Communication method and related equipment

Abstract

The invention provides a communication method and related equipment, which are used for enabling the computing power of a communication node to be applied to the updating of a learning model, improving the model generalization performance of a first sub-model in the learning model, and meeting the personalized requirements of the first node on the first sub-model in the learning model. In the method, a first node receives first information, and the first information is used for determining fusion parameters obtained by processing a first sub-model through first local data of N second nodes; wherein the first sub-model is included in a learning model, and the learning model further comprises a second sub-model; the first node sends second information, the second information is used for indicating an updated parameter of a first sub-model, and the updated first sub-model is obtained by processing the first sub-model based on the fusion parameter and second local data of the first node; wherein the updated parameters of the first sub-model are used for determining an updated learning model.

Description

A communication method and related equipment Technical Field

[0001] The present application relates to the field of communications, and in particular to a communication method and related equipment. Background Art

[0002] Wireless communication can be the transmission communication between two or more communication nodes without propagating through conductors or cables. The communication nodes generally include network devices and terminal devices.

[0003] Currently, in wireless communication systems, communication nodes generally possess both signal transceiver and computing capabilities. For example, network devices with computing capabilities primarily provide computing power to support signal transceiver capabilities (for example, calculating the time and frequency domain resources required to carry signals), enabling communication between the network device and other communication nodes.

[0004] However, in communication networks, communication nodes may have excess computing power beyond just supporting the aforementioned communication tasks. Therefore, how to utilize this computing power is a pressing technical issue.

[0005] Summary of the Invention

[0006] The present application provides a communication method and related equipment for enabling the computing power of a communication node to be applied to updating a learning model, and while improving the model generalization performance of a first sub-model in the learning model, it can also meet the first node's personalized needs for the first sub-model in the learning model.

[0007] In a first aspect, the present application provides a communication method, which is executed by a first node (the first node may be a communication device), or the method is executed by some components in the first node (such as a processor, a chip or a chip system, etc.), or the method can also be implemented by a logic module or software that can realize all or part of the functions of the first node. In the first aspect and its possible implementation, the method is described as being executed by the first node. In the method, the first node receives first information, and the first information is used to determine the fusion parameters obtained by processing the first local data of N second nodes on the first sub-model respectively, where N is a positive integer; wherein the first sub-model is included in a learning model, and the learning model also includes a second sub-model; the first node sends second information, and the second information is used to indicate the parameters of the updated first sub-model, and the updated first sub-model is obtained by processing the first sub-model based on the fusion parameters and the second local data of the first node; wherein the parameters of the updated first sub-model are used to determine the updated learning model.

[0008] Based on the above technical solution, after receiving the first information, the first node can determine the fusion parameters obtained by processing the first sub-model respectively with the first local data of the N second nodes based on the first information; thereafter, the first node can process the first sub-model based on the fusion parameters and the second local data of the first node to obtain the updated first sub-model. In other words, the first node can implement the model update of the first sub-model in the learning model based on the fusion parameters obtained by processing the first sub-model respectively with the first local data of other nodes (i.e., the N second nodes) and the second local data of the first node. Thus, using the first local data of other nodes and the second local data of the first node as the basis for updating the first sub-model can improve the model generalization performance of the first sub-model in the learning model while also meeting the first node's personalized needs for the first sub-model in the learning model.

[0009] Furthermore, in the above technical solution, the first node may further transmit second information indicating updated parameters of the first sub-model. The updated parameters of the first sub-model are used to determine an updated learning model. Thus, during the learning model update process, the updated learning model determined based on the updated first sub-model can also improve model generalization performance and meet the personalized needs of the node.

[0010] It should be understood that the learning model can be understood as a model used for AI learning and reasoning through AI methods, and the learning model can be replaced by AI model, AI learning model, machine learning model, neural network, AI network, etc.

[0011] It should be understood that the nodes participating in the learning model update may include a first node and N second nodes. The nodes participating in the learning model update may be communication nodes, that is, the first node and the N second nodes may all be communication nodes, and the communication nodes include network devices and / or terminal devices.

[0012] Optionally, the nodes participating in the learning model update may also include other nodes in addition to the first node and the N second nodes, which is not limited here.

[0013] Optionally, different nodes participating in the learning model update can communicate with each other in a variety of ways. For example, different nodes can be nodes in a decentralized learning model, and the system composed of different nodes can be understood as a fully distributed system without a central node; in this case, different nodes can communicate with neighboring nodes to participate in the learning model update. For another example, different nodes can be nodes in a centralized learning model, and the system composed of different nodes can be understood as a distributed system with a central node (or control node); in this case, different nodes can communicate with the central node to participate in the learning model update.

[0014] It should be noted that the learning model may include at least two sub-models, such as a first sub-model and a second sub-model. Optionally, the learning model may also include other models in addition to the first sub-model and the second sub-model, which are not limited here. Among them, the learning model can be used to perform an AI task, and the first sub-model and the second sub-model can be used to perform sub-tasks of the AI ​​task, that is, the AI ​​task performed by the first sub-model and the AI ​​task performed by the second sub-model can be understood as part of the AI ​​task performed by the learning model.

[0015] For example, the AI ​​task performed by the learning model could be an autonomous driving task. Generally, image recognition is a necessary component of autonomous driving tasks. For example, image recognition tasks such as human image recognition, license plate image recognition, and obstacle image recognition can be subtasks of the autonomous driving task. Accordingly, the AI ​​task performed by the first or second submodel can be any of these image recognition tasks.

[0016] For another example, the AI ​​task performed by the learning model can be an image classification task (or N-classification task). Generally, image classification can include image classification between different animals (for example, classifying images of cats and dogs), image classification between different plants (for example, classifying images of herbaceous plants and woody plants), etc. In other words, tasks such as image classification between different animals and image classification between different plants can be subtasks of the image classification task. Accordingly, the AI ​​task performed by the first sub-model or the second sub-model can be any of the image classification tasks between different animals and the image classification tasks between different plants.

[0017] For example, the AI ​​task performed by the learning model can be a machine translation task. Generally, word segmentation tasks, grammatical analysis tasks, sentence rewriting tasks, etc. can be subtasks of the machine translation task. Accordingly, the AI ​​task performed by the first sub-model or the second sub-model can be any of the word segmentation tasks, grammatical analysis tasks, and sentence rewriting tasks.

[0018] It should be understood that the updated model (including the updated first sub-model, the updated second sub-model, the updated learning model, etc.) can be understood as updating / iterating / improving the parameters in the model.

[0019] Optionally, the parameters of the updated model of the first sub-model indicated by the second information may include model parameters and / or gradient parameters.

