Communication method and related device

Abstract

A communication method and a related device, in the method, a first communication device can process one of sub-models included in a model based on an indication of first information. Therefore, compared with the process that the communication device processes the complete model (namely the first model), the process that the first communication device processes one of the sub-models included in the model, the complexity of the first communication device on the model processing process can be reduced, and meanwhile, the model processing time delay can be reduced. In some implementations, the first sub-model processed by the first communication device may be a processing result of other communication devices, and / or the second sub-model processed by the first communication device may be used for model processing of other communication devices. In this way, different communication devices can process the sub-models contained in the model in a chained (or tandem) manner, so as to realize cooperation of a plurality of communication devices.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of communication, and in particular to a communication method and related apparatus. BACKGROUND

[0002] With the development of communication technology, in a communication system, in addition to the traditional communication service, the service performed by the communication device can also include other new services, such as artificial intelligence (AI) service. Generally, the communication system capable of processing AI service can also be referred to as AI system.

[0003] At present, the communication device can serve as a participating node of the AI system and provide its own computing power and data. For example, the communication device can perform one or more model processing (such as model training, model updating, or model fine-tuning, etc.) processes on a model based on local data to obtain another model. In one possible way, increasing the complexity of the model (such as increasing the number of parameters of the model, increasing the number of neural network layers contained in the model, etc.) can effectively improve the performance of the model.

[0004] However, in the case of gradually increasing the complexity of the model, it is possible to cause the above-mentioned model processing process to consume a relatively long time, and even it is possible to cause the above-mentioned model processing process to be unable to be executed due to the limitation of the resources (such as computing power resources, storage resources, etc.) of the communication device. Therefore, how to optimize the model processing process is a technical problem to be solved. SUMMARY

[0005] The present application provides a communication method and related apparatus, which are used for reducing the complexity of the model processing process of the communication device and reducing the model processing delay, so as to improve the model processing efficiency.

[0006] The first aspect of the present application provides a communication method, which is performed by a first communication device. The first communication device can be a communication apparatus (e.g., a terminal device or a network device), or the first communication device can be a part of the communication apparatus (e.g., a circuit or a chip responsible for communication functions (e.g., a Modem chip (also referred to as a baseband chip), a system on chip (SoC) chip, such as an SoC chip including a modem core, or a system in package (SIP) chip), etc.), or the first communication device can also be a logic module or software capable of implementing all or part of the functions of the communication apparatus. In the method, the first communication device receives first information, which is used to indicate a first sub-model. The first sub-model is one of sub-models in a first model, and the first model includes at least two sub-models. The first communication device processes the first sub-model to obtain a second sub-model. The first communication device sends second information, which is used to indicate the second sub-model. The first sub-model is obtained by processing a model by another communication device, and / or the second sub-model is used for model processing by another communication device.

[0007] Based on the above scheme, the first information received by the first communication device is used to indicate the first sub-model in the at least two sub-models included in the first model. Thereafter, the first communication device can send the second information, which is used to indicate the second sub-model obtained by processing the first sub-model. In other words, the first communication device can process one of the sub-models included in the model based on the indication of the first information. Thus, compared with the process of model processing of the complete model (i.e., the first model) by the communication device, the process of processing one of the sub-models included in the model by the first communication device can reduce the complexity of the model processing process of the first communication device, and also reduce the model processing delay.

[0008] In addition, the number of first communication devices can be one or more. In the case where the number of first communication devices is greater than 1, the sender (e.g., the second communication device) of the first information can indicate the same or different sub-models to different first communication devices. In the distributed scenario where the different first communication devices participate in the model processing process as distributed nodes, compared with the process of model processing based on the complete model (i.e., the first model) by each distributed node, the processing complexity of the distributed nodes can be reduced, and the processing delay of the distributed nodes can also be reduced, so as to improve the model processing efficiency in the distributed scenario.

[0009] In addition, the first sub-model indicated by the first information is obtained by the other communication apparatus through model processing, and / or the second sub-model indicated by the second information is used for model processing of the other communication apparatus. In other words, the first sub-model processed by the first communication apparatus can be a processing result of the other communication apparatus, and / or the second sub-model processed by the first communication apparatus can be used for model processing of the other communication apparatus. In this way, different communication apparatuses can process sub-models contained in the model in a chain (or series) manner to realize cooperation of multiple communication apparatuses.

[0010] Optionally, the first communication apparatus sends capability information of the first communication apparatus; and the capability information is used to determine the first information. In this way, the sender of the first information (for example, the second communication apparatus) can determine the first sub-model matched with the capability information of the first communication apparatus, so that the first communication apparatus can perform model processing based on the sub-model matched with its own capability, and model processing failure caused by mismatch between the sub-model and the capability can be avoided, so as to improve the success rate of model processing.

[0011] Optionally, the capability information includes one or more of the following: computing power information (for example, the computing power information can indicate one or more of total computing power, used computing power, and idle computing power) of the first communication apparatus, storage information (for example, the storage information can indicate one or more of total storage space, used storage space, and idle storage space) of the first communication apparatus, communication information (for example, the communication information can indicate one or more of antenna information of the first communication apparatus, communication chip information of the first communication apparatus, and channel information between the first communication apparatus and the other communication apparatus), or other information.

[0012] In this application, the model can include a reference model, an AI model, a neural network model, an AI neural network model, a machine learning model, or an AI processing model. Similarly, the sub-model can include a sub-reference model, an AI sub-model, a neural network sub-model, an AI neural network sub-model, a machine learning sub-model, or an AI processing sub-model.

[0013] In this application, the communication apparatus processes one sub-model to obtain another sub-model, and the processing can include one or more of training, fine-tuning, fine-tuning, updating, iteration, and optimization. For example, the first communication apparatus processes the first sub-model through one or more of training, fine-tuning, fine-tuning, updating, iteration, and optimization to obtain the second sub-model.

[0014] In a possible implementation manner of the first aspect, the first information includes an identifier of the first sub-model.

[0015] Based on the above scheme, the first information is used to indicate the first sub-model by the identification of the first sub-model, so that the first communication device can quickly determine the first sub-model based on the identification, and the overhead of indicating the first sub-model can be reduced.

[0016] Optionally, the relationship between the identification of the one or more sub-models and the model structure of the one or more sub-models is predefined. In this way, the first communication device can quickly determine the model structure of the first sub-model based on the identification of the first sub-model, and the overhead of indicating the model structure can be reduced.

[0017] In a possible implementation of the first aspect, the first information is further used to indicate at least one of:

[0018] the model parameters of the first sub-model (for example, the model parameters can include frozen parameters and / or non-frozen parameters);

[0019] the input data set of the first sub-model;

[0020] the output data set of the first sub-model;

[0021] the input position of the input data set of the first sub-model in one or more model units included in the first sub-model;

[0022] the output position of the output data set of the first sub-model in one or more model units included in the first sub-model.

[0023] Optionally, the at least one of the above can be preconfigured.

[0024] Optionally, the second communication device can also indicate the at least one of the above in other ways, for example, the second communication device indicates the at least one of the above by other information / messages / signaling different from the first information.

[0025] In this application, the model unit can be replaced by other terms, such as reference unit, model element, model component, model component, or model reference unit, etc.

[0026] Based on the above scheme, in addition to indicating the first sub-model, the first information can also be used to indicate the at least one of the above, so that the first communication device can implement the processing of the first sub-model based on the at least one.

[0027] In a possible implementation of the first aspect, the first model includes at least two model units, and each of the at least two sub-models includes one or more model units of the at least two model units; wherein the first sub-model includes M model units of the at least two model units, M is a positive integer; and the first information includes the identification of the M model units.

[0028] Based on the above scheme, the first model can comprise at least two model units, and correspondingly, in the at least two sub-models comprised by the first model, each sub-model can comprise one or more model units of the at least two model units. Thus, the sender of the first information (e.g., the second communication apparatus) can indicate the first sub-model by the identification of the M model units comprised by the first sub-model, so that the first communication apparatus can quickly determine the first sub-model based on the identification, and meanwhile, the overhead of the indication of the first sub-model can be reduced.

[0029] Optionally, in the at least two model units comprised by the first model, the identification of each model unit is pre-defined in relation to the model structure of the model unit. In this way, the first communication apparatus can quickly determine the model structure of the M model units comprised by the first sub-model based on the identification of the M model units, and the overhead of the indication of the model structure can be reduced.

[0030] In a possible implementation of the first aspect, the first information is further used to indicate a connection relationship of the M model units.

[0031] Based on the above scheme, in addition to indicating the M model units comprised by the first sub-model by the identification of the M model units, the first information can be used to indicate the connection relationship of the M model units, so that the first communication apparatus can determine the first sub-model based on the connection relationship and the M model units.

[0032] Optionally, the second communication apparatus can also indicate the connection relationship of the M model units in other manners, e.g., the second communication apparatus indicates the connection relationship of the M model units by other information / messages / signaling different from the first information.

[0033] In a possible implementation of the first aspect, the connection relationship of the at least two model units is pre-configured.

[0034] Based on the above scheme, in the at least two model units comprised by the first model, the connection relationship of the at least two model units is pre-configured. In this way, the overhead of the indication of the connection relationship of the at least two model units (e.g., the indication of the connection relationship of the M model units) can be saved.

[0035] In a possible implementation of the first aspect, the first information is further used to indicate at least one of:

[0036] model parameters of part or all of the M model units;

[0037] configuration parameters of part or all of the M model units;

[0038] an input data set of some or all of the M model units;

[0039] an output data set of some or all of the M model units.

[0040] Optionally, the at least one item can be preconfigured.

[0041] Optionally, the second communication device can also indicate the at least one item in other manners, for example, the second communication device indicates the at least one item through other information / messages / signaling different from the first information.

[0042] Based on the above scheme, the first information can be used to indicate the at least one item in addition to indicating the first sub-model, so that the first communication device can implement processing of the first sub-model based on the at least one item.

[0043] In a possible implementation manner of the first aspect, the method further includes: the first communication device sending indication information used to indicate a local data set; and wherein the local data set is used to determine the first model.

[0044] Based on the above scheme, the first communication device can also send indication information used to indicate a local data set, so that a receiver of the indication information can determine the first model based on the local data set of the first communication device.

[0045] Optionally, the local data set can be non-private data, de-identification data, etc. of the first communication device. In this way, it is beneficial to protect the data security of the first communication device and avoid privacy leakage.

[0046] In a possible implementation manner of the first aspect, the first information is also used to indicate a processing result of the first sub-model.

[0047] Based on the above scheme, the first communication device can also indicate sending of the second information, so that a receiver of the second information can obtain a model processing result (i.e., a second sub-model) of the first communication device based on the second information, so as to enable the receiver to perform a model management (for example, the receiver performs model aggregation, model updating, etc. based on the second sub-model) process based on the second sub-model.

[0048] Optionally, the second communication device can also indicate sending of the processing result of the first sub-model in other manners, for example, the second communication device indicates sending of the processing result of the first sub-model through other information / messages / signaling different from the first information.

[0049] In a possible implementation manner of the first aspect, the first communication device sending the second information comprises: the first communication device sending the second information at or after a first time point; and a difference between the first time point and a second time point is determined by the time offset information, and the second time point is associated with the first sub-model.

[0050] Based on the above scheme, the first time point can be a time point at which the first communication device supervises processing of the first sub-model. The first time point can be before an end time point of the processing of the first sub-model by the first communication device. For example, the first communication device can supervise during training of the first sub-model. In this way, the first communication device can send the second information at or after the first time point, so that a receiver (for example, the second communication device) of the second information can receive the second information as soon as possible before the end time point of the training, thereby reducing the overall processing time delay of the plurality of first communication devices on different sub-models.

[0051] For example, if the first communication device discovers that the training performance is lower than the target threshold in advance, the second communication device can learn this information as soon as possible through the above process, so that the second communication device can select other communication devices for training, so as to reduce the overall processing time delay of the plurality of first communication devices on different sub-models.

[0052] In addition, the data of the processing of the sub-model by different communication devices can have timeliness. In the above scheme, the overall processing time delay of the plurality of first communication devices on different sub-models is reduced, and the data of each communication device can be avoided to be invalid to a certain extent.

[0053] It should be understood that the first time point can be a time point at which the first communication device obtains / determines / obtains the performance of the second sub-model, or the first time point can be a time point at which the first communication device obtains / determines / obtains the difference between the performance of the second sub-model and the threshold. For example, the first time point can be referred to as a supervision time point, a detection time point, and the like.

[0054] In a possible implementation manner of the first aspect, the first communication device sending the second information comprises: the first communication device sending the second information at or before a third time point, wherein a difference between the third time point and a second time point is determined by the time offset information, and the second time point is associated with the first sub-model.

[0055] Based on the above scheme, the third time point can be a deadline for the first communication device to process the first sub-model. For example, the first communication device can perform supervision before or at the deadline for training the first sub-model. In this way, the first communication device can send the second information at or after the first time point, so that the receiver (e.g., the second communication device) of the second information can receive the second information as soon as possible before the training end time point, thereby reducing the overall processing delay of multiple first communication devices for different sub-models.

[0056] For example, if the first communication device discovers that the training performance is lower than the target threshold in advance, the second communication device can learn this information as soon as possible, and the second communication device can select other communication devices for training.

[0057] Optionally, the deadline can be replaced by other descriptions, such as a termination time point, or a last time unit (such as a symbol, a subframe, a frame, etc.).

[0058] In a possible implementation of the first aspect, the method further includes: receiving, by the first communication device, third information, the third information being used to indicate the time offset information.

[0059] Based on the above scheme, the first communication device can also determine the time offset information based on the received third information, so that the first communication device can send the second information based on the specified time offset information.

[0060] In a possible implementation of the first aspect, the method further includes: sending, by the first communication device, fourth information, the fourth information being used to indicate the second time point, the second time point being a start time point of training the first sub-model.

[0061] Based on the above scheme, the first communication device can also send the fourth information indicating the second time point, so that the receiver (e.g., the second communication device) of the fourth information can determine the corresponding time offset information based on the start time point of training the first sub-model, to assist the decision of the receiver.

[0062] In a possible implementation of the first aspect, the time offset information is preconfigured.

[0063] Based on the above scheme, the time offset information used to determine the sending time point of the second information can be preconfigured, which can save the configuration overhead of the time offset information.

[0064] In a possible implementation of the first aspect, the second time point is associated with the first sub-model, including: the second time point is a receiving time point of the first information, or the second time point is a start time point of training the first sub-model.

[0065] Based on the above scheme, the second time can be implemented in the above-mentioned multiple ways to improve the flexibility of the scheme implementation.

