration with the network device, registering the fine-tuned model of the terminal device to the network device through the model registration process.

[0541] As described above, for Dec-only model fine-tuning, when the decoder model is deployed on the terminal device, the network device instructs the terminal device to perform Dec-only model fine-tuning via RRC or DCI. The network device collects data, performs encoder model inference, and sends inference information (such as intermediate inference results and / or labels) to the terminal device. The terminal device inputs the received inference information into the decoder model to continue inference, calculates the loss value based on the labels, and then calculates the gradient. The decoder model is updated through backpropagation of the gradient. Furthermore, after the decoder model deployed on the terminal device has completed fine-tuning, the terminal device can register the newly fine-tuned model with the network device in step 6. The network device assigns the new model ID to the terminal device for subsequent model pairing.

[0542] Optionally, before step 2, the network device can also perform scenario and data analysis and obtain the model ID to be assigned based on the analysis results, that is, step 1 is an optional step.

[0543] Please refer to the example shown in Figure 4f, which takes the encoder deployed on the terminal device and the decoder model deployed on the network device as an example. In this example, model fine-tuning with the model update method of Codec is taken as an example.

[0544] Step 1. Analyze the network device's execution scenario and / or data to determine the fine-tuning of the execution codec.

[0545] Step 2. The network device sends a Codec fine-tuning instruction to the terminal device.

[0546] Step 3. The terminal device performs Enc inference.

[0547] Step 4. The terminal device sends inference information (such as inference intermediate results and / or tags) to the network device.

[0548] Step 5. The network device performs Dec inference and / or updates and obtains intermediate gradients.

[0549] Step 6. The network device sends the intermediate gradient to the terminal device.

[0550] Step 7. The terminal device performs an Enc update.

[0551] Optionally, steps 1 to 7 can be executed in one or more rounds. The specific number of rounds can be pre-configured or predefined, or determined by the network device based on one or more parameters such as performance, power consumption, or latency.

[0552] Step 8. After performing one or more rounds of model fine-tuning, the terminal device performs model registration with the network device, registering the fine-tuned model of the terminal device to the network device through the model registration process.

[0553] As described above, for codec model fine-tuning, when the decoder model is deployed on a network device, the network device needs to instruct the terminal device to fine-tune the codec model via RRC or DCI. The terminal device collects data, performs encoder model inference, and sends inference information (such as intermediate inference results and / or labels) to the network device. The network device inputs the received intermediate inference results into the decoder model to continue inference, calculates the loss value based on the labels, and then calculates the gradient. The decoder model is updated through backpropagation of the gradient. The network device sends the backpropagated intermediate gradient to the terminal device so that the terminal device can update the encoder model. Furthermore, after the encoder model deployed on the terminal device has been fine-tuned, the terminal device can register the newly fine-tuned model with the network device in step 7. The network device assigns the new model ID to the terminal device for subsequent model pairing.

[0554] Optionally, before step 2, the network device can also perform scenario and data analysis and obtain the model ID to be assigned based on the analysis results, that is, step 1 is an optional step.

[0555] Please refer to the example shown in Figure 4g, which takes the encoder deployed on the terminal device and the decoder model deployed on the network device as an example. In this example, model fine-tuning with the model update method of Codec is taken as an example.

[0556] Step 1. Analyze the network device's execution scenario and / or data to determine the fine-tuning of the execution codec.

[0557] Step 2. The network device sends a Codec fine-tuning instruction to the terminal device.

[0558] Step 3. The network device performs Enc inference.

[0559] Step 4. The network device sends inference information (such as intermediate inference results and / or tags) to the terminal device.

[0560] Step 5. The terminal device performs Dec inference and / or updates, and obtains intermediate gradients.

[0561] Step 6. The terminal device sends the intermediate gradient to the network device.

[0562] Step 7. The network device performs an Enc update.

[0563] Optionally, steps 1 to 7 can be executed in one or more rounds. The specific number of rounds can be pre-configured or predefined, or determined by the network device based on one or more parameters such as performance, power consumption, or latency.

[0564] Step 8. After performing one or more rounds of model fine-tuning, the terminal device performs model registration with the network device, registering the fine-tuned model of the terminal device to the network device through the model registration process.