[0020] Optionally, the basis for determining the updated learning model may include only the parameters of the updated first sub-model. Alternatively, the basis for determining the updated learning model may include the parameters of the updated first sub-model, and may also include the parameters of other updated sub-models (e.g., the second sub-model). Taking the AI ​​task performed by the above-mentioned learning model as an example, where the learning model only needs to optimize a single sub-task (e.g., the single sub-task is a license plate image recognition task), the process of updating the learning model can be understood as updating the first sub-model in the learning model for performing the license plate image recognition task. Accordingly, the basis for determining the updated learning model may include only the parameters of the updated first sub-model. Where the learning model needs to optimize M (M is an integer greater than or equal to 2) sub-tasks, the process of updating the learning model may be understood as updating the M sub-models in the learning model for performing M sub-tasks (the M sub-models include the first sub-model). Accordingly, the basis for determining the updated learning model may include the parameters of the updated first sub-model, and the parameters of other sub-models (e.g., the second sub-model).

[0021] Optionally, the local information (including local data and / or local capabilities, etc.) of the nodes participating in the learning model update may be different, and accordingly, the sub-models that different nodes focus on may be different. In the present application, the sub-model that the first node focuses on includes the first sub-model, which can be referred to as the focused sub-model of the first node, the target sub-model, etc.; and the other sub-models in the learning model (such as the second sub-model) can be referred to as the non-focused sub-model of the first node, other sub-models, etc. In other words, the sub-model that the first node focuses on includes the first sub-model and the non-focused sub-model of the first node includes the second sub-model; accordingly, the sub-model that any of the N second nodes focuses on includes the second sub-model and the non-focused sub-model of any of the N second nodes includes the first sub-model. For example, taking the AI ​​task performed by the above-mentioned learning model as an example of an autonomous driving task, the local data of the first node can include data collected by a parking lot camera. For this reason, the sub-model that the first node focuses on can be a sub-model that performs a license plate image recognition task; and the local data of the second node can include data collected by a camera near the sidewalk. For this reason, the sub-model that the second node focuses on can be a sub-model that performs a human image recognition task.

[0022] In a possible implementation manner of the first aspect, the first information includes the fusion parameter.

[0023] Based on the above technical solution, the first information received by the first node may include the fusion parameter, so that the first node can directly obtain the fusion parameter based on the received first information, and subsequently the first node can determine the updated first sub-model based on the fusion parameter.

[0024] In a possible implementation manner of the first aspect, the method further includes: the first node receiving third information, where the third information is used to indicate parameters of the updated learning model.

[0025] Based on the above technical solution, the first node can also receive third information and determine the parameters of the updated learning model based on the third information, so that the first node can determine the updated learning model based on the parameters.

[0026] Optionally, the first information including the fusion parameters and the third information may come from the same node. For example, the same node may be a central node between the first node and the N second nodes. In another example, the same node may be a neighboring node of the first node, and the neighboring node is capable of fusing parameters obtained by processing the first sub-model using the first local data of the N second nodes to obtain the fusion parameters.

[0027] In a possible implementation of the first aspect, the first information includes N sub-information, and the N sub-information respectively comes from the N second nodes; wherein, the N sub-information are respectively used to indicate the parameters obtained by processing the first sub-model using the first local data of the N second nodes.

[0028] Based on the above technical solution, the first information received by the first node may include N sub-information respectively from the N second nodes, wherein the parameters indicated by the N sub-information can be used to determine the fusion parameter, so that the first node can obtain the fusion parameter based on the received N sub-information, and subsequently the first node can determine the updated first sub-model based on the fusion parameter.

[0029] Optionally, the first node may receive the N sub-information by communicating with the N second nodes respectively through communication links with the N second nodes. Alternatively, the first node may receive the N sub-information by communicating with K nodes (for example, the K nodes may be one or more of the N second nodes, or the K nodes may be relay nodes between the first node and the N second nodes, etc.) through communication links with the K nodes.

[0030] In a possible implementation of the first aspect, the method also includes: the first node receives K fourth information, where the fourth information is used to indicate the parameters of the updated second sub-model, and K is an integer greater than or equal to 1; the parameters of the updated second sub-model are used to determine the updated learning model.

[0031] Optionally, K and N have the same value, or K is smaller than N.

[0032] Based on the above technical solution, the first node can also receive K fourth information, and determine the updated learning model based on the parameters of the updated second sub-model indicated by the fourth information and the parameters of the updated first sub-model, so that the first node can obtain the updated learning model through local fusion processing.

[0033] In a possible implementation manner of the first aspect, K is an integer greater than 1, and the method further includes: the first node sending fifth information, where the fifth information is used to indicate a fusion result obtained based on the K fourth information.

[0034] Based on the above technical solution, when the number of fourth information received by the first node is greater than 1, the first node can perform fusion processing on the parameters of the K updated second sub-models respectively indicated by the K information, and indicate the fusion result of the fusion processing through the fifth information, so that the recipient of the fifth information can subsequently determine the updated learning model based on the fusion result of the K updated second sub-models.

[0035] In a possible implementation of the first aspect, the method further includes: the first node sending sixth information, where the sixth information is used to indicate the loss and / or gradient obtained by processing the second sub-model by the first local data of the first node.

[0036] Based on the above technical solution, the first node can also send sixth information for indicating the loss and / or gradient obtained by processing the second sub-model with the first local data of the first node, so that the recipient of the sixth information can obtain the loss and / or gradient, and subsequently update the second sub-model based on the loss and / or gradient.

[0037] In a possible implementation of the first aspect, the method further includes: the first node sending or receiving first downlink control information (DCI) and second DCI, the first DCI being used to indicate resources carrying the first information, and the second DCI being used to indicate resources carrying the second information; wherein both the first DCI and the second DCI include a learning phase indication field, and a value of the learning phase indication field of the first DCI is different from a value of the learning phase indication field of the second DCI.

[0038] Based on the above technical solution, the resources carrying the first information and the resources carrying the second information can be scheduled through different DCIs, wherein the first node can serve as the scheduler of the resources or the first node can serve as the scheduled party of the resources, so that the first node can realize resource scheduling by sending or receiving the first DCI and the second DCI.

[0039] In addition, the first DCI and the second DCI may include a field for indicating a learning phase, and different resources may be indicated by different values ​​of the field.

[0040] In a possible implementation manner of the first aspect, the first local data is query set data, and the second local data is support set data.

[0041] Based on the above technical solution, the learning model can be used in a meta-learning scenario. Generally, nodes in a meta-learning scenario can include support data used to train the model to an intermediate version, and can also include query data used to calculate loss based on the intermediate version. In other words, the first local data used by nodes participating in the learning model update to process the model to obtain loss and / or gradient can be the query data, and the second local data used to process the model can be the support data.

[0042] In a possible implementation manner of the first aspect, the fusion parameter includes a fused gradient and / or a fused loss.

[0043] Based on the above technical solution, the fusion parameters obtained by respectively processing the first sub-model with the first local data of the N second nodes can be obtained by fusing the gradients and / or losses obtained by respectively processing the first sub-model with the first local data of the N second nodes, so as to improve the model generalization ability of the updated first sub-model and the updated learning model subsequently obtained based on the fusion parameters.