[0066] In a possible implementation of the first aspect, the first communication device sends the second information at the first time or after the first time, including: in a case where a first condition is met, the first communication device sends the second information at the first time or after the first time; and the first condition includes at least one of the following: the model performance of the second sub-model is higher than a threshold, or the fifth information is received, the fifth information being used to indicate termination of processing of the first sub-model.

[0067] Based on the above scheme, the first communication device can also send the second information based on the first condition to improve the model processing efficiency. For example, in a case where the first condition includes that the model performance of the second sub-model is higher than a threshold, the first communication device can send the second information when the performance of the second sub-model is higher, so that the receiver of the second information can obtain the sub-model with higher performance. For another example, in a case where the first condition includes that the first communication device receives the fifth information, the first communication device can send the second information based on the termination processing indication of the opposite end.

[0068] Optionally, the second information can include a reason value, the reason value being used to indicate the first condition.

[0069] Optionally, the fifth information can be triggered based on sixth information, that is, the second communication device can send the fifth information to the first communication device based on an abnormal situation indicated by the sixth information after receiving the sixth information. The implementation of the sixth information can refer to the description below.

[0070] In a possible implementation of the first aspect, the first communication device sends the sixth information at the first time or after the first time in a case where a second condition is met, the sixth information being used to indicate that the first sub-model processing is abnormal; and the second condition includes at least one of the following: the model performance of the second sub-model is lower than a threshold, or the processing capability of the first communication device does not match the first sub-model.

[0071] Based on the above scheme, the first communication device can also send the sixth information based on the second condition, so that the scheme can adapt to the model processing abnormality scenario. For example, in a case where the second condition includes that the model performance of the second sub-model is lower than a threshold, the first communication device can send the sixth information when the performance of the second sub-model is lower, so that the receiver of the sixth information can learn about the abnormal situation of the model processing of the first communication device as soon as possible. For another example, in a case where the second condition includes that the processing capability of the first communication device does not match the first sub-model, the receiver of the sixth information can learn about the abnormal situation that the processing capability of the first communication device does not match the indicated first sub-model as soon as possible.

[0072] Optionally, the sixth information can comprise a cause value, the cause value being used to indicate the second condition.

[0073] Optionally, the process that the first communication device sends the second information is an optional step. For example, in the case that the first communication device sends the sixth information, it is possible that the first communication device does not obtain the second sub-model based on the first sub-model, and therefore the first communication device can not be able to send the second information indicating the second sub-model. Or, in the case that the first communication device sends the sixth information, it is possible that the first communication device does not obtain the second sub-model with better performance based on the first sub-model, and therefore the first communication device does not need to send the second information, which can avoid the second communication device performing processing based on the second sub-model with too low performance, and can reduce the overhead.

[0074] Optionally, in the case that the first communication device sends the sixth information, the first communication device can also send the second information. Since the second sub-model indicated by the second information can be obtained by the first communication device based on its own computing power, data set, etc., therefore even if the performance of the second sub-model is poor, the second communication device can still perform processing based on the second sub-model, and the local computing power and / or local data set of one or more first communication devices can be utilized as much as possible.

[0075] The second aspect of the present application provides a communication method, which is performed by a second communication device. The second communication device can be a communication device (such as a terminal device or a network device), or the second communication device can be a part of a communication device (for example, a circuit or a chip responsible for communication functions (such as a Modem chip (also known as a baseband chip), a SoC chip, such as a SoC chip containing a modem core, or a SIP chip, etc.), or the second communication device can also be a logic module or software capable of realizing all or part of the functions of the communication device. In the method, the second communication device sends first information, the first information being used to indicate a first sub-model, the first sub-model being one of the sub-models in a first model, the first model comprising at least two sub-models; wherein the first sub-model is used to obtain a second sub-model through processing by a first communication device; the second communication device receives second information, the second information being used to indicate the second sub-model; wherein the first sub-model is obtained by model processing of other communication devices, and / or the second sub-model is used for model processing of other communication devices.

[0076] Based on the above scheme, the first information sent by the second communication device to the first communication device is used to indicate a first sub-model in the at least two sub-models contained in the first model. Thereafter, the first communication device can send second information to the second communication device, the second information being used to indicate a second sub-model obtained based on processing of the first sub-model. In other words, the first communication device can process one of the sub-models contained in the model based on the indication of the first information. Thus, compared with the process of model processing of the first communication device based on the complete model (i.e., the first model), the process of model processing of the first communication device based on one of the sub-models contained in the model can reduce the complexity of the model processing of the first communication device and also reduce the model processing delay.

[0077] In addition, the number of the first communication devices can be one or more. In the case where the number of the first communication devices is greater than one, the sender (e.g., the second communication device) of the first information can indicate the same or different sub-models to different first communication devices. In the distributed scenario where the different first communication devices participate in the model processing as distributed nodes, compared with the process of model processing of each distributed node based on the complete model (i.e., the first model), the process of model processing of the distributed nodes based on one of the sub-models contained in the model can reduce the processing complexity of the distributed nodes and also reduce the processing delay of the distributed nodes, so as to improve the model processing efficiency in the distributed scenario.

[0078] In addition, the first sub-model indicated by the first information is obtained by model processing of other communication devices, and / or the second sub-model indicated by the second information is used for model processing of other communication devices. In other words, the first sub-model processed by the first communication device can be the processing result of other communication devices, and / or the second sub-model processed by the first communication device can be used for model processing of other communication devices. In this way, different communication devices can process the sub-models contained in the model in a chain (or series) manner to realize cooperation of multiple communication devices.

[0079] In a possible implementation of the second aspect, the first information includes an identifier of the first sub-model.

[0080] Based on the above scheme, the second communication device indicates the first sub-model by the identifier of the first sub-model, so that the first communication device can quickly determine the first sub-model based on the identifier and also reduce the overhead of indication of the first sub-model.

[0081] Optionally, the relationship between the identifier of the one or more sub-models and the model structure of the one or more sub-models is predefined. In this way, the first communication device can quickly determine the model structure of the first sub-model based on the identifier of the first sub-model, so as to reduce the overhead of indication of the model structure.

[0082] In a possible implementation of the second aspect, the first information is further used to indicate at least one of the following:

[0083] a model parameter of the first sub-model;

[0084] an input data set of the first sub-model;

[0085] an output data set of the first sub-model;

[0086] an input position of the input data set of the first sub-model in one or more model units included in the first sub-model;

[0087] an output position of the output data set of the first sub-model in one or more model units included in the first sub-model.

[0088] According to the above scheme, in addition to indicating the first sub-model, the first information can be used to indicate the at least one, so that the first communication device can implement processing on the first sub-model based on the at least one.

[0089] In a possible implementation of the second aspect, the first model includes at least two model units, and each sub-model in the at least two sub-models includes one or more model units in the at least two model units; wherein the first sub-model includes M model units in the at least two model units, M is a positive integer; and the first information includes an identifier of the M model units.

[0090] According to the above scheme, the first model can include at least two model units, and each sub-model in the at least two sub-models included in the first model can include one or more model units in the at least two model units. Thus, the second communication device can indicate the first sub-model by the identifier of the M model units included in the first sub-model, so that the first communication device can quickly determine the first sub-model based on the identifier, and the overhead of indicating the first sub-model can be reduced.

[0091] In a possible implementation of the second aspect, the first information is further used to indicate a connection relationship of the M model units.

[0092] According to the above scheme, in addition to indicating the M model units included in the first sub-model by the identifier of the M model units, the first information can be used to indicate the connection relationship of the M model units, so that the first communication device can determine the first sub-model based on the connection relationship and the M model units.

[0093] In a possible implementation of the second aspect, the connection relationship of the at least two model units is pre-configured.

[0094] Based on the above scheme, in the at least two model units included in the first model, the connection relationship of the at least two model units is preconfigured. In this way, the overhead of indicating the connection relationship of the at least two model units (for example, indicating the connection relationship of M model units) can be saved.

[0095] In a possible implementation of the second aspect, the first information is further used to indicate at least one of:

[0096] model parameters of part or all of the M model units;

[0097] configuration parameters of part or all of the M model units;

[0098] input data sets of part or all of the M model units;

[0099] output data sets of part or all of the M model units.

[0100] Based on the above scheme, in addition to indicating the first sub-model, the first information can be used to indicate the at least one, so that the first communication device can implement processing of the first sub-model based on the at least one.

[0101] In a possible implementation of the second aspect, the method further includes: the second communication device receiving indication information used to indicate a local data set of the first communication device; and wherein the local data set is used to determine the first model.

[0102] Based on the above scheme, the second communication device can also receive indication information used to indicate the local data set from the first communication device, so that the second communication device can determine the first model based on the local data set of the first communication device.

[0103] In a possible implementation of the second aspect, the first information is further used to indicate sending of a processing result of the first sub-model.

[0104] Based on the above scheme, the first communication device can also indicate sending of the second information based on the indication, so that the second communication device can obtain the model processing result (that is, the second sub-model) of the first communication device based on the second information, so that the second communication device can perform a model management (for example, the second communication device performs model aggregation, model updating, etc. based on the second sub-model) process based on the second sub-model.

[0105] Optionally, the second communication device can also indicate sending of the processing result of the first sub-model in other ways, for example, the second communication device indicates sending of the processing result of the first sub-model through other information / messages / signaling different from the first information.

[0106] In a possible implementation of the second aspect, the second communication device receiving the second information comprises: the second communication device receiving the second information at or after the first time point, wherein a difference between the first time point and a second time point is determined by the time offset information, and the second time point is associated with the first sub-model.

[0107] Based on the above scheme, the first time point can be a time point at which the first communication device supervises processing of the first sub-model by the first communication device, and the first time point can be before an end time point of the processing of the first sub-model by the first communication device, for example, the first communication device can supervise during the training of the first sub-model. In this way, the first communication device can send the second information at or after the first time point, so that the second communication device can receive the second information as soon as possible before the end time point of the training, thereby reducing the overall processing delay of the plurality of first communication devices on different sub-models.

[0108] For example, if the first communication device finds that the training performance is lower than the target threshold in advance, the second communication device can learn this information as soon as possible through the above process, so that the second communication device can select other communication devices for training to reduce the overall processing delay of the plurality of first communication devices on different sub-models.

[0109] In addition, the data of the processing of the sub-model by different communication devices can have timeliness, and in the above scheme, the overall processing delay of the plurality of first communication devices on different sub-models is reduced, and the data of each communication device can also be avoided to a certain extent. invalid.

[0110] It should be understood that the first time point can be a time point at which the first communication device obtains / determines / obtains the performance of the second sub-model, or the first time point can be a time point at which the first communication device obtains / determines / obtains the difference between the performance of the second sub-model and the threshold. For example, the first time point can be referred to as a supervision time point, a detection time point, etc.

[0111] In a possible implementation of the first aspect, the second communication device receiving the second information comprises: the second communication device receiving the second information at or before the third time point, wherein a difference between the third time point and a second time point is determined by the time offset information, and the second time point is associated with the first sub-model.

[0112] Based on the above scheme, the third time point can be a deadline for the first communication device to process the first sub-model. For example, the first communication device can perform supervision before or at the deadline for training the first sub-model. In this way, the first communication device can send the second information at or after the first time point, so that the second communication device can receive the second information as soon as possible before the training end time point, thereby reducing the overall processing delay of multiple first communication devices for different sub-models.

[0113] For example, if the first communication device discovers that the training performance is lower than the target threshold in advance, the second communication device can learn this information as soon as possible, and the second communication device can select other communication devices for training.

[0114] Optionally, the deadline can be replaced by other descriptions, such as a termination time point, or a last time unit (such as a symbol, a subframe, a frame, etc.).

[0115] In a possible implementation of the second aspect, the method further includes: the second communication device sending third information, the third information being used to indicate the time offset information.

[0116] Based on the above scheme, the second communication device can also send third information to the first communication device, so that the first communication device determines the time offset information based on the received third information, and then the first communication device can send the second information based on the specified time offset information.

[0117] In a possible implementation of the second aspect, the method further includes: the second communication device receiving fourth information, the fourth information being used to indicate the second time point, the second time point being a start time point of training the first sub-model.

[0118] Based on the above scheme, the second communication device can also receive fourth information indicating the second time point, so that the second communication device can determine the corresponding time offset information based on the start time point of training the first sub-model, to assist the decision of the receiver.

[0119] In a possible implementation of the second aspect, the time offset information is preconfigured.

[0120] Based on the above scheme, the time offset information used to determine the sending time point of the second information can be preconfigured, which can save the configuration overhead of the time offset information.

[0121] In a possible implementation of the second aspect, the second time point is associated with the first sub-model, including: the second time point is a receiving time point of the first information, or the second time point is a start time point of training the first sub-model.

[0122] Based on the above scheme, the second time can be implemented in the above-mentioned multiple ways to improve the flexibility of the scheme implementation.

[0123] In a possible implementation of the second aspect, the second communication device receives the second information at the first time or after the first time, including: receiving the second information at the first time or after the first time, in a case that a first condition is met; and wherein the first condition includes at least one of: the model performance of the second sub-model is higher than a threshold, or the second communication device sends fifth information, the fifth information being used to indicate termination of processing of the first sub-model.

[0124] Based on the above scheme, the first communication device can also send the second information based on the first condition to improve the model processing efficiency. For example, in a case that the first condition includes that the model performance of the second sub-model is higher than a threshold, the first communication device can send the second information when the performance of the second sub-model is higher, so that the receiver of the second information can obtain the sub-model with higher performance. For another example, in a case that the first condition includes that the first communication device receives the fifth information, the first communication device can send the second information based on the termination processing indication of the opposite end.

[0125] Optionally, the second information can include a reason value, the reason value being used to indicate the first condition.

[0126] Optionally, the fifth information can be triggered based on sixth information, that is, the second communication device can send the fifth information to the first communication device based on an abnormal situation indicated by the sixth information after receiving the sixth information. The implementation of the sixth information can refer to the description below.

[0127] In a possible implementation of the second aspect, the second communication device receives the sixth information at the first time or after the first time, in a case that a second condition is met, the sixth information being used to indicate that the first sub-model processing is abnormal; and wherein the second condition includes at least one of: the model performance of the second sub-model is lower than a threshold, or the processing capability of the first communication device does not match the first sub-model.

[0128] Based on the above scheme, the first communication device can also send the sixth information based on the second condition, so that the scheme can adapt to the scene of model processing abnormality. For example, in a case that the second condition includes that the model performance of the second sub-model is lower than a threshold, the first communication device can send the sixth information when the performance of the second sub-model is lower, so that the receiver of the sixth information can learn about the abnormal situation of the model processing of the first communication device as soon as possible. For another example, in a case that the second condition includes that the processing capability of the first communication device does not match the first sub-model, the receiver of the sixth information can learn about the abnormal situation that the processing capability of the first communication device does not match the indicated first sub-model as soon as possible.