[0565] As described above, for codec model fine-tuning, when the decoder model is deployed on the terminal device, the network device instructs the terminal device to fine-tune the codec model via RRC or DCI. The network device collects data, performs encoder model inference, and sends inference information (such as intermediate inference results and / or labels) to the terminal device. The terminal device inputs the received intermediate inference results into the decoder model to continue inference, calculates the loss value based on the inference results and the received label information, and then calculates the gradient. The decoder model is updated through backpropagation of the gradient. The terminal device sends the backpropagated intermediate gradient to the network device so that the network device can update the encoder model. Optionally, after the encoder model deployed on the terminal device has been fine-tuned, the terminal device can register the newly fine-tuned model with the network device in step 8. The network device assigns the model ID to be assigned to the new model on the terminal device for subsequent model pairing.

[0566] Optionally, before step 2, the network device can also perform scenario and data analysis and obtain the model ID to be assigned based on the analysis results, that is, step 1 is an optional step.

[0567] Please refer to the example shown in Figure 4h, which uses the encoder deployed on a network device and the decoder model deployed on a terminal device as an example. In this example, the model training is carried out with the model update method being Dec-only.

[0568] Step 1. The network device performs scenario and / or data analysis to determine whether to perform Dec-only training.

[0569] Step 2. The network device sends a Dec-only training instruction to the terminal device.

[0570] Step 3. The terminal device performs Dec training.

[0571] Step 4. After completing the Dec training, the terminal device sends a training completion instruction to the network device.

[0572] Step 5. Perform model registration between the terminal device and the network device. The trained model of the terminal device is registered to the network device through the model registration process.

[0573] As can be seen from the above process, for Dec-only model training, when the decoder model is deployed on the terminal device, the network device instructs the terminal device to perform Dec-only model training via RRC or DCI. The terminal device trains the decoder model and sends a model training completion indication message to the network device after training is complete. Furthermore, after the decoder model deployed on the terminal device completes training, the terminal device can register the newly trained model with the network device in step 5. The network device then assigns a model ID to the new model on the terminal device for subsequent model pairing.

[0574] Optionally, before step 2, the network device can also perform scenario and data analysis and obtain the model ID to be assigned based on the analysis results, that is, step 1 is an optional step.

[0575] Optionally, for Dec-only model training, when the decoder model is deployed on the network device, the terminal device may not be aware of the model update, or the network device may indicate to the terminal device via RRC or DCI that it is currently performing Dec-only model training, and instruct the terminal device to fall back to non-AI mode for communication.

[0576] Please refer to the example shown in Figure 4i, which takes the encoder deployed on a network device and the decoder model deployed on a terminal device as an example. In this example, the model training is taken as an example with the model update method being Codec.

[0577] Step 1. Analyze the network device's execution scenario and / or data to determine the training of the codec to be executed.

[0578] Step 2. The network device sends a Codec training instruction to the terminal device.

[0579] Step 3. The network device performs Enc inference.

[0580] Step 4. The network device and the terminal device exchange a reference model and / or dataset, which is used for model training.

[0581] Step 5. The terminal device performs Dec training.

[0582] Step 6. After completing the Dec training, the terminal device sends a training completion instruction to the network device.

[0583] Step 7. Perform model registration between the terminal device and the network device. The trained model of the terminal device is registered to the network device through the model registration process.

[0584] As described above, for codec model training, when the decoder model is deployed on the terminal device, the network device instructs the terminal device to train the codec model via RRC or DCI. The network device trains the encoder model, and the terminal device trains the decoder model. During training, there may be interaction between the reference model and / or the training dataset. After completing training, the terminal device sends a model training completion indication message to the network device. Furthermore, after the decoder model deployed on the terminal device completes training, the terminal device can register the newly trained model with the network device in step 7. The network device then assigns a model ID to the new model on the terminal device for subsequent model pairing.

[0585] Similarly, when the decoder model is deployed on a network device, the network device trains the decoder model, and the terminal device trains the encoder model.

[0586] Optionally, before step 2, the network device can also perform scenario and data analysis and obtain the model ID to be assigned based on the analysis results, that is, step 1 is an optional step.

[0587] Referring to Figure 5, this application embodiment provides a communication device 500. This communication device 500 can implement the functions of the first communication device (or second communication device) in the above method embodiments, and therefore can also achieve the beneficial effects of the above method embodiments. In this application embodiment, the communication device 500 can be the first communication device (or the second communication device), or it can be an integrated circuit or component inside the first communication device (or the second communication device), such as a chip, baseband chip, modem chip, SoC chip (e.g., an SoC chip containing a modem core), SIP chip, communication module, chip system, processor, etc.