[0044] It should be understood that the first information can be used to determine the fusion parameters obtained by processing the first sub-model by the first local data of N second nodes respectively. Generally, the fused gradient contained in the fusion parameters can refer to processing the parameters of a sub-model to obtain the gradient, and the fused loss contained in the fusion parameters can refer to processing the parameters of a learning model to determine the loss (or loss function).

[0045] Optionally, when the fusion parameter includes the fused gradient, for the first node, the first node may locally process the first sub-model based on the second local data of the first node to obtain an intermediate version of the first sub-model, and the first node may process the intermediate version of the first sub-model based on the fused gradient to obtain an updated first sub-model. Where the fusion parameter includes the fused loss, for the first node, the first node may locally process the learning model based on the fused loss to obtain an intermediate version of the learning model, and then the first node may process the first sub-model in the intermediate version of the learning model based on the second local data of the first node to obtain an updated first sub-model.

[0046] The second aspect of the present application provides a communication method, which is executed by a second node (the second node may be a communication device), or the method is executed by some components in the second node (such as a processor, a chip or a chip system, etc.), or the method can also be implemented by a logic module or software that can realize all or part of the functions of the second node. In the second aspect and its possible implementation, the method is described as being executed by the second node. In this method, the second node sends a first information, and the first information is used to determine the fusion parameters obtained by processing the first local data of N second nodes on the first sub-model respectively, where N is a positive integer; wherein the first sub-model is included in a learning model, and the learning model also includes a second sub-model; the second node receives a second information, and the second information is used to indicate the parameters of the updated first sub-model, and the updated first sub-model is obtained by updating the first sub-model based on the fusion parameters and the second local data of the first node; wherein the parameters of the updated first sub-model are used to determine the updated learning model.

[0047] Based on the above technical solution, after the second node sends the first information, the first node can determine the fusion parameters obtained by processing the first sub-model respectively with the first local data of the N second nodes based on the received first information; thereafter, the first node can process the first sub-model based on the fusion parameters and the second local data of the first node to obtain the updated first sub-model. In other words, the first node can implement the model update of the first sub-model in the learning model based on the fusion parameters obtained by processing the first sub-model respectively with the first local data of other nodes (i.e., N second nodes) and the second local data of the first node. Thus, using the first local data of other nodes and the second local data of the first node as the basis for updating the first sub-model can improve the model generalization performance of the first sub-model in the learning model while also meeting the first node's personalized needs for the first sub-model in the learning model.

[0048] Furthermore, in the above technical solution, the second node may also receive second information indicating updated parameters of the first sub-model. The updated parameters of the first sub-model are used to determine an updated learning model. Thus, during the learning model update process, the updated learning model determined based on the updated first sub-model can also improve model generalization performance and meet the personalized needs of the node.

[0049] It should be understood that the learning model can be understood as a model used for AI learning and reasoning through AI methods, and the learning model can be replaced by AI model, AI learning model, machine learning model, neural network, AI network, etc.

[0050] It should be understood that the nodes participating in the learning model update may include a first node and N second nodes. The nodes participating in the learning model update may be communication nodes, that is, the first node and the N second nodes may all be communication nodes, and the communication nodes include network devices and / or terminal devices.

[0051] Optionally, the nodes participating in the learning model update may also include other nodes in addition to the first node and the N second nodes, which is not limited here.

[0052] Optionally, different nodes participating in the learning model update can communicate with each other in a variety of ways. For example, different nodes can be nodes in a decentralized learning model, and the system composed of different nodes can be understood as a fully distributed system without a central node; in this case, different nodes can communicate with neighboring nodes to participate in the learning model update. For another example, different nodes can be nodes in a centralized learning model, and the system composed of different nodes can be understood as a distributed system with a central node (or control node); in this case, different nodes can communicate with the central node to participate in the learning model update.

[0053] It should be noted that the learning model may include at least two sub-models, such as a first sub-model and a second sub-model. Optionally, the learning model may also include other models in addition to the first sub-model and the second sub-model, which are not limited here. Among them, the learning model can be used to perform an AI task, and the first sub-model and the second sub-model can be used to perform sub-tasks of the AI ​​task, that is, the AI ​​task performed by the first sub-model and the AI ​​task performed by the second sub-model can be understood as part of the AI ​​task performed by the learning model.

[0054] For example, the AI ​​task performed by the learning model could be an autonomous driving task. Generally, image recognition is a necessary component of autonomous driving tasks. For example, image recognition tasks such as human image recognition, license plate image recognition, and obstacle image recognition can be subtasks of the autonomous driving task. Accordingly, the AI ​​task performed by the first or second submodel can be any of these image recognition tasks.

[0055] For example, the AI ​​task performed by the learning model can be a machine translation task. Generally, word segmentation tasks, grammatical analysis tasks, sentence rewriting tasks, etc. can be subtasks of the machine translation task. Accordingly, the AI ​​task performed by the first sub-model or the second sub-model can be any of the word segmentation tasks, grammatical analysis tasks, and sentence rewriting tasks.

[0056] It should be understood that the updated model (including the updated first sub-model, the updated second sub-model, the updated learning model, etc.) can be understood as updating / iterating / improving the parameters in the model.

[0057] Optionally, the basis for determining the updated learning model may include only the parameters of the updated first sub-model. Alternatively, the basis for determining the updated learning model may include the parameters of the updated first sub-model, and may also include the parameters of other updated sub-models (e.g., the second sub-model). Taking the AI ​​task performed by the above-mentioned learning model as an example, where the learning model only needs to optimize a single sub-task (e.g., the single sub-task is a license plate image recognition task), the process of updating the learning model can be understood as updating the first sub-model in the learning model for performing the license plate image recognition task. Accordingly, the basis for determining the updated learning model may include only the parameters of the updated first sub-model. Where the learning model needs to optimize M (M is an integer greater than or equal to 2) sub-tasks, the process of updating the learning model may be understood as updating the M sub-models in the learning model for performing M sub-tasks (the M sub-models include the first sub-model). Accordingly, the basis for determining the updated learning model may include the parameters of the updated first sub-model, and the parameters of other sub-models (e.g., the second sub-model).

[0058] Optionally, the local information (including local data and / or local capabilities, etc.) of the nodes participating in the learning model update may be different, and accordingly, the sub-models that different nodes focus on may be different. In this application, the sub-model that the first node focuses on includes the first sub-model, which can be called the focus sub-model of the first node, the target sub-model, etc.; and the other sub-models in the learning model (such as the second sub-model) can be called the non-focus sub-model of the first node, other sub-models, etc. In other words, the sub-model that the first node focuses on includes the first sub-model and the non-focus sub-model of the first node includes the second sub-model; accordingly, the sub-model that any of the N second nodes focuses on includes the second sub-model and the non-focus sub-model of any of the N second nodes includes the first sub-model. Taking the AI ​​task performed by the above-mentioned learning model as an example, which can be an autonomous driving task, the local data of the first node can include data collected by a parking lot camera. For this reason, the sub-model that the first node focuses on can be a sub-model that performs a license plate image recognition task; and the local data of the second node can include data collected by a camera near the sidewalk. For this reason, the sub-model that the second node focuses on can be a sub-model that performs a human image recognition task.