[0129] Optionally, the sixth information can comprise a cause value, the cause value being used to indicate the second condition.

[0130] Optionally, the process that the first communication device sends the second information is an optional step. For example, in the case that the first communication device sends the sixth information, it is possible that the first communication device does not obtain the second sub-model based on the first sub-model, and therefore the first communication device can not be able to send the second information indicating the second sub-model. Or, in the case that the first communication device sends the sixth information, it is possible that the first communication device does not obtain the second sub-model with better performance based on the first sub-model, and therefore the first communication device does not need to send the second information, which can avoid the second communication device processing based on the second sub-model with too low performance, and can reduce the overhead.

[0131] Optionally, in the case that the first communication device sends the sixth information, the first communication device can also send the second information. Since the second sub-model indicated by the second information can be obtained by the first communication device based on its own computing power, data set, etc., therefore even if the performance of the second sub-model is poor, the second communication device can still process based on the second sub-model, and the local computing power and / or local data set of one or more first communication devices can be utilized as much as possible.

[0132] The third aspect of the present application provides a communication device, which is a first communication device, comprising a transceiver unit and a processing unit; the transceiver unit is configured to receive first information, the first information being used to indicate a first sub-model, the first sub-model being one of sub-models in a first model, the first model comprising at least two sub-models; the processing unit is configured to process the first sub-model to obtain a second sub-model; the processing unit determines second information; the transceiver unit is further configured to send the second information, the second information being used to indicate the second sub-model; wherein the first sub-model is obtained by other communication devices through model processing, and / or the second sub-model is used for model processing of other communication devices.

[0133] In the third aspect of the present application, the component modules of the communication device can also be used to perform the steps performed in the various possible implementation manners of the first aspect, and achieve the corresponding technical effects, which can be referred to the first aspect for details, and will not be described here.

[0134] The fourth aspect of the present application provides a communication device, which is a second communication device, comprising a transceiver unit and a processing unit, the processing unit being configured to determine first information; the transceiver unit being configured to send the first information, the first information being used to indicate a first sub-model, the first sub-model being one of sub-models in a first model, the first model comprising at least two sub-models; wherein the first sub-model is used to be processed by a first communication device to obtain a second sub-model; the transceiver unit is further configured to receive second information, the second information being used to indicate the second sub-model; wherein the first sub-model is obtained by other communication devices through model processing, and / or the second sub-model is used for model processing of the other communication devices.

[0135] In the fourth aspect of the present application, the constituent modules of the communication device can also be used to perform the steps performed in the various possible implementation manners of the second aspect and achieve the corresponding technical effects, which can be referred to the second aspect for details and will not be described here.

[0136] The fifth aspect of the present application provides a communication device, comprising at least one processor coupled with a memory; the memory is used to store programs or instructions; the at least one processor is used to execute the programs or instructions, so that the device implements the method in any one of the possible implementation manners of the first aspect to the second aspect. Optionally, the communication device can comprise the memory.

[0137] The sixth aspect of the present application provides a communication device, comprising at least one logic circuit and an input-output interface; the logic circuit is used to execute the method in any one of the possible implementation manners of the first aspect to the second aspect.

[0138] The seventh aspect of the present application provides a communication system, comprising the first communication device and the second communication device.

[0139] The eighth aspect of the present application provides a computer readable storage medium, which is used to store one or more computer execution instructions, when the computer execution instructions are executed by a processor, the processor executes the method in any one of the possible implementation manners of the first aspect to the second aspect.

[0140] The ninth aspect of the present application provides a computer program product (or computer program), when the computer program in the computer program product is executed by a processor, the processor executes the method in any one of the possible implementation manners of the first aspect to the second aspect.

[0141] The tenth aspect of the present application provides a chip or a chip system, which comprises at least one processor for supporting a communication device to implement the method of any possible implementation manner of any one of the first aspect to the second aspect. For example, the chip can be a baseband chip, a modem chip, an SoC chip (such as an SoC chip containing a modem core), a SIP chip, or a communication module, etc.

[0142] In a possible design, the chip or the chip system can further comprise a memory for storing necessary program instructions and data of the communication device. The chip system can be composed of a chip, or can contain a chip and other discrete devices. Optionally, the chip system further comprises an interface circuit for providing program instructions and / or data for the at least one processor.

[0143] The technical effects brought by any design manner in the third aspect to the tenth aspect can be referred to the technical effects brought by different design manners in the first aspect to the second aspect, which will not be repeated here. BRIEF DESCRIPTION OF DRAWINGS

[0144] Figures la-lc A schematic diagram of a communication system provided by the present application is shown;

[0145] Figures 2a-2g A schematic diagram of an AI processing process related to the present application is shown;

[0146] Figure 3 An interaction schematic diagram of a communication method provided by the present application is shown;

[0147] Figures 4a-4j Some schematic diagrams of a model provided by the present application are shown;

[0148] Figures 5-9 A schematic diagram of a communication device provided by the present application is shown. DETAILED DESCRIPTION

[0149] First, some terms in the embodiments of the present application are explained to facilitate understanding by those skilled in the art.

[0150] (1) Terminal device: can be a wireless terminal device capable of receiving network device scheduling and indication information, and the wireless terminal device can be a device providing voice and / or data connectivity to a user, or a handheld device with wireless connection function, or other processing devices connected to a wireless modem.

[0151] A terminal device can communicate with one or more core networks or the Internet via a radio access network (RAN), and the terminal device can be a mobile terminal device, such as a mobile phone (or called "cellular" phone, mobile phone), a computer, and a data card, for example, which can be a portable, pocket, hand-held, computer- built-in, or vehicle-mounted mobile device that exchanges voice and / or data with a radio access network. For example, a personal communication service (PCS) phone, a cordless phone, a session initiation protocol (SIP) phone, a wireless local loop (WLL) station, a personal digital assistant (PDA), a tablet, a computer with wireless transceiver, and the like. The wireless terminal device can also be referred to as a system, a subscriber unit, a subscriber station, a mobile station (MS), a remote station, an access point (AP), a remote terminal, an access terminal, a user terminal, a user agent, a subscriber station (SS), customer premises equipment (CPE), a terminal, user equipment (UE), a mobile terminal (MT), and the like.

[0152] By way of example and not limitation, in embodiments of the present application, the terminal device can also be a wearable device. The wearable device can also be referred to as a smart wearable device or a smart wearable device, etc., which is a general term for devices that can be designed and developed by applying wearable technology to daily wear, such as glasses, gloves, watches, clothing, and shoes, etc. The wearable device is a portable device that can be directly worn on the body or integrated into the user's clothes or accessories. The wearable device is not only a hardware device, but also a powerful function realized through software support and data interaction, cloud interaction. The general wearable smart device includes a full function, large size, and can realize complete or partial functions without relying on a smart phone, such as smart watches or smart glasses, etc., and focuses on a certain application function, and needs to cooperate with other devices such as a smart phone, such as various smart wristbands, smart helmets, smart jewelry, etc. for monitoring vital signs.

[0153] The terminal can also be a drone, a robot, a terminal in device-to-device (D2D) communication, a terminal in vehicle to everything (V2X), a virtual reality (VR) terminal device, an augmented reality (AR) terminal device, a wireless terminal in industrial control, a wireless terminal in self driving, a wireless terminal in telemedicine or telehealth services, a wireless terminal in smart grid, a wireless terminal in transportation safety, a wireless terminal in smart city, a wireless terminal in smart home, etc.

[0154] In addition, the terminal device can also be a terminal device in a communication system evolved after the 5th generation (5G) communication system (such as 5G Advanced or 6th generation (6G) communication system, etc.) or a terminal device in a future evolved public land mobile network (PLMN), etc. For example, 5G Advanced or 6G network can further expand the form and function of 5G communication terminal, and 6G terminal includes but is not limited to vehicle, cellular network terminal (integrating satellite terminal function), drone, internet of things (IoT) device.

[0155] In the embodiments of the present application, the terminal device can also obtain an artificial intelligence (AI) service provided by the network device. Optionally, the terminal device can also have AI processing capability.

[0156] (2) Network device: can be a device in a wireless network, for example, the network device can be a RAN node (or device) for accessing the terminal device to the wireless network, which can also be referred to as a base station. At present, some examples of RAN devices are: base station, evolved NodeB (eNodeB), base station gNB (gNodeB) in 5G communication system, transmission reception point (TRP), evolved Node B (eNB), radio network controller (RNC), Node B (NB), home base station (for example, home evolved Node B, or home Node B, HNB), baseband unit (BBU) or wireless fidelity (Wi-Fi) access point (AP) and the like. In addition, in a network structure, the network device can include a central unit (CU) node, or a distributed unit (DU) node, or a RAN device including a CU node and a DU node.

[0157] Optionally, the RAN node can also be a macro base station, a micro base station or an indoor station, a relay node or a donor node, or a wireless controller in a cloud radio access network (CRAN) scenario. The RAN node can also be a server, a wearable device, a vehicle or a vehicle-mounted device, etc. For example, the access network device in vehicle-to-everything (V2X) technology can be a road side unit (RSU).

[0158] In another possible scenario, a terminal is assisted by multiple RAN nodes to implement wireless access, and different RAN nodes respectively implement part of functions of a base station. For example, a RAN node can be a CU, a DU, a CU-control plane (CP), a CU-user plane (UP), or a radio unit (RU), etc. The CU and the DU can be separately arranged, or can be included in the same network element, for example, in a baseband unit (BBU). The RU can be included in a radio frequency device or a radio frequency unit, for example, in a remote radio unit (RRU), an active antenna unit (AAU), a radio head (RH), or a remote radio head (RRH).

[0159] In different systems, the CU (or CU-CP and CU-UP), the DU, or the RU can also have different names, but those skilled in the art can understand their meanings. For example, in an open RAN (O-RAN or ORAN) system, the CU can also be referred to as an O-CU (open CU), the DU can also be referred to as an O-DU, the CU-CP can also be referred to as an O-CU-CP, the CU-UP can also be referred to as an O-CU-UP, and the RU can also be referred to as an O-RU. For the convenience of description, the CU, the CU-CP, the CU-UP, the DU, and the RU are taken as examples for description in this application. Any one of the CU (or the CU-CP, the CU-UP), the DU, and the RU in this application can be implemented by a software module, a hardware module, or a combination of a software module and a hardware module.

[0160] Communication between the access network device and the terminal device follows a certain protocol layer structure. The protocol layer can include a control plane protocol layer and a user plane protocol layer. The control plane protocol layer can include at least one of the following: a radio resource control (RRC) layer, a packet data convergence protocol (PDCP) layer, a radio link control (RLC) layer, a media access control (MAC) layer, or a physical (PHY) layer, etc. The user plane protocol layer can include at least one of the following: a service data adaptation protocol (SDAP) layer, a PDCP layer, an RLC layer, a MAC layer, or a physical layer, etc.

[0161] For the correspondence between the network elements in the ORAN system and the protocol layer functions that can be implemented by the network elements, refer to Table 1 below.

[0162] Table 1

[0163] ORAN network elements 3GPP protocol layer functions O-CU-CP RRC+PCDP - control plane (PDCP-C) O-CU-UP SDAP+PCDP - user plane (PDCP-U) O-DU RLC+MAC+PHY-high O-RU PHY-low

[0164] The network device can be another device that provides a wireless communication function for the terminal device. Embodiments of the present application do not limit the specific technology and specific device form adopted by the network device. For the convenience of description, embodiments of the present application do not limit.

[0165] The network device can also include a core network device, for example, a mobility management entity (MME) in a fourth generation (4G) network, a home subscriber server (HSS), a serving gateway (S-GW), a policy and charging rules function (PCRF), a public data network gateway (PDN gateway or P-GW), an access and mobility management function (AMF), a user plane function (UPF), or a session management function (SMF) in a 5G network, and other network elements. In addition, the core network device can also include other core network devices in a 5G network and a next-generation network of the 5G network.

[0166] In embodiments of the present application, the network device described above can also be an AI-capable network node that can provide AI services for terminals or other network devices, for example, AI nodes, computing power nodes, AI-capable RAN nodes, AI-capable core network elements, and the like on the network side (access network or core network).

[0167] In embodiments of the present application, the device for implementing the function of the network device can be a network device or a device capable of supporting the network device to implement the function, such as a chip system, which can be arranged in the network device. In the technical solutions provided in embodiments of the present application, the device for implementing the function of the network device is taken as an example to describe the technical solutions provided in embodiments of the present application.

[0168] (3) Configuration and pre-configuration: in this application, both configuration and pre-configuration will be used. Among them, configuration refers to that the network device / server sends some parameter configuration information or parameter values to the terminal through messages or signaling, so that the terminal determines the communication parameters or resource in transmission according to the values or information. Pre-configuration is similar to configuration, which can be parameter information or parameter values agreed by the network device / server and the terminal device in advance, or parameter information or parameter values adopted by the base station / network device or the terminal device according to the standard protocol, or parameter information or parameter values pre-stored in the base station / server or the terminal device. This application does not limit it.

[0169] Further, these values and parameters can be changed or updated.

[0170] (4) The terms "system" and "network" in the embodiments of the present application can be used interchangeably. "Multiple" means two or more. "And / or" describes the association between the associated objects, which means that there can be three relationships, for example, A and / or B, which means that A exists alone, A and B exist together, and B exists alone, where A and B can be singular or plural. The character " / " generally represents an "or" relationship between the associated objects. "At least one of the following" or similar expressions means any combination of these items, including any combination of single item or multiple items. For example, "at least one of A, B and C" includes A, B, C, AB, AC, BC or ABC. In addition, unless otherwise specified, the ordinal numbers "first", "second", etc. mentioned in the embodiments of the present application are used to distinguish multiple objects, and are not used to limit the order, time sequence, priority or importance of the multiple objects.

[0171] (5) In the embodiments of the present application, "sending" and "receiving" represent the direction of signal transmission. For example, "sending information to XX" can be understood as that the destination of the information is XX, which can include direct sending through the air interface, or indirect sending through the air interface by other units or modules. "Receiving information from YY" can be understood as that the source of the information is YY, which can include direct receiving from YY through the air interface, or indirect receiving from YY through the air interface by other units or modules. "Sending" can also be understood as the "output" of the chip interface, and "receiving" can also be understood as the "input" of the chip interface.

[0172] In other words, sending and receiving can be carried out between devices, such as between network devices and terminal devices, or within devices, such as between components, modules, chips, software modules or hardware modules within a device through buses, wires or interfaces.

[0173] It can be understood that the information can be processed, such as encoding and modulation, between the source end and the destination end of the information transmission, but the destination end can understand the effective information from the source end. Similar expressions in this application can be similarly understood, and will not be repeated here.