[0588] Optionally, the transceiver unit 502 may include a transmitting unit and a receiving unit, which are used to perform transmitting and receiving respectively.

[0589] In one possible implementation, when the device 500 is used to execute the method performed by the first communication device in the preceding embodiments, the device 500 includes a processing unit 501 and a transceiver unit 502; the processing unit 501 is used to acquire first information, which is used to indicate the inference state information of a first model group, the first model group including a first model deployed on the first communication device and a second model deployed on the second communication device; wherein the input of the first model includes the output of the second model, or the input of the second model includes the output of the first model; the first communication device determines a second model group, which is related to the inference state information of the first model group, the second model group including a third model deployed on the first communication device and a fourth model deployed on the second communication device; wherein the input of the third model includes the output of the fourth model, or the input of the fourth model includes the output of the third model; the transceiver unit 502 is used to send second information, which indicates the third model and / or the fourth model.

[0590] In one possible implementation, when the device 500 is used to execute the method performed by the second communication device in the preceding embodiments, the device 500 includes a transceiver unit 502; the transceiver unit 502 is used to receive second information, the second information indicating a third model and / or a fourth model included in the second model group, the third model being deployed on the first communication device, and the fourth model being deployed on the second communication device; wherein the input of the third model includes the output of the fourth model, or the input of the fourth model includes the output of the third model; wherein the second model group is determined based on the inference state information of the first model group; the first model group includes a first model deployed on the first communication device and a second model deployed on the second communication device; wherein the input of the first model includes the output of the second model, or the input of the second model includes the output of the first model.

[0591] In one possible design, when the communication device 500 is a terminal device or a communication module within a terminal, the functionality of the processing unit 501 can be implemented by one or more processors. Specifically, the processor may include a modem chip, a SoC chip (such as a SoC chip containing a modem core), or a SIP chip. The functionality of the transceiver unit 502 can be implemented by transceiver circuitry.

[0592] In one possible design, when the communication device 500 is a circuit or chip in a terminal responsible for communication functions, such as a modem chip, a SoC chip, or a SoC chip or SIP chip containing a modem core, the function of the processing unit 501 can be implemented by a circuit system in the aforementioned chip that includes one or more processors or processor cores. The function of the transceiver unit 502 can be implemented by the interface circuitry or data transceiver circuitry on the aforementioned chip.

[0593] Optionally, the information execution process of the unit of the above-mentioned communication device 500 can be specifically referred to in the description of the method embodiment shown above in this application, and will not be repeated here.

[0594] Please refer to Figure 6, which is another schematic structural diagram of the communication device 600 provided in this application. The communication device 600 includes a logic circuit 601 and an input / output interface 602. The communication device 600 can be a chip or an integrated circuit.

[0595] In Figure 5, the transceiver unit 502 can be a communication interface, which can be the input / output interface 602 in Figure 6, and the input / output interface 602 can include an input interface and an output interface. Alternatively, the communication interface can also be a transceiver circuit, which can include an input interface circuit and an output interface circuit.

[0596] In one possible implementation, when the device 600 is used to execute the method performed by the first communication device in the preceding embodiments, the logic circuit 601 is used to acquire first information, which indicates the inference state information of a first model group, the first model group including a first model deployed on the first communication device and a second model deployed on the second communication device; wherein the input of the first model includes the output of the second model, or the input of the second model includes the output of the first model; the first communication device determines a second model group, which is related to the inference state information of the first model group, the second model group including a third model deployed on the first communication device and a fourth model deployed on the second communication device; wherein the input of the third model includes the output of the fourth model, or the input of the fourth model includes the output of the third model; the input / output interface 602 is used to send second information, which indicates the third model and / or the fourth model.

[0597] In one possible implementation, when the device 600 is used to execute the method performed by the second communication device in the preceding embodiments, the input / output interface 602 is used to receive second information, which indicates a third model and / or a fourth model included in the second model group, wherein the third model is deployed on the first communication device and the fourth model is deployed on the second communication device; wherein the input of the third model includes the output of the fourth model, or the input of the fourth model includes the output of the third model; wherein the second model group is determined based on the inference state information of the first model group; the first model group includes a first model deployed on the first communication device and a second model deployed on the second communication device; wherein the input of the first model includes the output of the second model, or the input of the second model includes the output of the first model.