[0059] In a possible implementation manner of the second aspect, the first information includes the fusion parameter.

[0060] Based on the above technical solution, the first information sent by the second node may include the fusion parameter, so that the first node can directly obtain the fusion parameter based on the received first information, and subsequently the first node can determine the updated first sub-model based on the fusion parameter.

[0061] In a possible implementation manner of the second aspect, the method further includes: the second node sending third information, where the third information is used to indicate parameters of the updated learning model.

[0062] Based on the above technical solution, the second node can also send third information, so that after receiving the third information, the first node can determine the parameters of the updated learning model based on the third information and determine the updated learning model based on the parameters.

[0063] Optionally, for example, the second node may be a central node of the first node and the N second nodes. In another example, the second node may be a neighboring node of the first node, and the second node has the ability to fuse parameters obtained by processing the first sub-model using the first local data of the N second nodes to obtain fused parameters.

[0064] In a possible implementation of the second aspect, the first information includes one or more sub-information among N sub-information; wherein, the N sub-information are respectively used to indicate parameters obtained by processing the first local data of the N second nodes on the first sub-model.

[0065] Based on the above technical solution, the first information sent by the second node may be one or more sub-information among N sub-information, wherein the parameters indicated by the N sub-information may be used to determine the fusion parameter, so that after the first node receives the first information, the first node may obtain the fusion parameter based on the received N sub-information, and subsequently the first node may determine the updated first sub-model based on the fusion parameter.

[0066] In a possible implementation of the second aspect, the method further includes: the second node sending fourth information, where the fourth information is used to indicate the parameters of the updated second sub-model; and the parameters of the updated second sub-model are used to determine the updated learning model.

[0067] Based on the above technical solution, the second node can also send fourth information, so that after receiving the fourth information, the first node can determine the updated learning model based on the parameters of the updated second sub-model indicated by the fourth information and the parameters of the updated first sub-model. Subsequently, the first node can obtain the updated learning model through local fusion processing.

[0068] In a possible implementation manner of the second aspect, the method further includes: the second node receiving fifth information, where the fifth information is used to indicate a fusion result obtained based on the K fourth information, where K is an integer greater than 1.

[0069] Based on the above technical solution, when the number of fourth information received by the first node is greater than 1, the first node can perform fusion processing on the parameters of the K updated second sub-models respectively indicated by the K information, and indicate the fusion result of the fusion processing through the fifth information, so that after the second node receives the fifth information, the second node can subsequently determine the updated learning model based on the fusion result of the K updated second sub-models.

[0070] In a possible implementation of the second aspect, the method further includes: the second node receiving sixth information, where the sixth information is used to indicate the loss and / or gradient obtained by processing the second sub-model by the first local data of the first node.

[0071] Based on the above technical solution, the second node can also receive sixth information indicating the loss and / or gradient obtained by processing the second sub-model by the first local data of the first node, so that the second node can obtain the loss and / or gradient, and subsequently update the second sub-model based on the loss and / or gradient.

[0072] In a possible implementation of the second aspect, the method also includes: the second node sends or receives first downlink control information DCI and second DCI, the first DCI is used to indicate the resources carrying the first information, and the second DCI is used to indicate the resources carrying the second information; wherein, the first DCI and the second DCI both include a learning phase indication field, and the value of the learning phase indication field of the first DCI is different from the value of the learning phase indication field of the second DCI.

[0073] Based on the above technical solution, the resources carrying the first information and the resources carrying the second information can be scheduled through different DCIs, wherein the second node can serve as the scheduler of the resources or the second node can serve as the scheduled party of the resources, so that the first node can realize resource scheduling by sending or receiving the first DCI and the second DCI.

[0074] In addition, the first DCI and the second DCI may include a field for indicating a learning phase, and different resources may be indicated by different values ​​of the field.

[0075] In a possible implementation manner of the second aspect, the first local data is query set data, and the second local data is support set data.

[0076] Based on the above technical solution, the above learning model can be a learning model in a meta-learning scenario. Generally, the node in the meta-learning scenario can include support set data for training the model to an intermediate version, and can also include query set data for calculating the loss based on the intermediate version.

[0077] In other words, the first local data used to process the model to obtain loss and / or gradient in the node participating in the learning model update may be query set data, and the second local data used to process the model may be support set data.

[0078] In a possible implementation manner of the second aspect, the fusion parameter includes a fused gradient and / or a fused loss.

[0079] Based on the above technical solution, the fusion parameters obtained by processing the first sub-model respectively with the first local data of the second node can be obtained by fusing the gradients and / or losses obtained by processing the first sub-model respectively with the first local data of the second node, so as to improve the model generalization ability of the updated first sub-model and the updated learning model subsequently obtained based on the fusion parameters.

[0080] In a third aspect of the present application, a communication device is provided, which is a first node, or a component of the first node (such as a processor, chip, or chip system), or a logic module or software capable of implementing all or part of the functions of the first node. In the third aspect and its possible implementations, the communication device is described as a first node, and the first node can be a terminal device or a network device.

[0081] The device includes a processing unit and a transceiver unit; the transceiver unit is used to receive first information, and the first information is used to determine the fusion parameters obtained by processing the first local data of N second nodes on the first sub-model respectively, where N is a positive integer; wherein the first sub-model is included in the learning model, and the learning model also includes the second sub-model; the processing unit is used to determine second information, and the second information is used to indicate the parameters of the updated first sub-model, and the updated first sub-model is obtained by processing the first sub-model based on the fusion parameters and the second local data of the first node; wherein the parameters of the updated first sub-model are used to determine the updated learning model; the transceiver unit is also used to send the second information.

[0082] In a possible implementation manner of the third aspect, the first information includes the fusion parameter.

[0083] In a possible implementation manner of the third aspect, the transceiver unit is further used to receive third information, where the third information is used to indicate parameters of the updated learning model.

[0084] In a possible implementation of the third aspect, the first information includes N sub-information, and the N sub-information respectively comes from the N second nodes; wherein, the N sub-information are respectively used to indicate the parameters obtained by processing the first local data of the N second nodes on the first sub-model.

[0085] In a possible implementation of the third aspect, the transceiver unit is also used to receive K fourth information, where the fourth information is used to indicate the parameters of the updated second sub-model, where K is an integer greater than or equal to 1; the parameters of the updated second sub-model are used to determine the updated learning model.

[0086] In a possible implementation manner of the third aspect, K is an integer greater than 1, and the transceiver unit is further used to send fifth information, where the fifth information is used to indicate a fusion result obtained based on the K fourth information.

[0087] In a possible implementation manner of the third aspect, the transceiver unit is further used to send sixth information, where the sixth information is used to indicate the loss and / or gradient obtained by processing the second sub-model by the first local data of the first node.