[0174] (6) In the embodiments of the present application, "indication" can include direct indication and indirect indication, and can also include explicit indication and implicit indication. The information indicated by a certain information (indication information described below) is referred to as to-be-indicated information. In the implementation process, there are many ways to indicate the to-be-indicated information, for example, but not limited to, the to-be-indicated information can be directly indicated, such as the to-be-indicated information itself or the index of the to-be-indicated information. The to-be-indicated information can also be indirectly indicated by indicating other information, where the other information and the to-be-indicated information have an association relationship. The to-be-indicated information can also be indicated only by a part, and the other part of the to-be-indicated information is known or agreed in advance. For example, the indication of a specific information can be achieved by means of the arrangement order of each information agreed in advance (for example, protocol predefined), thereby reducing the indication overhead to a certain extent. The specific manner of indication is not limited in the present application. It can be understood that the indication information can be used to indicate the to-be-indicated information for the sender of the indication information, and the indication information can be used to determine the to-be-indicated information for the receiver of the indication information.

[0175] In the present application, the same or similar parts of each embodiment can be mutually referred to, unless otherwise specified. In the various embodiments of the present application, and the various methods / designs / implementation manners in each embodiment, the terms and / or descriptions of different embodiments, and the various methods / designs / implementation manners in each embodiment are consistent and can be mutually referred to, unless otherwise specified and logically conflicted. The technical features of different embodiments, and the various methods / designs / implementation manners in each embodiment can be combined to form new embodiments, methods, or implementation manners according to their inherent logical relationship. The implementation manners of the present application described below do not constitute a limitation on the protection scope of the present application.

[0176] The present application can be applied to a long term evolution (LTE) system, a new radio (NR) system, or a communication system evolved after 5G (such as 6G, etc.). The communication system includes at least one network device and / or at least one terminal device.

[0177] Please refer to Figure la , which is a schematic diagram of a communication system in the present application. Figure laIn the example shown, the network device and the six terminal devices are respectively terminal device 1, terminal device 2, terminal device 3, terminal device 4, terminal device 5 and terminal device 6. Figure la In the example shown, the terminal device 1 is an intelligent tea cup, the terminal device 2 is an intelligent air conditioner, the terminal device 3 is an intelligent fuel dispenser, the terminal device 4 is a vehicle, the terminal device 5 is a mobile phone, and the terminal device 6 is a printer.

[0178] As shown in Figure la , the sending entity of the AI configuration information can be a network device. The receiving entity of the AI configuration information can be terminal device 1-terminal device 6. At this time, the network device and the terminal device 1-terminal device 6 form a communication system, in which the terminal device 1-terminal device 6 can send data to the network device, and the network device receives the data sent by the terminal device 1-terminal device 6. The network device can send configuration information to the terminal device 1-terminal device 6.

[0179] As shown in Figure la , the terminal device 4-terminal device 6 can also form a communication system. Among them, the terminal device 5 acts as a network device, that is, the sending entity of the AI configuration information; the terminal device 4 and the terminal device 6 act as terminal devices, that is, the receiving entity of the AI configuration information. For example, in a vehicle networking system, the terminal device 5 sends AI configuration information to the terminal device 4 and the terminal device 6, and receives data sent by the terminal device 4 and the terminal device 6; correspondingly, the terminal device 4 and the terminal device 6 receive the AI configuration information sent by the terminal device 5, and send data to the terminal device 5.

[0180] As shown in Figure la , in the communication system, different devices (including network devices, network devices and terminal devices, and / or terminal devices and terminal devices) can perform AI-related services in addition to performing communication-related services.

[0181] As shown in Figure lb , 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.

[0182] As shown in Figure lc , taking a terminal device including a television and a mobile phone as an example, the television and the mobile phone can also perform communication-related services and AI-related services.

[0183] The technical solutions provided in the present application can be applied to a wireless communication system (for example Figure la , Figure lb orFigure lc For example, the AI network element can be introduced in the communication system provided in the present application to implement part or all of the AI related operations. The AI network element can also be referred to as an AI node, an AI device, an AI entity, an AI module, an AI model, or an AI unit, etc. The AI network element can be built-in in a network element of the communication system. For example, the AI network element can be an AI module built-in in an access network device, a core network device, a cloud server, or an operation, administration and maintenance (OAM) to implement AI related functions. The OAM can be a network management of the core network device and / or a network management of the access network device. Alternatively, the AI network element can also be a network element independently arranged in the communication system. Optionally, an AI entity can also be included in a terminal or a chip built-in in the terminal to implement AI related functions.

[0184] The concepts that can be involved in the present application will be briefly introduced below.

[0185] AI can enable a machine to have human intelligence, for example, the machine can apply software and hardware of a computer to simulate some intelligent behaviors of a human being. In order to realize artificial intelligence, a machine learning method can be adopted. In the machine learning method, the machine learns (or trains) a model by using training data. The model represents the mapping between the input and the output. The learned model can be used for inference (or prediction), that is, the model can be used to predict the output corresponding to a given input. The output can also be referred to as an inference result (or a prediction result).

[0186] Machine learning can include supervised learning, unsupervised learning, and reinforcement learning. The unsupervised learning can also be referred to as non-supervised learning.

[0187] Supervised learning learns the mapping relationship from sample values to sample labels by using a machine learning algorithm according to the collected sample values and sample labels, and uses an AI model to express the learned mapping relationship. The process of training the machine learning model is the process of learning the mapping relationship. In the training process, the sample values are input into the model to obtain the predicted values of the model, and the model parameters are optimized by calculating the error between the predicted values of the model and the sample labels (ideal values). After the mapping relationship is learned, the learned mapping can be used to predict new sample labels. The learned mapping relationship of the supervised learning can include linear mapping or non-linear mapping. According to the type of label, the learned task can be divided into classification task and regression task.

[0188] 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.

[0189] 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.

[0190] 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.

[0191] 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.

[0192] like Figure 2a The diagram shown is a schematic representation of a neuron structure. Assume the neuron's input is x = [x0, x1, ..., x...]. n The weights corresponding to each input are w = [w0, w1, ..., w] 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 The inputs are weighted according to the weights. The bias for the weighted sum of the inputs is, for example, b. The form of the activation function can be various. Assuming that the activation function of a neuron is y = f(z) = max(0, z), the output of the neuron is: For another example, the activation function of a neuron is y = f(z) = z, the output of the neuron is: where b can be various possible types such as a decimal, an integer (for example, 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.

[0193] In addition, a neural network generally includes multiple layers, and each layer can include one or more neurons. By increasing the depth and / or width of the neural network, the expressive ability of the neural network can be improved, and the neural network can provide stronger information extraction and abstract modeling capabilities for complex systems. The depth of the neural network can refer to the number of layers included in the neural network, and the number of neurons included in each layer can be referred to as the width of the layer. In an implementation manner, the neural network includes an input layer and an output layer. The input layer of the neural network processes the received input information through neurons, and transmits the processing result to the output layer, and the output layer obtains the output result of the neural network. In another implementation manner, the neural network includes an input layer, a hidden layer, and an output layer. The input layer of the neural network processes the received input information through neurons, and transmits the processing result to the intermediate hidden layer. The hidden layer calculates the received processing result to obtain a calculation result, and transmits the calculation result to the output layer or the next adjacent hidden layer, and finally the output layer obtains the output result of the neural network. The neural network can include one hidden layer, or include multiple sequentially connected hidden layers, which is not limited.

[0194] The neural network is, for example, a deep neural network (DNN). According to the construction manner of the network, the DNN can include a feedforward neural network (FNN), a convolutional neural network (CNN), and a recurrent neural network (RNN).

[0195] Figure 2b An FNN network is shown in a schematic diagram. The characteristic of the FNN network is that the neurons of adjacent layers are completely connected two by two. This characteristic makes the FNN usually need a large amount of storage space, and leads to a high calculation complexity.

[0196] 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.

[0197] 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.

[0198] 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.

[0199] 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.

[0200] The implementation process of the neural network will be described below with reference to the accompanying drawings.

[0201] 1. Fully connected neural network, also known as multilayer perceptron (MLP).

[0202] like Figure 2c As shown, 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.

[0203] Optionally, considering neurons in two adjacent layers, the output h of a neuron in the next layer 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:

[0204] h = f(wx + b).

[0205] Where w is the weight matrix, b is the bias vector, and f is the activation function.

[0206] Alternatively, the output of the neural network can be recursively expressed as:

[0207] y = f z (w z f z-1 (…)+b z ).

[0208] 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.

[0209] 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.

[0210] Optionally, the training method involves using a loss function to evaluate the output of the neural network.

[0211] like Figure 2d As shown, 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. Figure 2d The term "relative advantage (e.g., optimal advantage)" is used. This is understandable. Figure 2d The neural network parameters corresponding to the "better points (e.g., the best points)" in the data can be used as neural network parameters in the trained AI model information.

[0212] Alternatively, the gradient descent process can be represented as:

[0213]

[0214] 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.

[0215] Alternatively, the backpropagation process can utilize the chain rule for partial derivatives.

[0216] likeFigure 2e As shown, the gradient of the previous layer parameters can be recursively calculated by the gradient of the next layer parameters, which can be expressed as:

[0217]

[0218] where w ij is the weight of node j connecting node i, s i is the input weighted sum on node i.

[0219] 2. Federated Learning (FL).

[0220] The concept of federated learning effectively solves the difficulties faced by the current development of artificial intelligence. Under the premise of fully guaranteeing the privacy and security of user data, it promotes the cooperation of various edge devices and central servers to efficiently complete the learning task of the model.

[0221] 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 algorithm process of FedAvg is roughly as follows:

[0222] (1) The central end initializes the model to be trained and broadcasts it to all clients.

[0223] (2) In the tthround t∈[1,T], the client k∈[1,K] trains the received global model for E epochs based on the local data set to obtain the local training result and reports it 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 n , G k , and G m , respectively.

[0224] (3) The central node collects the local training results from all (or part) of the clients. Assuming that the client set that uploads the local model in the tthround is The central end weights the new global model by the number of samples of the corresponding client to obtain the new global model, and the specific update rule is After that, the central end broadcasts the latest version of the global model to all clients for a new round of training.

[0225] (4) Repeat steps (2) and (3) until the model converges or the number of training rounds reaches the upper limit.

[0226] Optionally, the client reports the local model The trained local gradient may also be reported, and the center node averages all the local gradients reported by the clients and updates the global model according to the average gradient.

[0227] As can be seen, in the FL framework, the data set exists at the distributed nodes (such as the clients), that is, the distributed nodes collect the local data set and perform local training, and report the local result (model or gradient) obtained by training to the center node. The center node itself can have no data set and can be responsible for fusing the training results of the distributed nodes to obtain a global model and issuing the global model to the distributed nodes.

[0228] 3. Decentralized learning.

[0229] As shown in Figure 2g , it is a completely distributed system without a center node. The design goal f(x) of the decentralized learning system is generally the average of the goals f i (x) of each node, that is, where n is the number of distributed nodes, and x is the parameter to be optimized. In machine learning, x is the parameter of a machine learning (such as a neural network) model. Each node calculates the local gradient i (x) using local data and local goal f and sends it to the neighbor nodes that are communicatively reachable. After receiving the gradient information sent by the neighbor nodes, any node can update the parameter x of the local model according to the following formula:

[0230]

[0231] where, represents the parameter of the local model updated for the k+1th time (k is a natural number) in the ith node, represents the parameter of the local model updated for the kth time in the ith node (if k is 0, it represents the parameter of the local model that has not been updated in the ith node), and a k represents an adjustment coefficient, N i is the neighbor node set of node i, and |N i | represents the number of elements in the neighbor node set of node i, that is, the number of neighbor nodes of node i. Through the information interaction between nodes, the decentralized learning system will finally learn a unified model.

[0232] The technical solutions provided in the present application can be applied in a communication system (for example Figure la or Figure lb or Figure lcIn the shown system), in a communication system, a communication node generally has signal transceiving capability and computing capability. Taking a network device with computing capability as an example, the computing capability of the network device is mainly to provide computing power support for the signal transceiving capability (for example: to perform sending processing and receiving processing on signals), so as to realize the communication task of the network device and other communication nodes.

[0233] With the development of communication technology, in a communication system, in addition to traditional communication services, a communication device can also perform other new services, such as artificial intelligence (AI) services. Generally, a communication system capable of processing AI services can also be referred to as an AI system.

[0234] At present, a communication device can serve as a participating node of an AI system and provide its own computing power and data. For example, the communication device can perform a plurality of model processing (such as model training, model updating, or model fine-tuning) processes on a model based on local data to obtain another model. In one possible manner, increasing the complexity of the model (such as increasing the number of parameters of the model or increasing the number of neural network layers included in the model) can effectively improve the performance of the model.

[0235] However, in the case where the complexity of the model gradually increases, the above-mentioned model processing process may consume a relatively long time, and even the above-mentioned model processing process may not be able to be performed due to the limitation of the resources (such as computing power resources and storage resources) of the communication device. Therefore, how to optimize the model processing process is a technical problem to be solved.

[0236] As an example, when the communication device trains based on general data, it is possible that the model is too matched with the training data set, and falls into an overfitting state, thereby reducing the generalization of the model in inference. That is, the model cannot adapt to the actual scene and / or task in inference. For example, the environment map used by the radio frequency map (RF MAP) model in training may be different from the actual environment map, and the position of the communication device used by the RF MAP model in training may also be different from the actual position of the communication device. If the training data is collected according to the actual physical environment, the overhead of data collection will be very large, and therefore it is necessary to further fine-tune the model according to the actual physical environment. In some scenarios, the resources (such as computing power resources or storage resources) of a single communication device are relatively limited, and when the trained model is large, a single communication device is difficult to independently complete the training. Considering the manner of cooperation of multiple communication devices, the model processing process can be optimized.

[0237] However, in the traditional cooperation process of multiple communication devices, as described in the foregoing Figure 2fIn the shown example, different communication devices can serve as distributed nodes, and different distributed nodes are all processed based on the same model. However, the resources of a single communication device are limited, and this optimization method may still cause the single communication device to be unable to execute the model processing process due to resource limitations.

[0238] To solve the above problems, the present application provides a communication method and related devices, which will be described in detail below with reference to the accompanying drawings.

[0239] Please refer to Figure 3 An implementation example of the communication method provided by the present application is shown in the figure, and the method includes the following steps.

[0240] It should be noted that in the following, Figure 3 In the above example, the first communication device and other communication devices (for example, the second communication device) are taken as an example to illustrate the execution subject of the interaction scenario, but the present application does not limit the execution subject of the interaction scenario. For example, the communication device can be a communication device (such as a terminal device or a network device), or a chip, a baseband chip, a modem chip, a SoC chip (such as a SoC chip containing a modem core), a SIP chip, a communication module, a chip system, a processor, a logic module or software, etc.