[0598] The logic circuit 601 and the input / output interface 602 can also perform other steps performed by the first or second communication device in any embodiment and achieve corresponding beneficial effects, which will not be elaborated here.

[0599] In one possible implementation, the processing unit 501 shown in FIG5 can be the logic circuit 601 in FIG6.

[0600] Optionally, the logic circuit 601 can be a processing device, the functions of which can be partially or entirely implemented in software.

[0601] Optionally, the processing apparatus may include a memory and a processor, wherein the memory is used to store a computer program, and the processor reads and executes the computer program stored in the memory to perform the corresponding processing and / or steps in any of the method embodiments.

[0602] Optionally, the processing device may consist of only a processor. A memory for storing computer programs is located outside the processing device, and the processor is connected to the memory via circuitry / wires to read and execute the computer programs stored in the memory. The memory and processor may be integrated together or physically independent of each other.

[0603] Optionally, the processing device may be one or more chips, or one or more integrated circuits. For example, the processing device may be one or more field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), system-on-chips (SoCs), central processing units (CPUs), network processors (NPs), digital signal processors (DSPs), microcontroller units (MCUs), programmable logic devices (PLDs), or other integrated chips, or any combination of the above chips or processors.

[0604] Please refer to Figure 7, which shows the communication device 700 involved in the above embodiments provided in the embodiments of this application. Specifically, the communication device 700 can be the communication device as a terminal device in the above embodiments. The example shown in Figure 7 is that the terminal device is implemented through the terminal device (or the components in the terminal device).

[0605] The present invention provides a possible logical structure diagram of the communication device 700, which may include, but is not limited to, at least one processor 701 and a communication port 702.

[0606] In Figure 5, the transceiver unit 502 can be a communication interface, which can be the communication port 702 in Figure 7. The communication port 702 can include an input interface and an output interface. Alternatively, the communication port 702 can also be a transceiver circuit, which can include an input interface circuit and an output interface circuit.

[0607] Further optionally, the device may also include at least one of a memory 703 and a bus 704. In the embodiments of this application, the at least one processor 701 is used to control the operation of the communication device 700.

[0608] Furthermore, the processor 701 can be a central processing unit, a general-purpose processor, a digital signal processor, an application-specific integrated circuit, a field-programmable gate array, or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. The processor can also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a digital signal processor and a microprocessor, etc. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0609] Optionally, the communication device 700 shown in FIG7 can be used to implement the steps implemented by the terminal device in the aforementioned method embodiments and to achieve the corresponding technical effects of the terminal device. The specific implementation of the communication device shown in FIG7 can be referred to the description in the aforementioned method embodiments, and will not be repeated here.

[0610] Please refer to Figure 8, which is a structural schematic diagram of the communication device 800 involved in the above embodiments provided in the embodiments of this application. Specifically, the communication device 800 can be a communication device as a network device in the above embodiments. The example shown in Figure 8 is that the network device is implemented through a network device (or a component in the network device). The structure of the communication device can be referred to the structure shown in Figure 8.

[0611] The communication device 800 includes at least one processor 811 and at least one network interface 814. Optionally, the communication device further includes at least one memory 812, at least one transceiver 813, and one or more antennas 815. The processor 811, memory 812, transceiver 813, and network interface 814 are connected, for example, via a bus. In this embodiment, the connection may include various interfaces, transmission lines, or buses, etc., and this embodiment is not limited thereto. The antenna 815 is connected to the transceiver 813. The network interface 814 enables the communication device to communicate with other communication devices through a communication link. For example, the network interface 814 may include a network interface between the communication device and core network equipment, such as an S1 interface, or a network interface between the communication device and other communication devices (e.g., other network devices or core network equipment), such as an X2 or Xn interface.

[0612] In Figure 5, the transceiver unit 502 can be a communication interface, which can be the network interface 814 in Figure 8. The network interface 814 can include an input interface and an output interface. Alternatively, the network interface 814 can also be a transceiver circuit, which can include an input interface circuit and an output interface circuit.