[0088] In a possible implementation manner of the third aspect, the transceiver unit is further configured to send or receive first downlink control information DCI and second DCI, where the first DCI is used to indicate resources carrying the first information, and the second DCI is used to indicate resources carrying the second information;

[0089] The first DCI and the second DCI both include a learning phase indication field, and a value of the learning phase indication field of the first DCI is different from a value of the learning phase indication field of the second DCI.

[0090] In a possible implementation manner of the third aspect, the first local data is query set data, and the second local data is support set data.

[0091] In a possible implementation manner of the third aspect, the fusion parameter includes a fused gradient and / or a fused loss.

[0092] In a possible implementation manner of the third aspect, the updated parameters of the first sub-model include model parameters and / or gradient parameters.

[0093] In a fourth aspect, the present application provides a communication device, which is a second node, or a component of the second node (such as a processor, chip, or chip system), or a logic module or software capable of implementing all or part of the functions of the second node. In the fourth aspect and its possible implementations, the communication device is described as a second node, and the second node can be a terminal device or a network device.

[0094] The device includes a processing unit and a transceiver unit; the processing unit is used to determine first information, and the first information is used to determine fusion parameters obtained by processing the first local data of N second nodes on the first sub-model respectively, where N is a positive integer; wherein the first sub-model is included in the learning model, and the learning model also includes the second sub-model; the transceiver unit is used to send the first information; the transceiver unit is also used to receive second information, and the second information is used to indicate the parameters of the updated first sub-model, and the updated first sub-model is obtained by processing the first sub-model based on the fusion parameters and the second local data of the first node; wherein the parameters of the updated first sub-model are used to determine the updated learning model.

[0095] In a possible implementation manner of the fourth aspect, the first information includes the fusion parameter.

[0096] In a possible implementation manner of the fourth aspect, the transceiver unit is further used to send third information, where the third information is used to indicate parameters of the updated learning model.

[0097] In a possible implementation of the fourth aspect, the first information includes one or more sub-information among N sub-information; wherein, the N sub-information are respectively used to indicate the parameters obtained by processing the first local data of the N second nodes on the first sub-model.

[0098] In a possible implementation of the fourth aspect, the transceiver unit is further used to send fourth information, where the fourth information is used to indicate the parameters of the updated second sub-model; the parameters of the updated second sub-model are used to determine the updated learning model.

[0099] In a possible implementation manner of the fourth aspect, the transceiver unit is further used to receive fifth information, where the fifth information is used to indicate a fusion result obtained based on the K fourth information, where K is an integer greater than 1.

[0100] In a possible implementation of the fourth aspect, the transceiver unit is further used to receive sixth information, where the sixth information is used to indicate the loss and / or gradient obtained by processing the second sub-model by the first local data of the first node.

[0101] In a possible implementation of the fourth aspect, the transceiver unit is further used to send or receive first downlink control information DCI and second DCI, the first DCI being used to indicate the resources carrying the first information, and the second DCI being used to indicate the resources carrying the second information; wherein both the first DCI and the second DCI include a learning phase indication field, and the value of the learning phase indication field of the first DCI is different from the value of the learning phase indication field of the second DCI.

[0102] In a possible implementation manner of the fourth aspect, the first local data is query set data, and the second local data is support set data.

[0103] In a possible implementation manner of the fourth aspect, the fusion parameter includes a fused gradient and / or a fused loss.

[0104] In a possible implementation manner of the fourth aspect, the parameters of the updated first sub-model include model parameters and / or gradient parameters.

[0105] In a fifth aspect, the present application provides a communication device, comprising at least one processor coupled to a memory; the memory is used to store programs or instructions; the at least one processor is used to execute the program or instructions so that the device implements the method of the aforementioned first aspect or any possible implementation method of the first aspect.

[0106] In the sixth aspect of the present application, a communication device is provided, comprising at least one processor coupled to a memory; the memory is used to store programs or instructions; the at least one processor is used to execute the program or instructions so that the device implements the aforementioned second aspect or any possible implementation method of the second aspect.

[0107] A seventh 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 as described in the first aspect or any possible implementation of the first aspect.

[0108] In an eighth aspect, 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 as described in the second aspect or any possible implementation of the second aspect.

[0109] In one possible design, the communication device provided in the fifth, sixth, seventh or eighth aspect above may be a chip or a chip system.

[0110] In the ninth aspect of the present application, a computer-readable storage medium is provided, which is used to store one or more computer-executable instructions. When the computer-executable instructions are executed by a processor, the processor executes the method in any possible implementation of any aspect of the first to second aspects above.

[0111] The tenth aspect of the present application provides a computer program product (or computer program). When the computer program product is executed by the processor, the processor executes any possible implementation method of any aspect of the first to second aspects above.

[0112] In the eleventh aspect of the present application, a chip system is provided, which includes at least one processor for supporting a communication device to implement the functions involved in any possible implementation of any one of the first to second aspects above.

[0113] In one possible design, the chip system may further include a memory for storing program instructions and data necessary for the first communication device. The chip system may be composed of a chip or may include a chip and other discrete components. Optionally, the chip system may further include an interface circuit for providing program instructions and / or data to the at least one processor.

[0114] The twelfth aspect of the present application provides a communication system, which includes the communication device of the third aspect and the communication device of the fourth aspect, and / or the communication system includes the communication device of the fifth aspect and the communication device of the sixth aspect, and / or the communication system includes the communication device of the seventh aspect and the communication device of the eighth aspect.

[0115] Among them, the technical effects brought about by any design method in the third aspect to the twelfth aspect can refer to the technical effects brought about by the different design methods in the above-mentioned first aspect to the second aspect, and will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0116] FIG1a is a schematic diagram of a communication system provided by the present application;

[0117] FIG1b is another schematic diagram of the communication system provided by the present application;

[0118] FIG1c is another schematic diagram of the communication system provided by the present application;

[0119] FIG2a is a schematic diagram of the AI ​​processing process involved in this application;

[0120] FIG2 b is another schematic diagram of the AI ​​processing process involved in this application;

[0121] FIG2c is another schematic diagram of the AI ​​processing process involved in this application;

[0122] FIG2 d is another schematic diagram of the AI ​​processing process involved in this application;

[0123] FIG2e is a schematic diagram of the AI ​​processing process involved in this application;

[0124] FIG2f is another schematic diagram of the AI ​​processing process involved in this application;

[0125] FIG2g is another schematic diagram of the AI ​​processing process involved in this application;

[0126] FIG3 is an interactive schematic diagram of the communication method provided by this application;

[0127] FIG4 is a schematic diagram of a communication scenario provided by this application;

[0128] FIG5 is another schematic diagram of a communication scenario provided by this application;

[0129] FIG6 is another interactive schematic diagram of the communication method provided by this application;

[0130] FIG7 is another interactive schematic diagram of the communication method provided by this application;

[0131] FIG8 is a schematic diagram of a communication device provided by the present application;

[0132] FIG9 is another schematic diagram of a communication device provided by the present application;

[0133] FIG10 is another schematic diagram of a communication device provided by the present application;

[0134] FIG11 is another schematic diagram of the communication device provided in this application. DETAILED DESCRIPTION

[0135] First, some of the terms used in the embodiments of the present application are explained to facilitate understanding by those skilled in the art.