[0241] As an example, the first communication device can be a terminal device and the second communication device can be a network device.

[0242] As another example, the first communication device can be a network device and the second communication device can be a terminal device.

[0243] As another example, the first communication device and the second communication device are both terminal devices, that is, Figure 3 The above-mentioned solution can be applied to a sidelink communication scenario.

[0244] S301. The second communication device sends first information, and correspondingly, the first communication device receives the first information. The first information is used to indicate a first sub-model, and the first sub-model is one of the first model, and the first model includes at least two sub-models.

[0245] In addition, after step S301, the first communication device can process the first sub-model to obtain a second sub-model.

[0246] S302. The first communication device sends second information, and correspondingly, the second communication device receives the second information. The second information is used to indicate the second sub-model. The first sub-model is obtained by the other communication device through model processing, and / or the second sub-model is used for model processing of the other communication device.

[0247] In this application, the model can include a reference model, an AI model, a neural network model, an AI neural network model, a machine learning model, or an AI processing model. Similarly, the sub-model can include a sub-reference model, an AI sub-model, a neural network sub-model, an AI neural network sub-model, a machine learning sub-model, or an AI processing sub-model.

[0248] In this application, the communication device processes one sub-model to obtain another sub-model, and the processing can include one or more of training, fine-tuning, fine-tuning, updating, iteration, and optimization. For example, the first communication device processes the first sub-model through one or more of training, fine-tuning, fine-tuning, updating, iteration, and optimization to obtain the second sub-model.

[0249] Based on Figure 3 As shown in the scheme, the first information received by the first communication device in step S301 is used to indicate the first sub-model in the at least two sub-models contained in the first model. Thereafter, the first communication device can send second information in step S302, which is used to indicate the second sub-model obtained by processing based on the first sub-model. In other words, the first communication device can process one of the sub-models contained in the model based on the indication of the first information. Thus, compared with the process of the communication device processing the complete model (i.e. the first model), the process of the first communication device processing one of the sub-models contained in the model can reduce the complexity of the model processing process of the first communication device, while also reducing the model processing delay.

[0250] In addition, the number of first communication devices can be one or more. In the case where the number of first communication devices is greater than 1, the sender of the first information (e.g. the second communication device) can indicate the same or different sub-models to different first communication devices. In the distributed scenario where the different first communication devices participate in the model processing process as distributed nodes, compared with the process of each distributed node processing the complete model (i.e. the first model), the processing complexity of the distributed nodes can be reduced while the processing delay of the distributed nodes is also reduced, so as to improve the model processing efficiency in the distributed scenario.

[0251] Optionally, for a certain first communication device, the first communication device can be configured to process one or more sub-models. For example, in step S301, the first information received by the first communication device can indicate a first sub-model and one or more other sub-models, so that the second information sent by the first communication device in step S302 can indicate a second sub-model obtained based on the first sub-model and sub-models obtained based on the one or more other sub-models.

[0252] In addition, the first sub-model indicated by the first information is obtained by the other communication device through model processing, and / or the second sub-model indicated by the second information is used for model processing of the other communication device. In other words, the first sub-model processed by the first communication device can be a processing result of the other communication device, and / or the second sub-model obtained by the first communication device can be used for model processing of the other communication device. In this way, different communication devices can process sub-models contained in the model in a chain (or series) manner to realize cooperation of multiple communication devices.

[0253] In a possible implementation, before step S301, the first communication device sends capability information of the first communication device; and the capability information is used to determine the first information. In this way, the sender of the first information (for example, the second communication device) can determine the first sub-model matched with the capability information of the first communication device, so that the first communication device can perform model processing based on the sub-model matched with its own capability, and model processing failure caused by sub-model and capability mismatch can be avoided, so as to improve the success rate of model processing.

[0254] Optionally, the capability information includes one or more of the following: computing power information (for example, the computing power information can indicate one or more of total computing power, used computing power, and idle computing power), storage information (for example, the storage information can indicate one or more of total storage space, used storage space, and idle storage space), communication information (for example, the communication information can indicate one or more of antenna information of the first communication device, communication chip information, and channel information between the first communication device and the other communication device), or other information.

[0255] In Figure 3 In a possible implementation of the method shown in FIG. 3, Figure 3 The method shown in FIG. 3 further includes: the first communication device sends indication information used to indicate a local data set; and the local data set is used to determine the first model. Thus, the receiver of the indication information (for example, the second communication device) can determine the first model based on the local data set of the first communication device.

[0256] Similarly, in the case that the number of the first communication devices is multiple, each of the multiple first communication devices can send the indication information for indicating the local data set to the second communication device, so that the receiver (e.g., the second communication device) of the indication information can determine the first model based on the local data set of the multiple first communication devices.

[0257] For example, the second communication device can perform initial training based on the local data set of one or more first communication devices to obtain an initial model (or an initial reference model, a total reference model, etc.), which is the first model.

[0258] For another example, the second communication device can fine-tune (or fine-tune, update, etc.) a pre-trained model based on the local data set of one or more first communication devices to obtain an initial model (or an initial reference model, a total reference model, etc.), which is the first model. The pre-trained model can be pre-configured or obtained by training based on a general data set.

[0259] Optionally, the local data set can be non-private data, de-identified data, etc. of the first communication device. In this way, it is beneficial to protect the data security of the first communication device and avoid privacy leakage.

[0260] In a possible implementation, in step S301, the first information can indicate the first sub-model in multiple ways, which will be described in combination with some possible implementations.

[0261] Implementation one, the first information includes an identifier of the first sub-model.

[0262] In implementation one, the sender (e.g., the second communication device) of the first information indicates the first sub-model by the identifier of the first sub-model, so that the first communication device can quickly determine the first sub-model based on the identifier, and the overhead of indicating the first sub-model can be reduced.

[0263] Optionally, the identifier of one or more sub-models is pre-defined with the model structure of one or more sub-models. In this way, the first communication device can quickly determine the model structure of the first sub-model based on the identifier of the first sub-model, and the overhead of indicating the model structure can be reduced.

[0264] In a possible implementation of implementation one, in step S301, the first information is further used to indicate at least one of the following:

[0265] Information 1. Model parameters (e.g., the model parameters can include frozen parameters and / or non-frozen parameters) of the first sub-model.

[0266] Information 2. An input data set of the first sub-model.

[0267] Information 3. An output data set of the first sub-model.

[0268] Information 4. An input position of the input data set of the first sub-model in one or more model units included in the first sub-model.

[0269] Information 5. An output position of the output data set of the first sub-model in one or more model units included in the first sub-model.

[0270] Therefore, in addition to indicating the identity of the first sub-model, the first information can also be used to indicate the at least one item, so that the first communication device can implement processing on the first sub-model based on the at least one item.

[0271] Optionally, the at least one item can be preconfigured.

[0272] Optionally, the second communication device can also indicate the at least one item in other ways, for example, the second communication device indicates the at least one item through other information / messages / signaling different from the first information.

[0273] In this application, the model unit can be replaced by other terms, such as reference unit, model element, model component, model component, or model reference unit, etc.

[0274] As an example, as shown in the scenario shown in FIG. 1, the first model can include four parts R1, R2, R3, and R4 as shown in FIG. 2, and the second communication device can split the four parts included in the first model into three sub-models, including: Figure 4a Figure 4a Sub-model 1: R1, R2;

[0275] Sub-model 2: R2, R3, R4;

[0276] Sub-model 3: R3, R4.

[0277] In addition, the input data set of the sub-model 1 is d1, and the output data set is d3; the input data set of the sub-model 2 is d2, and the output data set is d5; the input data set of the sub-model 3 is d3, and the output data set is d5.

[0278] In the above implementation manner one, the mapping relationship between the identity (ID) of the sub-model and the model structure (or model type) can be pre-defined, and the mapping relationship can be pre-defined by a table, a formula, etc. For example, taking the mapping relationship pre-defined by a table as an example, as shown in Table 2 below.

[0279] Table 2

[0280] Table 2 ​

[0281]

[0282] wherein, Transformer is a widely used architecture in the field of AI.

[0283] As an example, Figure 4b The overall architecture of the transformer is given, including the encoder part and the decoder part.

[0284] As another example, Figure 4c The model structure of an encoder block and a decoder block in the transformer is given.

[0285] Optionally, taking the first communication device as a terminal device and the second communication device as a network device as an example. The second communication device can configure the model structure of one or more sub-models included in the first model through the RRC parameter, that is, the second communication device can pre-configure the model structure of various supported sub-models to the first communication device, and when the model structure of the configured sub-model is more than one, the model structure of the current sub-model is activated through the DCI.

[0286] Thereafter, based on the mapping relationship between the ID of the pre-defined or pre-configured sub-model and the sub-model structure, the second communication device indicates the ID of the sub-model to the first communication device, for example, through the DCI, and the first communication device can know the structure corresponding to the model.

[0287] As can be known from the foregoing description, in the first implementation manner, the first information can be used to indicate at least one of the information 1 to information 5 described in the foregoing description in addition to indicating the first sub-model. More implementation examples will be described below.

[0288] As an implementation example, the first information can indicate the model parameters of the sub-model (i.e., the information 1 in the foregoing description).

[0289] In this implementation example, the first information can include frozen parameter information. Wherein, the frozen parameter information is used to indicate that part of the parameters in the sub-model are frozen, and the frozen part remains unchanged in the processing process (for example, the training process) of the first communication device. For example, when the sub-model ID is ID 3 in Table 2, part of the heads in the multi-head attention is frozen, or when the sub-model ID is ID 0 in Table 2, part of the encoder blocks in the encoder is frozen.

[0290] Optionally, the frozen portion can be indicated by a freeze ID. The definition of the freeze ID is related to the sub-model structure, and different sub-model structures correspond to different freeze indication methods. When the sub-model is a multi-head attention, the freeze ID is X, indicating that the Xth head in the multi-head attention is frozen. When the sub-model is an encoder or decoder, the freeze ID is X, indicating that the Xth encoder or decoder block in the encoder or decoder is frozen.

[0291] Optionally, the freeze parameter information can indicate the model parameters corresponding to the frozen portion.

[0292] In this implementation example, the first information may include indication information, which indicates whether the non-frozen portion includes model parameters. Specifically, for the non-frozen portion, the second communication device may or may not indicate the corresponding model parameters. If the second communication device indicates the corresponding model parameters, the first information can also be used to indicate the model parameters corresponding to the non-frozen portion, enabling the first communication device to obtain the model parameters through the first information and perform subsequent model processing.

[0293] Optionally, the model parameters of the sub-model may not include the frozen part, that is, all the model parameters of the sub-model are unfrozen.

[0294] As another implementation example, the first information can indicate the dataset corresponding to the sub-model (i.e., information 2 and information 3 mentioned above). Optionally, the first information can also indicate the location corresponding to the dataset (i.e., information 4 and information 5 mentioned above). Here, since the sub-model may correspond to multiple datasets, and the processing location of these multiple datasets is not limited to the input and output locations of the model.

[0295] like Figure 4d The example shown includes a certain model. Figure 4dFor example, the R1 module and the R2 module in the model, the input data of the model is a data set d1, and the output data of the model is a data set d2. In addition, the data set d1 is also used as the input data of the second part of the model (i.e., the R2 module), and the input data of the second part of the model also includes a data set d3. Therefore, the first information can indicate the input position corresponding to each data set. The specific indication manner can be position ID indication, and the definition of the position ID is related to the sub-model structure. Different sub-model structures correspond to different position indication manners. For example, when the sub-model is a multi-head attention, the position ID is X, indicating that the input position corresponding to the data set is the Xth head in the multi-head attention. When the sub-model is an encoder or a decoder, the position ID is X, indicating that the input position corresponding to the data set is the Xth encoder or decoder block in the encoder or the decoder. One data set can correspond to multiple position IDs.

[0296] In the second implementation, the first model includes at least two model units, and each sub-model in the at least two sub-models includes one or more model units in the at least two model units. The first sub-model includes M model units in the at least two model units, and M is a positive integer. The first information includes the identifiers of the M model units.

[0297] In the second implementation, the first model can include at least two model units, and each sub-model in the at least two sub-models included in the first model can include one or more model units in the at least two model units. Therefore, the sender (for example, the second communication apparatus) of the first information can indicate the first sub-model by the identifiers of the M model units included in the first sub-model, so that the first communication apparatus can quickly determine the first sub-model based on the identifiers, and the overhead of indicating the first sub-model can be reduced.

[0298] Optionally, the identifier of each model unit in the at least two model units included in the first model is pre-defined in relation to the model structure of the model unit. In this way, the first communication apparatus can quickly determine the model structure of the M model units included in the first sub-model based on the identifiers of the M model units, and the overhead of indicating the model structure can be reduced.

[0299] As an implementation example of the second implementation manner (denoted as example 1), the first information is further used to indicate the connection relationship of the M model units. In other words, in addition to indicating the M model units contained in the first sub-model through the identification of the M model units, the first information can be used to indicate the connection relationship of the M model units, so that the first communication apparatus can determine the first sub-model based on the connection relationship and the M model units.

[0300] Optionally, in example 1, the second communication apparatus can also indicate the connection relationship of the M model units in other manners, for example, the second communication apparatus indicates the connection relationship of the M model units through other information / messages / signaling different from the first information.

[0301] In example 1, the first information can indicate the sub-model through a general model unit, that is, the first information can include the identification (ID) of the model unit.

[0302] In this example 1, the mapping relationship between the model unit ID and the model unit structure can be predefined, which can be predefined through a table, a formula, etc. For example, taking the mapping relationship predefined through a table as an example, as shown in Table 3.

[0303] Table 3

[0304]

[0305] In Table 3, model unit 0 is, model unit 1 is the query matrix Q in the attention structure in the Transformer, model unit 2 is the key-value matrix K in the attention structure in the Transformer, model unit 3 is the value matrix V in the attention structure in the Transformer, model unit 4 is a multiplication operation, model unit 5 is a softmax activation function, model unit 6 is a mask matrix, and model unit 7 is a scaling operation.

[0306] Optionally, taking the first communication apparatus as a terminal device and the second communication apparatus as a network device as an example. The second communication apparatus can configure the model structure of one or more model units contained in the first model through an RRC parameter, that is, the second communication apparatus can pre-configure the model structure of various supported model units to the first communication apparatus, and when the model structure of the configured model units is more than one, the model structure of the current model unit is activated through DCI.

[0307] Thereafter, based on the mapping relationship between the predefined or preconfigured model unit ID and the model unit, the second communication device indicates the ID of the sub-model to the first communication device, and the first communication device can learn the structure corresponding to the model. It should be understood that the second communication device can indicate multiple model unit IDs to the same first communication device, and the number of the same model unit can also be multiple, that is, the first information can contain one or more numbers of the same model unit ID.