[0613] The processor 811 is primarily used to process communication protocols and communication data, control the entire communication device, execute software programs, and process data from these programs, for example, to support the actions described in the embodiments of the communication device. The communication device may include a baseband processor and a central processing unit (CPU). The baseband processor is primarily used to process communication protocols and communication data, while the CPU is primarily used to control the entire terminal device, execute software programs, and process data from these programs. The processor 811 in Figure 8 can integrate the functions of both a baseband processor and a CPU. Those skilled in the art will understand that the baseband processor and CPU can also be independent processors interconnected via technologies such as buses. Those skilled in the art will understand that a terminal device may include multiple baseband processors to adapt to different network standards, and multiple CPUs to enhance its processing capabilities. The various components of the terminal device can be connected via various buses. The baseband processor can also be described as a baseband processing circuit or a baseband processing chip. The CPU can also be described as a central processing circuit or a central processing chip. The function of processing communication protocols and communication data can be built into the processor or stored in memory as a software program, which is then executed by the processor to implement the baseband processing function.

[0614] The memory is primarily used to store software programs and data. The memory 812 can exist independently or be connected to the processor 811. Optionally, the memory 812 can be integrated with the processor 811, for example, integrated into a single chip. The memory 812 can store program code that executes the technical solutions of the embodiments of this application, and its execution is controlled by the processor 811. The various types of computer program code being executed can also be considered as drivers for the processor 811.

[0615] Figure 8 shows only one memory and one processor. In actual terminal devices, there may be multiple processors and multiple memories. Memory can also be called storage medium or storage device, etc. Memory can be a storage element on the same chip as the processor, i.e., an on-chip storage element, or it can be a separate storage element; this application does not limit this.

[0616] Transceiver 813 can be used to support the reception or transmission of radio frequency (RF) signals between a communication device and a terminal. Transceiver 813 can be connected to antenna 815. Transceiver 813 includes a transmitter Tx and a receiver Rx. Specifically, one or more antennas 815 can receive RF signals. The receiver Rx of transceiver 813 receives the RF signals from the antennas, converts the RF signals into digital baseband signals or digital intermediate frequency (IF) signals, and provides the digital baseband signals or IF signals to processor 811 so that processor 811 can perform further processing on the digital baseband signals or IF signals, such as demodulation and decoding. Furthermore, the transmitter Tx in transceiver 813 is also used to receive modulated digital baseband signals or IF signals from processor 811, convert the modulated digital baseband signals or IF signals into RF signals, and transmit the RF signals through one or more antennas 815. Specifically, the receiver Rx can selectively perform one or more stages of downmixing and analog-to-digital conversion on the radio frequency signal to obtain a digital baseband signal or a digital intermediate frequency (IF) signal. The order of these downmixing and IF conversion processes is adjustable. The transmitter Tx can selectively perform one or more stages of upmixing and digital-to-analog conversion on the modulated digital baseband signal or digital IF signal to obtain a radio frequency signal. The order of these upmixing and IF conversion processes is also adjustable. The digital baseband signal and the digital IF signal can be collectively referred to as digital signals.

[0617] The transceiver 813 can also be called a transceiver unit, transceiver, transceiver device, etc. Optionally, the device in the transceiver unit that performs the receiving function can be regarded as the receiving unit, and the device in the transceiver unit that performs the transmitting function can be regarded as the transmitting unit. That is, the transceiver unit includes a receiving unit and a transmitting unit. The receiving unit can also be called a receiver, input port, receiving circuit, etc., and the transmitting unit can be called a transmitter, transmitter, or transmitting circuit, etc.

[0618] Optionally, the communication device 800 shown in FIG8 can be used to implement the steps implemented by the network device in the aforementioned method embodiments and achieve the corresponding technical effects of the network device. The specific implementation of the communication device 800 shown in FIG8 can be referred to the description in the aforementioned method embodiments, and will not be repeated here.

[0619] Please refer to Figure 9, which is a schematic diagram of the structure of the communication device involved in the above embodiments provided in the embodiments of this application.

[0620] Optionally, the communication device 900 includes, for example, modules, units, elements, circuits, or interfaces, etc., appropriately configured together to execute the technical solutions provided in this application. The communication device 900 may be the terminal device or network device described above, or a component (e.g., a chip) within these devices, used to implement the methods described in the following method embodiments. The communication device 900 includes one or more processors 901. The processor 901 may be a general-purpose processor or a dedicated processor, etc. For example, it may be a baseband processor or a central processing unit. The baseband processor can be used to process communication protocols and communication data, and the central processing unit can be used to control the communication device (e.g., RAN node, terminal, or chip, etc.), execute software programs, and process data from the software programs.