[0136] (1) Terminal device: It can be a wireless terminal device that can receive network device scheduling and instruction information. The wireless terminal device can be a device that provides voice and / or data connectivity to the user, or a handheld device with wireless connection function, or other processing device connected to a wireless modem.

[0137] Terminal devices can communicate with one or more core networks or the Internet via the RAN. Terminal devices can be mobile terminal devices, such as mobile phones (also known as "cellular" phones, mobile phones), computers, and data cards. For example, they can be portable, pocket-sized, handheld, computer-built-in, or vehicle-mounted mobile devices that exchange voice and / or data with the radio access network. Examples include personal communication service (PCS) phones, cordless phones, Session Initiation Protocol (SIP) phones, wireless local loop (WLL) stations, personal digital assistants (PDAs), tablet computers, and computers with wireless transceiver capabilities. Wireless terminal equipment can also be called system, subscriber unit, subscriber station, mobile station, mobile station (MS), remote station, access point (AP), remote terminal equipment (remote terminal), access terminal equipment (access terminal), user terminal equipment (user terminal), user agent, subscriber station (SS), customer premises equipment (CPE), terminal, user equipment (UE), mobile terminal (MT), etc.

[0138] As an example and not a limitation, in the embodiments of the present application, the terminal device may also be a wearable device. Wearable devices may also be referred to as wearable smart devices or smart wearable devices, etc., which are a general term for wearable devices that are intelligently designed and developed using wearable technology for daily wear, such as glasses, gloves, watches, clothing, and shoes. A wearable device is a portable device that is worn directly on the body or integrated into the user's clothes or accessories. Wearable devices are not only hardware devices, but also achieve powerful functions through software support, data interaction, and cloud interaction. Broadly speaking, wearable smart devices include those that are fully functional, large in size, and can achieve complete or partial functions without relying on smartphones, such as smart watches or smart glasses, etc., as well as those that only focus on a certain type of application function and need to be used in conjunction with other devices such as smartphones, such as various smart bracelets, smart helmets, and smart jewelry for vital sign monitoring.

[0139] The terminal may 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 remote medical, 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.

[0140] In addition, the terminal device may also be a terminal device in a communication system that has evolved after the fifth generation (5G) communication system (e.g., a sixth generation (6G) communication system) or a terminal device in a future public land mobile network (PLMN). For example, the 6G network can further expand the form and function of 5G communication terminals. 6G terminals include but are not limited to vehicles, cellular network terminals (with integrated satellite terminal functions), drones, and Internet of Things (IoT) devices.

[0141] In an embodiment of the present application, the terminal device may also obtain AI services provided by the network device. Optionally, the terminal device may also have AI processing capabilities.

[0142] (2) Network equipment: It can be a device in a wireless network. For example, the network equipment can be a RAN node (or device) that connects a terminal device to a wireless network, which can also be called a base station. Currently, some examples of RAN equipment include: base stations gNB (gNodeB) in 5G communication systems, transmission reception points (TRP), evolved Node B (eNB), radio network controllers (RNC), Node B (NB), home base stations (e.g., home evolved Node B, or home Node B, HNB), base band units (BBU), or wireless fidelity (Wi-Fi) access points AP, etc. In addition, in a network structure, the network equipment can include centralized unit (CU) nodes, distributed unit (DU) nodes, or RAN equipment including CU nodes and DU nodes.

[0143] The network device may be any other device that provides wireless communication functionality to the terminal device. The embodiments of this application do not limit the specific technology and device form used by the network device. For ease of description, the embodiments of this application do not limit this.

[0144] The network equipment may also include core network equipment, which may include, for example, a mobility management entity (MME), a home subscriber server (HSS), a serving gateway (S-GW), a policy and charging rules function (PCRF), and a public data network gateway (PDN gateway, P-GW) in a fourth generation (4G) network; and network elements such as an access and mobility management function (AMF), a user plane function (UPF), or a session management function (SMF) in a 5G network. In addition, the core network equipment may also include other core network equipment in a 5G network and a next generation network of a 5G network.

[0145] In an embodiment of the present application, the above-mentioned network device may also be a network node with AI capabilities, which can provide AI services for terminals or other network devices. For example, it may be an AI node on the network side (access network or core network), a computing node, a RAN node with AI capabilities, a core network element with AI capabilities, etc.

[0146] In the embodiments of the present application, the apparatus for implementing the function of the network device may be the network device, or may be a device capable of supporting the network device in implementing the function, such as a chip system, which may be installed in the network device. In the technical solutions provided in the embodiments of the present application, the technical solutions provided in the embodiments of the present application are described by taking the network device as an example.

[0147] (3) Configuration and pre-configuration: In this application, configuration and pre-configuration are used simultaneously. Configuration refers to the network device / server sending some parameter configuration information or parameter values ​​to the terminal through messages or signaling, so that the terminal can determine the communication parameters or resources during transmission based on these values ​​or information. Pre-configuration is similar to configuration, and can be parameter information or parameter values ​​pre-negotiated between the network device / server and the terminal device, or parameter information or parameter values ​​used by the base station / network device or terminal device as specified in the standard protocol, or parameter information or parameter values ​​pre-stored in the base station / server or terminal device. This application does not limit this.

[0148] Furthermore, these values ​​and parameters can be changed or updated.

[0149] (4) The terms "system" and "network" in the embodiments of the present application can be used interchangeably. "Multiple" refers to two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships can exist. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone, where A and B can be singular or plural. The character " / " generally indicates that the previous and next associated objects are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, "at least one of A, B and C" includes A, B, C, AB, AC, BC or ABC. In addition, unless otherwise specified, the ordinal numbers such as "first" and "second" mentioned in the embodiments of the present application are used to distinguish multiple objects, and are not used to limit the order, timing, priority or importance of multiple objects.

[0150] (5) “Sending” and “receiving” in the embodiments of the present application indicate the direction of signal transmission. For example, “sending information to XX” can be understood as the destination of the information being XX, which can include direct sending through the air interface, as well as indirect sending through the air interface by other units or modules. “Receiving information from YY” can be understood as the source of the information being YY, which can include direct receiving from YY through the air interface, as well as indirect receiving from YY through the air interface from 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.

[0151] In other words, sending and receiving can be performed between devices, for example, between a network device and a terminal device, or can be performed within a device, for example, sending or receiving between components, modules, chips, software modules or hardware modules within the device through a bus, wiring or interface.

[0152] It is understandable that information may be processed between the source and destination of information transmission, such as coding, modulation, etc., but the destination can understand the valid information from the source. Similar expressions in this application can be understood similarly and will not be repeated.