[0308] In addition, in example 1, the first information can also indicate the connection relationship between the model units (i.e., the “connection relationship of the M model units” described above). Generally, the connection relationship between the model units includes both serial and parallel relationships, which will be introduced respectively.

[0309] For example, the serial relationship can be embodied by the order of the ID. Assuming that the order of the model units indicated by the second communication device is 1->2->4->7->6->5->3->4, where the structure corresponding to the model unit ID is determined by Table 3, and each model unit ID is represented by 3 bits, a total of 8 model units. Then, according to the above indication order of the model units, the serial relationship shown in Table 4 can be obtained. Figure 4e

[0310] For another example, for the parallel relationship, it can be embodied by a specific parallel identifier (P_f). For example, when each model unit is indicated, a fixed 1 bit is added after the model unit ID to indicate the parallel identifier. When the 1 bit is 1, it indicates that the model unit is in parallel with the previous model unit, and when the 1 bit is 0, it indicates that the model unit is in series with the previous model unit.

[0311] As shown in Table 5, when the parallel identifier is inserted, the indication order of the model units is: Figure 4e

[0312] 1 (P_f=0)->2 (P_f=1)->4 (P_f=0)->7 (P_f=0)->6 (P_f=0)->5 (P_f=0)->3 (P_f=1)->4 (P_f=0).

[0313] As an implementation example of the second implementation manner (denoted as example 2), the connection relationship of the at least two model units is preconfigured. In other words, in the at least two model units included in the first model, the connection relationship of the at least two model units is preconfigured. In this way, the overhead of indicating the connection relationship of the at least two model units (for example, indicating the connection relationship of the M model units) can be saved.

[0314] ​​In Example 2, the second communication device can instruct a sub-model based on a specific model unit; that is, the first information sent by the second communication device may include the identifier (ID) of the model unit. In Example 2, a mapping relationship between the model unit ID and the first model can be predefined, and a model unit is a specific unit in the first model.

[0315] like Figure 4f The example shown illustrates the position of each model unit within the first model. In this case, the second communication device indicates a specific model unit ID via a first information reference. The first communication device can then obtain the sub-model based on a predefined mapping relationship. For example, if the second communication device indicates model units 0, 1, 2, 3, 4 to the UE, the first communication device can obtain... Figure 4f The sub-model corresponding to the dashed box in the middle.

[0316] Optionally, the type or structure of the model unit can be defined through a predefined table (such as Table 3 above).

[0317] In one possible implementation of implementation method two, the first information is also used to indicate at least one of the following:

[0318] Information AM refers to the model parameters of some or all of the model units in the model unit.

[0319] Information BM contains the configuration parameters of some or all of the model units.

[0320] Information CM is the input dataset of some or all of the model units in the model unit.

[0321] Information DM is the output dataset of some or all of the model units in the model unit.

[0322] Therefore, in addition to indicating the first sub-model, the first information can also be used to indicate at least one of the above, so that the first communication device can perform processing on the first sub-model based on the at least one.

[0323] Optionally, at least one of the above can be pre-configured.

[0324] Optionally, the second communication device may also indicate at least one of the above in other ways, such as by indicating at least one of the above through other information / messages / signalings different from the first information.

[0325] The following will describe information A through information D with more implementation examples.

[0326] As an example of implementation, the first information can indicate the model parameters of the model unit (i.e., the information A mentioned above).

[0327] In this implementation example, the first information can include frozen parameter information, which can indicate that part of the reference is frozen, and the frozen model unit remains unchanged in training. Wherein, whether the model parameters of one model unit are frozen can be embodied by a specific frozen identifier (F_f), for example, a fixed 1-bit is added after the model unit ID when indicating each model unit, which is used to indicate the frozen identifier, and the 1-bit is 1 when indicating that the model unit is frozen, and 0 when indicating that the model unit is not frozen.

[0328] For example, as shown in the model structure in Table 3, when the parallel identifier is inserted, the indication order of the model units is 1 (P_f=0; F_f=1) -> 2 (P_f=1; F_f=1) -> 4 (P_f=0) -> 7 (P_f=0) -> 6 (P_f=0) -> 5 (P_f=0) -> 3 (P_f=1) -> 4 (P_f=0; F_f=0). Only the model units including weights need to indicate the frozen identifier, such as the query matrix, the key-value matrix, and the value matrix in Table 3. The first communication device can determine whether the frozen identifier exists according to the type of the model unit. Figure 4e

[0329] Optionally, when a certain model unit is frozen, the first information can indicate the model parameters corresponding to the frozen reference.

[0330] In this implementation example, the first information can indicate whether the non-frozen unit includes the model parameters. Wherein, if the first information can indicate that the non-frozen unit includes the model parameters, the second communication device can further indicate the model parameters corresponding to the non-frozen unit, so that the first communication device performs model processing based on the model parameters indicated by the second communication device. Similarly, whether the non-frozen unit includes the model parameters can be embodied by a specific non-frozen parameter identifier (N_f), for example, a fixed 1-bit is added after the model unit ID when indicating each model unit, which is used to indicate whether the non-frozen unit includes the model parameters, and the 1-bit is 1 when indicating that the non-frozen unit includes the model parameters, and 0 when indicating that the non-frozen unit does not include the model parameters. Only the non-frozen model units including weights need to indicate the non-frozen parameter identifier.

[0331] Optionally, the first information can indicate that all the model units are non-frozen model units.

[0332] As an implementation example, the first information can indicate the configuration parameters of the model units (i.e., the information B in the foregoing). Wherein, the configuration parameters can include the dimensions or parameters of the model units, etc. For example, the configuration parameters can include the dimensions of the query matrix Q, the key-value matrix K, the value matrix V, and the mask matrix M in Table 3, and the dimensions of the convolution kernel of the CNN network. For another example, the configuration parameters can include the parameters corresponding to the scaling operation, etc. ​

[0333] As an implementation example, the first information can indicate the dataset of the model unit (i.e., information C and information D mentioned above). A model unit may not correspond to a dataset, or it may correspond to one or more datasets. Therefore, the second communication device can use the first information to indicate whether each model unit corresponds to a dataset, the number of corresponding datasets, and whether the corresponding dataset is an input dataset or an output dataset. (As mentioned above...) Figure 4d As shown, model unit R1 corresponds to input dataset 1; model unit R2 corresponds to input datasets d1 and d3, and output dataset d2.

[0334] exist Figure 3 In one possible implementation of the method shown, the first information in step S301 is also used to indicate the transmission of the processing result of the first sub-model. In other words, a second information can also be transmitted based on the indication, so that the recipient of the second information can obtain the model processing result (i.e., the second sub-model) of the first communication device based on the second information, so that the recipient can perform model management (e.g., the recipient performs model aggregation, model updating, etc. based on the second sub-model) based on the second sub-model.

[0335] Optionally, the second communication device may also instruct the transmission of the processing result of the first sub-model in other ways, such as by instructing the transmission of the processing result of the first sub-model through other information / messages / signalings different from the first information.

[0336] The following is given Figure 3 This is a specific example of the chained processing procedure. Taking the second communication device as the network device and the first communication device as the terminal device as an example, and assuming there are three first communication devices, the communication process between the second communication device and the three first communication devices can be understood as the communication process between the network device and three terminal devices, denoted as UE1, UE2, and UE3 respectively. As mentioned earlier... Figure 4a The example shown assumes that the network device indicates the sub-model using the method described in Example 1 above. The sub-model indicated by the network device to UE1 includes model units 1 and 2, the sub-model indicated to UE2 includes model units 2, 3, and 4, and the sub-model indicated to UE3 includes model units 3 and 4. The specific process includes the following steps:

[0337] 1. The network device first performs initial training to obtain the model units R1, R2, R3, R4 and training dataset d1, d2, d3, d4, d5 contained in the first model on the network device side.

[0338] 2. The network device indicates to UE1 the structure of model unit R2, the structure and parameters of model unit R1, that model unit R1 is frozen, and data sets d1 and d3. It should be understood that this procedure is an example implementation of the foregoing step S301.

[0339] Thereafter, UE1 trains the model based on d1 and d3, and obtains actual model unit R2’ after training, and UE1 indicates R2’ to the network device. It should be understood that this procedure is an example implementation of the foregoing step S302.

[0340] 3. The network device performs fine-tuning or compression processing on R2’, and obtains a new reference model R2. The network device indicates to UE2 the structure of model unit R3, the structure and parameters of model unit R2, that model unit R2 is frozen, and data sets d2 and d4. It should be understood that this procedure is an example implementation of the foregoing step S301.

[0341] Thereafter, UE1 trains the model based on d1 and d3, and obtains actual model unit R2’ after training, and UE1 indicates R2’ to the network device. It should be understood that this procedure is an example implementation of the foregoing step S302.

[0342] 4. The network device performs distillation processing on R3’, and obtains a new reference model R3. The network device indicates to UE3 the structure of model unit R4, the structure and parameters of model unit R3, that model unit R3 is frozen, and data sets d3 and d5. It should be understood that this procedure is an example implementation of the foregoing step S301.

[0343] Thereafter, UE1 trains the model based on d3 and d5, and obtains actual model unit R4’ after training, and UE2 indicates R4’ to the network device. It should be understood that this procedure is an example implementation of the foregoing step S302.

[0344] Through the above procedure, the network device can perform a chain processing procedure of the model through UE1, UE2 and UE3 in sequence, to realize cooperative processing of the model among various distributed nodes (i.e., various UEs).

[0345] In Figure 3In a possible implementation of the method, in step S302, the first communication device sending the second information comprises: the first communication device sending the second information at or after a first time point; wherein a difference between the first time point and a second time point is determined by the time offset information, and the second time point is associated with the first sub-model. The first time point can be a time point at which the first communication device supervises the processing of the first sub-model. The first time point can be before an end time point of the processing of the first sub-model by the first communication device. For example, the first communication device can supervise the training of the first sub-model, in other words, the first communication device can send the second information at or after the first time point, so that a receiver (for example, the second communication device) of the second information can receive the second information as soon as possible before the end time point of the training, thereby reducing the overall processing delay of the plurality of first communication devices on different sub-models.

[0346] For example, if the first communication device discovers that the training performance is lower than the target threshold in advance, the second communication device can learn this information as soon as possible, and the second communication device can select other communication devices for training to reduce the overall processing delay of the plurality of first communication devices on different sub-models.

[0347] In addition, the data of the processing of the sub-model by different communication devices can have timeliness, and in the above scheme, the data of each communication device can be avoided to a certain extent without invalidation under the condition of reducing the overall processing delay of the plurality of first communication devices on different sub-models.

[0348] It should be understood that the first time point can be a time point at which the first communication device obtains / determines / obtains the performance of the second sub-model, or the first time point can be a time point at which the first communication device obtains / determines / obtains the difference between the performance of the second sub-model and the threshold. For example, the first time point can be referred to as a supervision time point, a detection time point, etc.

[0349] In Figure 3 In another possible implementation of the method, in step S302, the first communication device sending the second information comprises: the first communication device sending the second information at or before a third time point, wherein a difference between the third time point and a second time point is determined by the time offset information, and the second time point is associated with the first sub-model. Specifically, the third time point can be a cutoff time point of the processing of the first sub-model by the first communication device, for example, the first communication device can supervise at or before the cutoff time point of the training of the first sub-model. In this way, the first communication device can send the second information at or after the first time point, so that a receiver (for example, the second communication device) of the second information can receive the second information as soon as possible before the end time point of the training, thereby reducing the overall processing delay of the plurality of first communication devices on different sub-models.

[0350] For example, if the first communication device finds that the training performance is lower than the target threshold in advance, the second communication device can learn the information in advance, and the second communication device can select other communication devices for training.

[0351] Optionally, the end time can be replaced by other descriptions, such as a termination time, or a last time unit (such as a symbol, a subframe, a frame, etc.), etc.

[0352] For example, in the chain processing of the above network device through UE1, UE2 and UE3 in turn, the next UE needs to start training after the previous UE completes training, so when the training time of one of the UEs is too long, it will affect the training of the next UE and the training delay of the whole model. In order to reduce the overall training delay, through the control of the above time offset information, the UE can send its own model processing result as soon as possible, and the network device can receive the model processing result sent by the UE as soon as possible, so as to reduce the overall processing delay of multiple UEs on different sub-models.

[0353] In a possible implementation, the method further includes: the first communication device receiving third information, the third information being used to indicate the time offset information. In other words, the first communication device can also determine the time offset information based on the received third information, so that the first communication device can send the second information based on the specified time offset information.

[0354] Optionally, the method further includes: the first communication device sending fourth information, the fourth information being used to indicate the second time, the second time being the starting time of training the first sub-model. Specifically, the first communication device can also send the fourth information indicating the second time, so that the receiver (for example, the second communication device) of the fourth information can determine the corresponding time offset information based on the starting time of training the first sub-model, to assist the decision of the receiver.

[0355] In another possible implementation, the time offset information is preconfigured. In other words, the time offset information used to determine the sending time of the second information can be preconfigured, which can save the configuration overhead of the time offset information.

[0356] In Figure 3 In a possible implementation of the method, the second time is associated with the first sub-model, including: the second time is the receiving time of the first information, or the second time is the starting time of training the first sub-model. Thus, the second time can be implemented in the above-mentioned various ways to improve the flexibility of the scheme implementation.

[0357] It should be noted that for a certain first communication device, the same sub-model can exist one or more rounds of processing processes, or exist multiple stages of processing, wherein the processing processes of different rounds or different stages can be obtained by the first communication device based on the same or different parameters (for example, a training parameter set, a number of fine-tuning times, etc.). Wherein, each round or each stage of processing process can correspond to a sending time. Hereinafter, taking an example that the first communication device performs one or more rounds of processing processes corresponding to one or more first times, it can be understood that the first time can also be replaced by the third time.

[0358] For example, taking an example that there are multiple rounds of processing processes, for the first round of processing processes, the first communication device sends the second information_1 for indicating the first round of processing results at or after the first time_1.

[0359] For example, for the second round of processing processes, the first communication device sends the second information_2 for indicating the second round of processing results at or after the first time_2.

[0360] By analogy, until the first communication device completes the one or more rounds of processing processes, and the sending of the processing results corresponding to the one or more rounds of processing processes.

[0361] In the above process, different first times can be determined by one or more time offset amounts contained in the time offset information, and the one or more time offset amounts can include offset amount_1, offset amount_2, … For example, the first time_1 corresponding to the first round can be determined by the offset amount_1 contained in the time offset information and the starting time corresponding to the first round; the first time_2 corresponding to the first round can be determined by the offset amount_2 contained in the time offset information and the starting time corresponding to the second round. Wherein, the starting times corresponding to different time offset amounts can be the same or different, which will be described in combination with more implementation examples.