[0621] Optionally, in one design, processor 901 may include program 903 (sometimes also referred to as code or instructions), which may be executed on processor 901 to cause communication device 900 to perform the methods described in the following embodiments. In yet another possible design, communication device 900 includes circuitry (not shown in FIG9).

[0622] Optionally, the communication device 900 may include one or more memories 902 storing a program 904 (sometimes referred to as code or instructions), which can be run on the processor 901 to cause the communication device 900 to perform the methods described in the above method embodiments.

[0623] Optionally, the processor 901 and / or memory 902 may include AI modules 907 and 908, which are used to implement AI-related functions. The AI ​​modules can be implemented through software, hardware, or a combination of both. For example, the AI ​​module may include a radio intelligence control (RIC) module. For instance, the AI ​​module may be a near real-time RIC or a non-real-time RIC.

[0624] Optionally, the processor 901 and / or memory 902 may also store data. The processor and memory may be configured separately or integrated together.

[0625] Optionally, the communication device 900 may further include a transceiver 905 and / or an antenna 906. The processor 901, sometimes referred to as a processing unit, controls the communication device (e.g., a RAN node or terminal). The transceiver 905, sometimes referred to as a transceiver unit, transceiver, transceiver circuit, or transceiver, is used to implement the transmission and reception functions of the communication device via the antenna 906.

[0626] In Figure 5, the processing unit 501 can be a processor 901. The transceiver unit 502 shown in Figure 5 can be a communication interface, which can be the transceiver 905 in Figure 9. The transceiver 905 can include an input interface and an output interface. Alternatively, the transceiver 905 can also be a transceiver circuit, which can include an input interface circuit and an output interface circuit.

[0627] This application also provides a computer-readable storage medium for storing one or more computer-executable instructions. When the computer-executable instructions are executed by a computer, the computer performs the method described in the possible implementations of the first or second communication device in the foregoing embodiments.

[0628] This application also provides a computer program product (or computer program) that, when executed by a computer, allows the computer to execute the method described above for the possible implementation of the first or second communication device.

[0629] This application also provides a chip system including at least one processor for supporting a communication device in implementing the functions involved in the possible implementations of the communication device described above. Optionally, the chip system further includes an interface circuit that provides program instructions and / or data to the at least one processor. In one possible design, the chip system may also include a memory for storing the program instructions and data necessary for the communication device. The chip system may be composed of chips or may include chips and other discrete devices, wherein the communication device may specifically be the first communication device or the second communication device in the aforementioned method embodiments.

[0630] This application also provides a communication system, which includes the first communication device in any of the above embodiments.

[0631] Optionally, the communication system may also include a second communication device.

[0632] In the embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, indirect coupling or communication connection between devices or units, and may be electrical, mechanical, or other forms. Whether a function is implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0633] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0634] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

Claims

1. A communication method characterized by comprising: The method comprises: obtaining first information, the first information being used to indicate inference state information of a first model group, the first model group comprising a first model deployed on a first communication device and a second model deployed on a second communication device; wherein an input of the first model comprises an output of the second model, or an input of the second model comprises an output of the first model; the inference state information of the first model group comprises at least one of performance of the first model group, a number of times of model inference using the first model group, or a time length of model inference using the first model group; determining a second model group, the second model group being related to the inference state information of the first model group, the second model group comprising a third model deployed on the first communication device and a fourth model deployed on the second communication device; wherein an input of the third model comprises an output of the fourth model, or an input of the fourth model comprises an output of the third model; sending second information, the second information indicating the third model and / or the fourth model.

2. The method of claim 1, wherein, The second information satisfies any one of the following conditions: the first model is different from the third model, the second model is the same as the fourth model, and the second information indicates the third model; the first model is the same as the third model, the second model is different from the fourth model, and the second information indicates the fourth model; or the first model is different from the third model, the second model is different from the fourth model, and the second information indicates both the third model and the fourth model.

3. The method of claim 2, wherein, In a case where the second information indicates the third model or the second information indicates the fourth model, the method further comprises: obtaining third information, the third information being used to indicate inference state information of the second model group; determining a third model group, the third model group being related to the inference state information of the second model group, the third model group comprising a fifth model deployed on the first communication device and a sixth model deployed on the second communication device; wherein an input of the fifth model comprises an output of the sixth model, or an input of the sixth model comprises an output of the fifth model; sending fourth information, the fourth information indicating the fifth model and / or the sixth model.