[0153] (6) In the embodiments of the present application, "indication" may include direct indication and indirect indication, and may also include explicit indication and implicit indication. The information indicated by a certain information (such as the indication information described below) is called information to be indicated. In the specific implementation process, there are many ways to indicate the information to be indicated, such as but not limited to, directly indicating the information to be indicated, such as the information to be indicated itself or the index of the information to be indicated. The information to be indicated may also be indirectly indicated by indicating other information, wherein the other information is associated with the information to be indicated; or only a part of the information to be indicated may be indicated, while the other part of the information to be indicated is known or agreed in advance. For example, the indication of specific information may be achieved by means of the arrangement order of each information agreed in advance (such as predefined by the protocol), thereby reducing the indication overhead to a certain extent. The present application does not limit the specific method of indication. It is understandable that for the sender of the indication information, the indication information can be used to indicate the information to be indicated, and for the receiver of the indication information, the indication information can be used to determine the information to be indicated.

[0154] In this application, unless otherwise specified, the same or similar parts between the various embodiments can refer to each other. In the various embodiments of this application, and the various methods / designs / implementations in each embodiment, if there is no special explanation and logical conflict, the terms and / or descriptions between different embodiments and the various methods / designs / implementations in each embodiment are consistent and can be referenced to each other. The technical features in different embodiments and the various methods / designs / implementations in each embodiment can be combined to form new embodiments, methods, or implementations according to their inherent logical relationships. The following description of the implementation methods of this application does not constitute a limitation on the scope of protection of this application.

[0155] 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.

[0156] Please refer to Figure 1a, which is a schematic diagram of a communication system in this application. Figure 1a exemplarily illustrates a network device and six terminal devices, namely terminal device 1, terminal device 2, terminal device 3, terminal device 4, terminal device 5, and terminal device 6. In the example shown in Figure 1a, terminal device 1 is a smart teacup, terminal device 2 is a smart air conditioner, terminal device 3 is a smart gas pump, terminal device 4 is a vehicle, terminal device 5 is a mobile phone, and terminal device 6 is a printer.

[0157] As shown in Figure 1a, the AI ​​configuration information sending entity can be a network device. The AI ​​configuration information receiving entity can be terminal devices 1-6. In this case, the network device and terminal devices 1-6 form a communication system. In this communication system, terminal devices 1-6 can send data to the network device, and the network device needs to receive data sent by terminal devices 1-6. At the same time, the network device can send configuration information to terminal devices 1-6.

[0158] For example, in Figure 1a, terminal devices 4 and 6 can also form a communication system. Terminal device 5 serves as a network device, i.e., the AI ​​configuration information sending entity; terminal devices 4 and 6 serve as terminal devices, i.e., the AI ​​configuration information receiving entities. For example, in a connected vehicle system, terminal device 5 sends AI configuration information to terminal devices 4 and 6, respectively, and receives data from them. Correspondingly, terminal devices 4 and 6 receive AI configuration information from terminal device 5 and send data to terminal device 5.

[0159] Taking the communication system shown in Figure 1a as an example, in addition to performing communication-related services, different devices (including between network devices and network devices, between network devices and terminal devices, and / or between terminal devices) may also perform AI-related services. For example, as shown in Figure 1b, taking the 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. For another example, as shown in Figure 1c, taking the terminal devices including a TV and a mobile phone as an example, communication-related services and AI-related services can also be performed between the TV and the mobile phone.

[0160] The technical solution provided in this application can be applied to wireless communication systems (such as the system shown in Figure 1a, Figure 1b or Figure 1c). The following is a brief introduction to the neural networks that may be involved in this application.

[0161] 1. Fully connected neural network.

[0162] Also known as a multilayer perceptron (MLP), an MLP consists of an input layer (left), an output layer (right), and multiple hidden layers (center). Each layer of the MLP contains several nodes, called neurons. Neurons in adjacent layers are connected to each other.

[0163] 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 connected to it in the previous layer and passes through an activation function, which can be expressed as: h=f(wx+b).

[0164] Among them, w is the weight matrix, b is the bias vector, and f is the activation function.

[0165] Alternatively, the output of the neural network can be recursively expressed as: y = f n (w n f n-1 (…)+b n ).

[0166] Where n is the index of the neural network layer, 1<=n<=N, where N is the total number of neural network layers.

[0167] 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, and the process of obtaining this mapping from random w and b using existing data is called neural network training.

[0168] Optionally, the specific training method is to use a loss function to evaluate the output of the neural network. As shown in Figure 2b, the error can be backpropagated, and the neural network parameters (including w and b) can be iteratively optimized through gradient descent until the loss function reaches a minimum value, i.e., the "optimal point" in Figure 2b. It is understandable that the neural network parameters corresponding to the "optimal point" in Figure 2b can be used as the neural network parameters in the trained AI model information.

[0169] Alternatively, the gradient descent process can be expressed as:

[0170] Among them, θ is the parameter to be optimized (including w and b), L is the loss function, and η is the learning rate, which controls the step size of gradient descent. represents the derivative operation, represents the derivative of θ with respect to L.

[0171] Alternatively, the backpropagation process can utilize the chain rule for partial derivatives. As shown in Figure 2c, the gradient of the previous layer parameters can be recursively calculated from the gradient of the next layer parameters, which can be expressed as:

[0172] Among them, w ij is the weight of node j connecting to node i, s i is the weighted sum of the inputs to node i.

[0173] 2. Federated Learning (FL)

[0174] The concept of federated learning effectively addresses the current challenges facing AI development. While fully protecting user data privacy and security, it enables collaboration between edge devices and central servers to efficiently complete model learning tasks. As shown in Figure 2d, the FL architecture is the most widely used training architecture in the FL field. The FedAvg algorithm is the foundation of FL. Its algorithmic process is roughly as follows:

[0175] (1) The center initializes the model to be trained And broadcast it to all client devices.

[0176] (2) In the round t∈[1,T], client k∈[1,K] based on the local dataset For the received global model Perform E epochs of training to obtain local training results Report it to the central node.

[0177] (3) The central node aggregates and collects the local training results from all (or some) clients. Assume that the client set that uploads the local model in round t is The center will use the number of samples of the corresponding client as the weight to perform weighted averaging to obtain a new global model. The specific update rule is: The center then sends the latest version of the global model Broadcast to all client devices for a new round of training.

[0178] (4) Repeat steps (2) and (3) until the model finally converges or the number of training rounds reaches the upper limit.

[0179] In addition to reporting local models You can also use the local gradient of training After reporting, the central node averages the local gradients and updates the global model according to the direction of the average gradient.

[0180] As you can see, in the FL framework, datasets exist on distributed nodes. Distributed nodes collect local datasets, perform local training, and report the local training results (models or gradients) to the central node. The central node itself does not have a dataset; it is only responsible for fusing the training results of distributed nodes to obtain a global model and send it to the distributed nodes.

[0181] 3. Decentralized learning.