[0362] Example A. The starting times corresponding to one or more time offset amounts contained in the time offset information can be the same.

[0363] As shown in the example of Figure 4g The starting times corresponding to different time offset amounts can all be the second time, which is the receiving time of the first information, or the starting time of training the first sub-model. In Figure 4gIn the example shown, the first moment_1 corresponding to the first round or the first stage can be determined by the offset amount_1 contained in the time offset information and the second moment, the first moment_2 corresponding to the second round or the second stage can be determined by the offset amount_2 contained in the time offset information and the second moment, and so on.

[0364] Example A. The start times corresponding to one or more time offset values ​​included in the time offset information can be different.

[0365] like Figure 4h As shown in the example, the start times corresponding to different time biases can be different, and the processing in different rounds can be continuous in time. Figure 4h In the example shown, the first time point _1 corresponding to the first round can be determined by the offset amount _1 contained in the time offset information and the second time point, the first time point _2 corresponding to the second round can be determined by the offset amount _2 contained in the time offset information and the termination time corresponding to the first round, or the first time point _2 corresponding to the second round can be determined by the offset amount _2 contained in the time offset information and the first time point _1 corresponding to the first round, and so on.

[0366] like Figure 4i As shown in the example, the start times corresponding to different time biases can be different, and the processing in different rounds can be discontinuous in time. Figure 4i In the example shown, the first time point _1 corresponding to the first round can be determined by the offset amount _1 included in the time offset information and the second time point. After the first time point _1, the first communication device may need a certain processing delay before it can start the processing of the second round, so there may be a certain time interval between the end time of the first round (as shown by the first time point _1 in the figure) and the start time of the second round (as shown by the time point x in the figure) (as shown by the time interval between the first time point _1 and the time point x in the figure). In this case, the first time point _2 corresponding to the second round can be determined by the offset amount _2 included in the time offset information and the time point x.

[0367] It should be understood that the "processing latency" for different first communication devices may be the same or different. This "processing latency" may include one or more of the following: data preparation latency, data loading latency, model switching latency, or other latency.

[0368] exist Figure 3In a possible implementation of the method, the first communication device sends the second information at the first time or after the first time, including: the first communication device sends the second information at the first time or after the first time, in a case where a first condition is met; and the first condition includes at least one of the following: the model performance of the second sub-model is higher than a threshold, or the fifth information is received, the fifth information being used to indicate termination of processing of the first sub-model. Specifically, the first communication device can also send the second information based on the first condition, so as to improve the model processing efficiency. For example, in a case where the first condition includes that the model performance of the second sub-model is higher than a threshold, the first communication device can send the second information when the performance of the second sub-model is higher, so that a receiver of the second information can obtain the sub-model with higher performance. For another example, in a case where the first condition includes that the first communication device receives the fifth information, the first communication device can send the second information based on a termination processing indication of a peer end.

[0369] Optionally, the fifth information can be triggered based on sixth information, that is, the second communication device can send the fifth information to the first communication device based on an abnormal situation indicated by the sixth information after receiving the sixth information. The implementation of the sixth information can refer to the description below.

[0370] Optionally, the second information can include a reason value, the reason value being used to indicate the first condition.

[0371] In Figure 3 In a possible implementation of the method, the first communication device sends the sixth information at the first time or after the first time, in a case where a second condition is met, the sixth information being used to indicate that the first sub-model processing is abnormal; and the second condition includes at least one of the following: the model performance of the second sub-model is lower than a threshold, or the processing capability of the first communication device does not match the first sub-model. Specifically, the first communication device can also send the sixth information based on the second condition, so that the scheme can adapt to a model processing abnormal scenario. For example, in a case where the second condition includes that the model performance of the second sub-model is lower than a threshold, the first communication device can send the sixth information when the performance of the second sub-model is lower, so that a receiver of the sixth information can learn about the abnormal situation of the model processing of the first communication device as soon as possible. For another example, in a case where the second condition includes that the processing capability of the first communication device does not match the first sub-model, a receiver of the sixth information can learn about the abnormal situation that the processing capability of the first communication device does not match the indicated first sub-model as soon as possible. The second communication device can learn about the information as soon as possible, and the second communication device can select other communication devices for training, thereby reducing the overall processing delay of multiple first communication devices on different sub-models.

[0372] Optionally, the sixth information can comprise a cause value, which is used to indicate the second condition.

[0373] Optionally, the process that the first communication device sends the second information is an optional step. For example, in the case that the first communication device sends the sixth information, it is possible that the first communication device does not obtain the second sub-model based on the first sub-model, and therefore the first communication device can not be able to send the second information indicating the second sub-model. Or, in the case that the first communication device sends the sixth information, it is possible that the first communication device does not obtain the second sub-model with better performance based on the first sub-model, and therefore the first communication device does not need to send the second information, which can avoid the second communication device processing based on the second sub-model with too low performance, and can reduce the overhead.

[0374] Optionally, in the case that the first communication device sends the sixth information, the first communication device can also send the second information. Since the second sub-model indicated by the second information can be obtained by the first communication device based on its own computing power, data set, etc., even if the performance of the second sub-model is poor, the second communication device can still process based on the second sub-model, and the local computing power and / or local data set of one or more first communication devices can be utilized as much as possible.

[0375] It should be noted that any of the model performance involved in the first condition and the model performance involved in the second condition can be implemented in various ways. For example, the model performance can comprise model accuracy, or the model performance can be reflected by the value of the model accuracy. The higher the value of the model accuracy, the better the model performance, and the lower the value of the model accuracy, the worse the model performance. For another example, the model performance can comprise the degree of drift of output data, or the model performance can be reflected by the degree of drift of output data.

[0376] Correspondingly, the first communication device can determine the model performance threshold value in a configured manner or a preconfigured manner, and determine whether the above-mentioned first condition (and / or the second condition) is met based on the relationship between the measured model performance and the model performance threshold value.

[0377] It can be understood that if the model performance measured by the first communication device is lower than the performance indicated by the model performance threshold value, the first communication device can report the processing abnormality of the model through the sixth information.

[0378] Optionally, in the case that the first communication device is a terminal device and the second communication device is a network device, the network device can also indicate to the terminal device the resource (i.e., the resource carrying the sixth information) for reporting the processing exception of the model. For example, the resource can include a physical uplink control channel (PUCCH) or a physical uplink shared channel (PUSCH) resource, etc.

[0379] Alternatively, the terminal device can request the network device for the resource (i.e., the resource carrying the sixth information) for reporting the processing exception of the model, and then send the sixth information on the resource indicated by the network device to realize the indication of the processing exception of the model.

[0380] It should be noted that in the above chain processing process of the model performed by the network device through UE1, UE2 and UE3 in turn, the next UE can start training only after the previous UE completes training, and therefore when the training of one UE takes too long, it will affect the training of the next UE and the training time delay of the entire model. In order to reduce the overall training time delay, other ways can also be used to achieve it, for example, the second communication device can indicate the same or different sub-models to at least two first communication devices, and the model processing time of the at least two first communication devices can be partially overlapped or fully overlapped, in this way, the time overhead caused by the model processing time of different first communication devices being staggered can be reduced.

[0381] For example, another example of the chain processing process shown in Figure 3 is given below. Taking the second communication device as a network device and the first communication device as a terminal device as an example, and taking the number of first communication devices as 3 as an example. That is, the communication process between the second communication device and the three first communication devices can be understood as the communication process between the network device and the three terminal devices, which are denoted as UE1, UE2, and UE3 respectively.

[0382] As shown in the example of Figure 4j , it is assumed that the network device indicates the sub-models in the manner of the foregoing example 1, the sub-model 1 indicated by the network device to UE1 includes model units 1 and 2, the sub-model 2 indicated by the network device to UE2 includes model units 2, 3, and 4, and the sub-model 3 indicated by the network device to UE3 includes model units 4 and 5, which includes the following processes:

[0383] 1. The network device first performs initial training to obtain the model units R1, R2, R3, R4, and R5 on the network device side, and training data sets d1, d2, d3, d4, d5, and d6.

[0384] 2. The network device indicates to UE1 the structure of model unit R2, the structure and parameters of model unit R1, that model unit R1 is frozen, and data sets d1 and d3.

[0385] Thereafter, UE1 trains the model based on d1 and d3, obtains actual model unit R2’ after training, and indicates R2’ to the network device.

[0386] 3. The network device indicates to UE3 the structure of model unit R4, the structure and parameters of model unit R5, that model unit R5 is frozen, and data sets d4 and d6.

[0387] Thereafter, UE3 trains the model based on d4 and d6, obtains actual model unit R4’ after training, and indicates R4’ to the network device.

[0388] 4. The network device fine-tunes or compresses R2’ and R4’, obtains new reference models R2 and R4. The network device indicates to UE2 the structure of model unit R3, the structure and parameters of model units R2 and R4, that model units R2 and R4 are frozen, and data sets d2 and d5.

[0389] Thereafter, UE2 trains the model based on d2 and d5, obtains actual model unit R3’ after training, and indicates R3’ to the network device.

[0390] In the above process, since training starts from both sides of the model at the same time, in order to further reduce the time delay of training, when one side trains faster, the training of the other side can be terminated, and the specific process is as follows:

[0391] After one side UE completes training, it feeds back the trained model to the network device. In addition, the network device determines whether to terminate the training of the other side according to the time of training report.

[0392] The other side UE reports a training exception indication according to the supervision time node and the corresponding target performance, and the determination method of the time of model processing is the same (such as the previous example A or example B or other related implementation processes).

[0393] In addition, the network device sends a training termination indication to the other side UE. The training termination indication can be sent before the other side sends a training exception, that is, the other side UE can receive the training termination indication before the supervision time node.

[0394] Please refer to Figure 5This application provides a communication device 500, which 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.

[0395] It should be noted that the transceiver unit 502 may include a transmitting unit and a receiving unit, which are used to perform transmitting and receiving respectively.

[0396] In one possible implementation, when the device 500 is for performing Figure 3 When the method executed by the first communication device in the relevant embodiments is performed, the device 500 includes a processing unit 501 and a transceiver unit 502; the transceiver unit 502 is used to receive first information, the first information being used to indicate a first sub-model, the first sub-model being one of the sub-models in a first model, the first model including at least two sub-models; the first communication device processes the first sub-model to obtain a second sub-model; the processing unit 501 is used to determine second information; the transceiver unit 502 is also used to send second information, the second information being used to indicate the second sub-model; wherein, the first sub-model is obtained by other communication devices through model processing, and / or, the second sub-model is used for model processing by other communication devices.

[0397] In one possible implementation, when the device 500 is for performing Figure 3 When the method executed by the third communication device in the relevant embodiments is used, the device 500 includes a processing unit 501 and a transceiver unit 502; the processing unit 501 is used to determine first information; the transceiver unit 502 is used to send the first information, the first information being used to indicate a first sub-model, the first sub-model being one of the sub-models in a first model, the first model including at least two sub-models; wherein, the first sub-model is used to obtain a second sub-model through processing by the first communication device; the transceiver unit 502 is also used to receive second information, the second information being used to indicate the second sub-model; wherein, the first sub-model is obtained by other communication devices through model processing, and / or, the second sub-model is used for model processing by other communication devices.

[0398] 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.

[0399] 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.

[0400] It should be noted that the information execution process of the unit of the above-mentioned communication device 500 can be specifically described in the method embodiment shown above in this application, and will not be repeated here.

[0401] Please see Figure 6 This 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.

[0402] in, Figure 5 The transceiver unit 502 shown can be a communication interface, which can be... Figure 6 The input / output interface 602 may include an input interface and an output interface. Alternatively, the communication interface may also be a transceiver circuit, which may include an input interface circuit and an output interface circuit.

[0403] In one possible implementation, when the device 600 is for performing Figure 3 When the method executed by the first communication device in the relevant embodiments is performed, the input / output interface 602 is used to receive first information, which is used to indicate a first sub-model. The first sub-model is one of the sub-models in the first model, which includes at least two sub-models. The logic circuit 601 is used to process the first sub-model to obtain a second sub-model. The processing unit determines the second information. The input / output interface 602 is also used to send the second information, which is used to indicate the second sub-model. The first sub-model is obtained by other communication devices through model processing, and / or the second sub-model is used for model processing by other communication devices.

[0404] In one possible implementation, when the device 600 is for performingFigure 3 In the method performed by the second communication device in the related embodiments and the method, the logic circuit 601 is configured to determine the first information; the input and output interface 602 is configured to send the first information, and the first information is used to indicate a first sub-model, the first sub-model is one of sub-models in a first model, and the first model includes at least two sub-models; the first sub-model is used to be processed by the first communication device to obtain a second sub-model; the input and output interface 602 is further configured to receive second information, and the second information is used to indicate the second sub-model; the first sub-model is obtained by model processing of the other communication device, and / or the second sub-model is used for model processing of the other communication device.

[0405] The logic circuit 601 and the input and output interface 602 can also perform other steps of the first communication device or the second communication device in any embodiment and achieve the corresponding beneficial effects, which are not described herein.

[0406] In a possible implementation manner, Figure 5 The processing unit 501 shown can be Figure 6 The logic circuit 601 in the processing unit 501.

[0407] Optionally, the logic circuit 601 can be a processing device, and the functions of the processing device can be partially or entirely implemented through software.

[0408] Optionally, the processing device can include a memory and a processor, where 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 one of the method embodiments.

[0409] Optionally, the processing device can only include the processor. The memory for storing the computer program is located outside the processing device, and the processor is connected with the memory through a circuit / wire to read and execute the computer program stored in the memory. The memory and the processor can be integrated together or can be physically independent of each other.

[0410] Optionally, the processing device can be one or more chips, or one or more integrated circuits. For example, the processing device can be one or more field-programmable gate arrays (FPGA), application specific integrated circuits (ASIC), system on chips (SoC), central processing units (CPU), network processors (NP), digital signal processors (DSP), microcontroller units (MCU), programmable logic devices (PLD) or other integrated chips, or any combination of the above chips or processors, etc.

[0411] Referring to Figure 7 The communication device 700 involved in the above embodiments provided for the embodiments of the present application, and the communication device 700 can be specifically the communication device as the terminal device in the above embodiments, Figure 7 The terminal device is implemented by the terminal device (or components in the terminal device) in the illustrated example.

[0412] A possible logical structure diagram of the communication device 700 is shown in the figure, and the communication device 700 can include but is not limited to at least one processor 701 and a communication port 702.

[0413] The communication device 700 can include but is not limited to at least one processor 701 and a communication port 702. Figure 5 The transceiver unit 502 shown in the figure can be a communication interface, which can be Figure 7 The communication port 702 in the communication device 700 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.