4. The method of claim 3, wherein, In a case where a first condition is met, any one of the following conditions is satisfied: the fifth model is obtained by model switching from the third model, and the fourth model is the same as the sixth model; the sixth model is obtained by model switching from the fourth model, and the third model is the same as the fifth model; or the fifth model is obtained by model switching from the third model, and the sixth model is obtained by model switching from the fourth model.

5. The method of claim 3, wherein, In a case where a second condition is met, any one of the following conditions is satisfied: the fifth model is obtained by fine-tuning from the third model, and the fourth model is the same as the sixth model; the sixth mode is obtained by fine-tuning from the fourth model, and the third model is the same as the fifth model; or The fifth model is obtained by model fine-tuning on the third model, and the sixth model is obtained by model fine-tuning on the fourth model.

6. The method of claim 3, wherein, In a case where the third condition is met, any of the following is met: The fifth model is obtained by model training on the third model, and the fourth model is the same as the sixth model; The sixth model is obtained by model training on the fourth model, and the third model is the same as the fifth model; or The fifth model is obtained by model training on the third model, and the sixth model is obtained by model training on the fourth model.

7. The method according to any one of claims 1 to 6, characterized in that, In a case where the fourth condition is met, any of the following is met: The third model is obtained by model switching on the first model, and the second model is the same as the fourth model; The fourth model is obtained by model switching on the second model, and the first model is the same as the third model; or The third model is obtained by model switching on the first model, and the fourth model is obtained by model switching on the second model.

8. The method according to any one of claims 1 to 6, characterized in that, In a case where the fifth condition is met, any of the following is met: The third model is obtained by model fine-tuning on the first model, and the second model is the same as the fourth model; The fourth mode is obtained by model fine-tuning on the second model, and the first model is the same as the third model; or the third model is obtained by model fine-tuning on the first model, and the fourth model is obtained by model fine-tuning on the second model.

9. The method according to any one of claims 1 to 6, characterized in that, In a case where the sixth condition is met, any of the following is met: The third model is obtained by model training on the first model, and the second model is the same as the fourth model; The fourth is obtained by model training on the second model, and the first model is the same as the third model; or In a case where the sixth condition is met, any of the following is met:

10. The method according to any one of claims 1 to 9, characterized in that, The third is obtained by model training on the first model, and the fourth is obtained by model training on the second model. The determining the second model group comprises:

11. The method of claim 10, wherein, determining the second model group based on the inference state information and the inference configuration information of the first model group.

12. The method according to any one of claims 1 to 11, characterized in that, The inference configuration information comprises one or more model identifiers and / or inference auxiliary information. The method further comprises: sending fifth information, the fifth information indicating AI capability information of the first communication device; and / or, receiving sixth information, the sixth information indicating AI capability information of the second communication device; wherein the AI capability information of the first communication device and / or the AI capability information of the second communication device is used to determine the second model group.

13. The method of claim 12, wherein the AI capability information of the first communication device indicates at least one of the following: one or more supported AI models, a supported model updating manner, whether a model deployed on the first communication device can independently perform model updating, or whether a model deployed on the first communication device can jointly perform model updating with an associated model; and / or, The AI capability information of the second communication device indicates at least one of the following: one or more AI models supported, a model updating manner supported, whether a model deployed on the second communication device independently performs model updating, or whether a model deployed on the second communication device jointly performs model updating with an associated model. The supported model updating manner includes at least one of model switching, model fine-tuning, or model training.

14. A communication method, comprising: The method comprises: receiving second information, the second information indicating a third model and / or a fourth model included in a second model group, the third model being deployed on a first communication device, and the fourth model being deployed on a second communication device; wherein input of the third model includes output of the fourth model, or input of the fourth model includes output of the third model; The inference state information of the second model group is related to inference state information of a first model group, the inference state information of the first model group including at least one of performance of the first model group, a number of times of model inference using the first model group, or a time length of model inference using the first model group; the first model group includes a first model deployed on a first communication device and a second model deployed on a second communication device; wherein input of the first model includes output of the second model, or input of the second model includes output of the first model.