[0414] Further optionally, the device can further include at least one of a memory 703 and a bus 704, and in the embodiments of the present application, the at least one processor 701 is configured to control and process the actions of the communication device 700.

[0415] Further, 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 device, transistor logic, hardware components, or any combination thereof. It can implement or execute various example logical blocks, modules, and circuits described in connection with the disclosure. The processor can also be a combination of computing components, such as a combination of one or more microprocessors, a combination of a digital signal processor and a microprocessor, and the like. For the sake of brevity and conciseness, the specific working processes of the system, device, and unit described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be described here.

[0416] It should be noted that, Figure 7 The communication apparatus 700 shown can be specifically used to implement the steps implemented by the terminal device in the foregoing method embodiments, and achieve the corresponding technical effects of the terminal device, Figure 7 The specific implementation of the communication apparatus can be referred to the foregoing method embodiments, and will not be described here.

[0417] Please refer to Figure 8 The foregoing communication apparatus 800 related to the embodiments of the present application is shown in the structural schematic diagram, and the communication apparatus 800 can be specifically the communication apparatus as the network device in the foregoing embodiments, Figure 8 The example shown is implemented by the network device (or components in the network device), and the structure of the communication apparatus can be referred to Figure 8 The structure shown.

[0418] The communication apparatus 800 includes at least one processor 811 and at least one network interface 814. Further optionally, the communication apparatus further includes at least one memory 812, at least one transceiver 813, and one or more antennas 815. The processor 811, the memory 812, the transceiver 813, and the network interface 814 are connected, for example, through a bus, and in the embodiments of the present application, the connection can include various interfaces, transmission lines, or buses, etc., which are not limited in the embodiments. The antenna 815 is connected to the transceiver 813. The network interface 814 is used to enable the communication apparatus to communicate with other communication devices through a communication link. For example, the network interface 814 can include a network interface between the communication apparatus and the core network device, such as an S1 interface, and the network interface can include a network interface between the communication apparatus and other communication apparatuses (such as other network devices or core network devices), such as an X2 or Xn interface.

[0419] Wherein, Figure 5 The transceiver unit 502 shown can be a communication interface, which can be Figure 8The network interface 814 in the communication device can include an input interface and an output interface. Alternatively, the network interface 814 can be a transceiver circuit, which can include an input interface circuit and an output interface circuit.

[0420] The processor 811 is mainly used for processing communication protocols and communication data, controlling the whole communication device, executing software programs, processing data of the software programs, for example, for supporting the communication device to perform the actions described in the embodiments. The communication device can include a baseband processor and a central processor. The baseband processor is mainly used for processing communication protocols and communication data, and the central processor is mainly used for controlling the whole terminal device, executing software programs, and processing data of the software programs. Figure 8 The processor 811 in the communication device can integrate the functions of the baseband processor and the central processor. Those skilled in the art can understand that the baseband processor and the central processor can also be independent processors interconnected by a bus or the like. Those skilled in the art can understand that the terminal device can include multiple baseband processors to adapt to different network modes, and the terminal device can include multiple central processors to enhance its processing capability. The various components of the terminal device can be connected by various buses. The baseband processor can also be referred to as a baseband processing circuit or a baseband processing chip. The central processor can also be referred to 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 the memory in the form of a software program, and the processor executes the software program to realize the baseband processing function.

[0421] The memory is mainly used for storing software programs and data. The memory 812 can exist independently and be connected to the processor 811. Alternatively, the memory 812 can be integrated with the processor 811, for example, integrated in a chip. The memory 812 can store program codes for executing the technical solutions of the embodiments of the present application, and the processor 811 controls the execution. Various computer programs executed can also be regarded as a driver of the processor 811.

[0422] Figure 8 Only one memory and one processor are shown. In actual terminal devices, multiple processors and multiple memories can exist. The memory can also be referred to as a storage medium or a storage device, etc. The memory can be a storage element on the same chip as the processor, i.e., an on-chip storage element, or an independent storage element, which is not limited in the embodiments of the present application.

[0423] The transceiver 813 can be configured to support the receiving or transmitting of radio frequency signals between the communication device and a terminal. The transceiver 813 can be connected to the antenna 815. The transceiver 813 includes a transmitter Tx and a receiver Rx. Specifically, the one or more antennas 815 can receive radio frequency signals, and the receiver Rx of the transceiver 813 is configured to receive the radio frequency signals from the antenna and convert the radio frequency signals into digital baseband signals or digital intermediate frequency signals, and provide the digital baseband signals or digital intermediate frequency signals to the processor 811 for further processing, such as demodulation processing and decoding processing, by the processor 811. In addition, the transmitter Tx in the transceiver 813 is also configured to receive modulated digital baseband signals or digital intermediate frequency signals from the processor 811, and convert the modulated digital baseband signals or digital intermediate frequency signals into radio frequency signals, and transmit the radio frequency signals through the one or more antennas 815. Specifically, the receiver Rx can selectively perform one or more levels of down-mixing processing and analog-to-digital conversion processing on the radio frequency signals to obtain the digital baseband signals or digital intermediate frequency signals, and the order of the down-mixing processing and the analog-to-digital conversion processing can be adjustable. The transmitter Tx can selectively perform one or more levels of up-mixing processing and digital-to-analog conversion processing on the modulated digital baseband signals or digital intermediate frequency signals to obtain the radio frequency signals, and the order of the up-mixing processing and the digital-to-analog conversion processing can be adjustable. The digital baseband signals and the digital intermediate frequency signals can be collectively referred to as digital signals.

[0424] The transceiver 813 can also be referred to as a transceiving unit, a transceiver, a transceiving device, etc. Optionally, the devices in the transceiving unit for implementing the receiving function can be regarded as a receiving unit, and the devices in the transceiving unit for implementing the transmitting function can be regarded as a transmitting unit, i.e., the transceiving unit includes the receiving unit and the transmitting unit. The receiving unit can also be referred to as a receiver, an input port, a receiving circuit, etc., and the transmitting unit can be referred to as a transmitter, a transmitter, or a transmitting circuit, etc.

[0425] It should be noted that, Figure 8 The communication device 800 shown can be specifically configured to implement the steps implemented by the network device in the foregoing method embodiments, and achieve the corresponding technical effects of the network device, Figure 8 The specific implementation modes of the communication device 800 shown can be referred to the descriptions in the foregoing method embodiments, which will not be described here one by one.

[0426] Please refer to Figure 9 The structure diagram of the communication device involved in the foregoing embodiments provided by the embodiments of the present application is shown.

[0427] It can be understood that the communication apparatus 900 includes, for example, modules, units, elements, circuits, or interfaces, and the like, which are appropriately configured together to perform the technical solutions provided in the present application. The communication apparatus 900 can be a terminal device or a network device described above, or can be a component (for example, a chip) of the devices, to implement the methods described in the following method embodiments. The communication apparatus 900 includes one or more processors 901. The processor 901 can be a general processor or a special-purpose processor, and the like. For example, it can 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 apparatus (such as a RAN node, a terminal, or a chip, and the like), execute software programs, and process data of the software programs.

[0428] Optionally, in one design, the processor 901 can include a program 903 (which can also be referred to as code or instructions at times) that can be run on the processor 901, so that the communication apparatus 900 performs the methods described in the following embodiments. In yet another possible design, the communication apparatus 900 includes a circuit (not shown) that can be implemented by hardware, software, or a combination of hardware and software. Figure 9

[0429] Optionally, the communication apparatus 900 can include one or more memories 902 having a program 904 (which can also be referred to as code or instructions at times) stored thereon, which can be run on the processor 901, so that the communication apparatus 900 performs the methods described in the above method embodiments.

[0430] Optionally, the processor 901 and / or the memory 902 can include an AI module 907, 908 for implementing AI-related functions. The AI module can be implemented by software, hardware, or a combination of software and hardware. For example, the AI module can include a radio intelligence control (RIC) module. For example, the AI module can be a near-real-time RIC or a non-real-time RIC.

[0431] Optionally, the processor 901 and / or the memory 902 can also store data. The processor and the memory can be separately arranged or integrated together.

[0432] Optionally, the communication apparatus 900 can also include a transceiver 905 and / or an antenna 906. The processor 901 can also be referred to as a processing unit, which controls the communication apparatus (such as a RAN node or a terminal). The transceiver 905 can also be referred to as a transceiving unit, a transceiver, a transceiving circuit, or a transceiver, and the like, which is used to realize the transceiving function of the communication apparatus through the antenna 906.

[0433] ​wherein, Figure 5 The processing unit 501 shown can be the processor 901. Figure 5 The transceiver unit 502 shown can be a communication interface, which can be the transceiver 905 in the communication device 1000, 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. Figure 9 The transceiver unit 502 shown can be a communication interface, which can be the transceiver 905 in the communication device 1000, 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.

[0434] The embodiments of the present application further provide a computer readable storage medium for storing one or more computer-executable instructions, when the computer-executable instructions are executed by a processor, the processor executes the method described in the possible implementation manners of the first communication device or the second communication device.

[0435] The embodiments of the present application further provide a computer program product (or computer program), when the computer program product is executed by the processor, the processor executes the method of the possible implementation manners of the first communication device or the second communication device.

[0436] The embodiments of the present application further provide a chip system, which includes at least one processor for supporting the communication device to implement the functions involved in the possible implementation manners of the communication device. Optionally, the chip system further includes an interface circuit, which provides program instructions and / or data for the at least one processor. In a possible design, the chip system can further include a memory, which is used to store necessary program instructions and data of the communication device. The chip system can be composed of a chip, or can include a chip and other discrete devices, and the communication device can be the first communication device or the second communication device in the foregoing method embodiments.

[0437] The embodiments of the present application further provide a communication system, which includes the first communication device and / or the second communication device in any of the foregoing embodiments.

[0438] In several embodiments provided in the present application, it should be understood that the disclosed system, device and method can be implemented in other manners. For example, the described device embodiments are merely schematic. For example, the division of the units is only a logical function division. There can be another division manner for the actual implementation, for example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed mutual couplings or direct couplings or communication connections between different units, can be indirect couplings or communication connections through some interfaces, devices or units, and can be in electrical, mechanical or other forms.

[0439] The units described as separated components can or can not be physically separated, and the components displayed as units can or can not be physical units, i.e., can be located in one place, or can be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purposes of the embodiments of the present application.

[0440] In addition, each functional unit in the embodiments of the present application can be integrated in one processing unit, or each unit can exist physically as a separate unit, or two or more units can be integrated in one unit. The integrated unit can be implemented in the form of hardware or in the form of a software functional unit. When the integrated unit is implemented in the form of 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 solutions of the present application essentially or substantially, or all or part of the technical solutions, can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the embodiments of the present application. The foregoing storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), magnetic disk or optical disk, and various other media that can store program codes.

Claims

1. A communication method, characterized in that, include: Receive first information, the first information being used to indicate a first sub-model, the first sub-model being one of the sub-models in a first model, the first model including at least two sub-models; The first sub-model is processed to obtain the second sub-model; Send a second message, the second message being used to instruct the second sub-model; wherein the first sub-model is obtained by other communication devices through model processing, and / or the second sub-model is used for model processing by other communication devices.

2. The method according to claim 1, characterized in that, The sending of the second information includes: The second information is sent at or after the first time point; wherein the difference between the first time point and the second time point is determined by time offset information, and the second time point is associated with the first sub-model.

3. The method according to claim 2, characterized in that, The method further includes: Receive third information, which is used to indicate the time offset information.

4. The method according to claim 3, characterized in that, The method further includes: Send a fourth message, which indicates the second time point, which is the start time for training the first sub-model.

5. The method according to claim 2, characterized in that, The time offset information is pre-configured.

6. The method according to any one of claims 2 to 5, characterized in that, The second time step is associated with the first sub-model and includes: The second time is the time when the first information is received, or the second time is the start time when training the first sub-model.

7. The method according to any one of claims 2 to 6, characterized in that, Sending the second information at or after the first moment includes: At or after the first moment, if the first condition is met, the second information is sent; The first condition includes at least one of the following: The performance of the second sub-model is higher than the threshold, or, Receive a fifth message, which is used to indicate the termination of the processing of the first sub-model.

8. The method according to any one of claims 2 to 7, characterized in that, At or after the first moment, if the second condition is met, a sixth message is sent, the sixth message being used to instruct the first sub-model to handle an anomaly; The second condition includes at least one of the following: The performance of the second sub-model is below the threshold, or the processing capability of the first communication device is not compatible with that of the first sub-model.

9. A communication method, characterized in that, include: Send first information, the first information being used to indicate a first sub-model, the first sub-model being one of the sub-models in a first model, the first model including at least two sub-models; wherein, the first sub-model is used to obtain a second sub-model through processing by a first communication device; Receive second information, the second information being used to indicate the second sub-model; wherein the first sub-model is obtained by other communication devices through model processing, and / or the second sub-model is used for model processing by other communication devices.

10. The method according to claim 9, characterized in that, The receiving of the second information includes: The second information is received at or after the first time point; wherein the difference between the first time point and the second time point is determined by time offset information, and the second time point is associated with the first sub-model.

11. The method according to claim 10, characterized in that, The method further includes: Send a third message, which is used to indicate the time offset information.

12. The method according to claim 11, characterized in that, The method further includes: Receive fourth information, which indicates the second time point, which is the start time for training the first sub-model.

13. The method according to claim 10, characterized in that, The time offset information is pre-configured.

14. The method according to any one of claims 10 to 13, characterized in that, The second time step is associated with the first sub-model and includes: The second time is the time when the first information is received, or the second time is the start time when training the first sub-model.

15. The method according to any one of claims 10 to 14, characterized in that, Receiving the second information at or after the first moment includes: The second information is received at or after the first moment, provided that the first condition is met. The first condition includes at least one of the following: The performance of the second sub-model is higher than the threshold, or, Send a fifth message, which indicates the termination of processing of the first sub-model.

16. The method according to any one of claims 10 to 15, characterized in that, At or after the first moment, if the second condition is met, a sixth message is received, the sixth message being used to instruct the first sub-model to handle an anomaly; The second condition includes at least one of the following: The performance of the second sub-model is below the threshold, or the processing capability of the first communication device is not compatible with that of the first sub-model.

17. A communication device, characterized in that, Includes a module for performing the method as described in any one of claims 1 to 16.

18. A communication device, characterized in that, It includes at least one processor, said at least one processor being used to perform the method as described in any one of claims 1 to 16.

19. The communication device according to claim 18, characterized in that, The communication device is a chip or chip system.

20. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program or instructions that, when executed by a communication device, implement the method as described in any one of claims 1 to 16.

21. A computer program product, characterized in that, It includes a computer program or instructions that, when executed by a computer, implement the method as described in any one of claims 1 to 16.

Images