15. The method of claim 14, wherein, The second information satisfies any one of the following: The first model is different from the third model, the second model is the same as the fourth model, and the second information indicates the third model; The first model is the same as the third model, the second model is different from the fourth model, and the second information indicates the fourth model; or The first model is different from the third model, the second model is different from the fourth model, and the second information indicates both the third model and the fourth model.

16. The method of claim 15, wherein, In a case where the second information indicates the third model or the second information indicates the fourth model, the method further comprises: receiving fourth information, the fourth information indicating a fifth model and / or a sixth model included in a third model group, the fifth model being deployed on the first communication device, and the sixth model being deployed on the second communication device; wherein input of the fifth model includes output of the sixth model, or input of the sixth model includes output of the fifth model; The inference state information of the third model is related to inference state information of a second model group.

17. The method of claim 16, wherein, In a case where a first condition is met, any one of the following is met: The fifth model is obtained by model switching from the third model, and the fourth model is the same as the sixth model; The sixth model is obtained by model switching from the fourth model, and the third model is the same as the fifth model; or The fifth model is obtained by model switching from the third model, and the sixth model is obtained by model switching from the fourth model.

18. The method of claim 16, wherein, In a case where a second condition is met, any one of the following is met: The fifth model is obtained by model fine-tuning of the third model, the fourth model is the same as the sixth model; The sixth model is obtained by model fine-tuning of the fourth model, the third model is the same as the fifth model; or The fifth model is obtained by model fine-tuning of the third model, and the sixth model is obtained by model fine-tuning of the fourth model.

19. The method of claim 16, wherein, In the case where the third condition is met, any of the following is met: The fifth model is obtained by model training of the third model, the fourth model is the same as the sixth model; The sixth model obtained by model training of the fourth model, the third model is the same as the fifth model; or The sixth model is obtained by model training of the fourth model, the third model is the same as the fifth model.

20. The method according to any one of claims 14 to 19, characterized in that, In the case where the fourth condition is met, any of the following is met: The third model is obtained by model switching of the first model, the second model is the same as the fourth model; The fourth model is obtained by model switching of the second model, the first model is the same as the third model; or The third model is obtained by model switching of the first model, and the fourth model is obtained by model switching of the second model.

21. The method according to any one of claims 14 to 19, characterized in that, In the case where the fifth condition is met, any of the following is met: The third model is obtained by model fine-tuning of the first model, the second model is the same as the fourth model; The fourth model obtained by model fine-tuning of the second model, the first model is the same as the third model; or The fourth model is obtained by model fine-tuning of the second model, the third model is the same as the first model.

22. The method according to any one of claims 14 to 19, characterized in that, In the case where the sixth condition is met, any of the following is met: The third model is obtained by model training of the first model, the second model is the same as the fourth model; The fourth model, obtained by model training of the second model, the first model is the same as the third model; or The second model group is determined based on inference state information and inference configuration information of the first model group.

23. The method according to any one of claims 14 to 22, characterized in that, The inference configuration information includes one or more model identifiers and / or inference auxiliary information.

24. The method of claim 23, wherein, The method further comprises:

25. The method according to any one of claims 14 to 24, characterized in that, sending fifth information, the fifth information indicating AI capability information of the first communication device; and / or, receiving sixth information, the sixth information indicating AI capability information of the second communication device; wherein the AI capability information of the first communication device and / or the AI capability information of the second communication device is used to determine the second model group.

26. The method of claim 25, wherein, ​ The first communication device AI capability information indicates at least one of the following: one or more AI models supported, a model updating manner supported, whether a model deployed on the first communication device is supported to independently perform model updating, or whether the model deployed on the first communication device is supported to jointly perform model updating with an associated model. And / or, The second communication device AI capability information indicates at least one of the following: one or more AI models supported; a model updating manner supported; whether a model deployed on the second communication device is supported to independently perform model updating; or whether the model deployed on the second communication device is supported to jointly perform model updating with an associated model. The model updating manner supported includes at least one of model switching, model fine-tuning, or model training.

27. A communications device, characterized by A module for performing the method of any of claims 1-26.

28. A communications device, characterized by At least one processor for performing the method of any of claims 1-26.

29. The communication apparatus according to claim 28, wherein The communication device is a chip or a chip system.

30. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program or instructions which, when executed, implement the method of any of claims 1-26.

31. A computer program product, characterised in that, A computer program or instructions which, when executed by a computer, implement the method of any of claims 1-26.

Images