[0182] Different from federated learning, another distributed learning architecture, decentralized learning, is shown in Figure 2e. Considering a fully distributed system without a central node, the design goal f(x) of a decentralized learning system is generally the goal f of each node. i The mean of (x), that is Where n is the number of distributed nodes, x is the parameter to be optimized. In machine learning, x is the parameter of the machine learning (such as neural network) model. Each node uses local data and local target f i (x) Calculate local gradient Then it is sent to the neighboring nodes that can be communicated with. After any node receives the gradient information sent by its neighbor, it can update the parameter x of the local model according to the following formula:

[0183] in, represents the parameters of the local model after the k+1th (k is a natural number) update in the i-th node, Represents the parameters of the local model after the kth update in the i-th node (if k is 0, it means is the parameter of the local model of the i-th node that does not participate in the update), α k Represents the tuning coefficient, Ni is the set of neighbor nodes of node i, |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 information interaction between nodes, the decentralized learning system will eventually learn a unified model.

[0184] 4. Scalable distributed learning

[0185] In the aforementioned federated learning and decentralized learning approaches, nodes need to exchange complete neural network model parameters or gradients. If a neural network model can be split into multiple sub-models, the parameters or gradients of some of these sub-models can be exchanged.

[0186] Exemplarily, the scalable neural network model can be divided into: a width scalable neural network model shown in FIG2f and a depth scalable neural network model shown in FIG2g, which are respectively constructed by horizontal superposition or vertical cascade of sub-models. For example, in FIG2f, for one of the neural network sub-models a in the complete neural network model, by vertically expanding the number of layers of the neural network, the neural network models b / c / d in FIG2f can be obtained. For another example, in FIG2g, for one of the neural network sub-models a in the complete neural network model, by horizontally expanding the number of neurons in the same layer of the neural network, the neural network models b / c / d in FIG2g can be obtained.

[0187] Distributed learning based on scalable neural network models is called scalable distributed learning. In a scalable distributed learning system, nodes can exchange parameters or gradients of sub-models instead of the parameters or gradients of the complete model, thereby reducing communication overhead.

[0188] The technical solutions provided in this application can be applied to wireless communication systems (e.g., the systems shown in Figures 1a and 1b). In wireless communication systems, communication nodes generally have both signal transceiver capabilities and computing capabilities. For example, network devices with computing capabilities primarily provide computing power to support signal transceiver capabilities (e.g., calculating the time and frequency domain resources that carry the signal), thereby enabling communication between the network device and other communication nodes.

[0189] However, in communication networks, communication nodes may have excess computing power beyond just supporting the aforementioned communication tasks. Therefore, how to utilize this computing power is a pressing technical issue.

[0190] In one possible implementation, the computing power of the communication nodes can be used in distributed learning scenarios. In a distributed learning scenario, the local data of different users may have different distributions, and different users have different performance requirements, that is, the AI ​​algorithm needs to have personalized characteristics, and different users can process local data based on personalized AI algorithms (such as machine learning models) to meet their personalized needs. As the user's location moves and the surrounding environment changes, the distribution of user data will also change. In order to quickly adapt to the new environment without excessively affecting performance, the AI ​​algorithm also needs to have small sample learning capabilities. Unlike federated learning, meta-learning is often used to complete few-shot learning and personalized learning. The steps of meta-learning are as follows:

[0191] Step 1: From the distribution Sample a set of I tasks from the tasks For example, each task is an image classification task (where image classification task can be understood as a type of N (N ≥ 2) classification task), that is, the data labels are N categories. Each task (for example, task i) contains a support set consisting of K × N samples. and a query set consisting of M×N samples K is the number of data samples for each category (a total of N categories) in the support set, and M is the number of data samples for each category (a total of N categories) in the query set. The support set can be used to train the model to an intermediate version, and the query set can be used to calculate the loss based on the intermediate version.

[0192] Step 2: For task i, use the support set Train the machine learning model and update its parameters to obtain: Among them, θ represents the original parameters of the model, θ′ i represents the updated model parameters, α represents the first learning rate, Representation based on support set The gradient value calculated by performing the gradient operation on the model.

[0193] Step 3: Using QuerySets Calculate the gradient of the loss of the updated model with respect to the original parameters θ Represents a query set based on The gradient value calculated by performing the gradient operation on the updated model in step 2.

[0194] Step 4: Update the model using the query set gradient of the above set of multiple tasks (task i, i = [1, 2, ..., I]) θ represents the original parameters of the model, β represents the second learning rate, Represents the gradient value calculated in all I tasks The harmony.

[0195] Through meta-learning, a better initialization model can be obtained in step 4. Subsequent nodes using this model can fine-tune this initialization model based on local data to obtain a personalized model.

[0196] However, in the above meta-learning method, each node participating in meta-learning uses the same model structure in steps 1 to 4 above, without considering the capabilities of different nodes and the differences in local dataset characteristics. It lacks flexibility and is difficult to meet the personalized needs of each node.

[0197] To address the above issues, this application provides a communication method and related devices for enabling the computing power of communication nodes to be applied to updating a learning model. This method improves the generalization performance of a first sub-model in the learning model while also meeting the first node's personalized requirements for the first sub-model in the learning model. This method is described in detail below with reference to the accompanying figures.

[0198] Please refer to FIG3 , which is a schematic diagram of the communication method provided in this application. The method includes the following steps.

[0199] It should be noted that, in FIG3 , the method is illustrated by taking the first node and the second node as the execution subjects of the interactive diagram as an example, but the present application does not limit the execution subjects of the interactive diagram. For example, the execution subject of S301 in FIG3 and the corresponding implementation is the first node, and the execution subject may also be a chip, chip system, or processor that supports the first node to implement the method, or a logic module or software that can implement all or part of the functions of the first node. The second node in S301-S302 in FIG3 and the corresponding implementation may also be replaced by a chip, chip system, or processor that supports the second node to implement the method, or may be replaced by a logic module or software that can implement all or part of the functions of the second node. Among them, the first node may be a terminal device or a network device, and the second node may also be a terminal device or a network device.

[0200] S301. A first node sends first information, and correspondingly, a second node receives the first information.

[0201] In step S301, the first information is used to determine fusion parameters obtained by processing the first local data of N second nodes on the first sub-model, where N is a positive integer. In addition, the first sub-model is included in a learning model, which also includes a second sub-model.

[0202] S302. The second node sends second information, and correspondingly, the first node receives the second information.

[0203] In step S302, the second information is used to indicate the parameters of the updated first sub-model, which is obtained by processing the first sub-model based on the fusion parameters and the second local data of the first node, and the parameters of the updated first sub-model are used to determine the updated learning model.

[0204] It should be understood that the learning model can be understood as a model used for AI learning and reasoning through AI methods, and the learning model can be replaced by AI model, AI learning model, machine learning model, neural network, AI network, etc.

[0205] It should be understood that the nodes participating in the learning model update may include a first node and N second nodes. The nodes participating in the learning model update may be communication nodes, that is, the first node and the N second nodes may all be communication nodes, and the communication nodes include network devices and / or terminal devices.

[0206] Optionally, the nodes participating in the learning model update may also include other nodes in addition to the first node and the N second nodes, which is not limited here.

[0207] Optionally, different nodes participating in the learning model update can communicate with each other in a variety of ways. For example, different nodes can be nodes in a decentralized learning model, and the system composed of different nodes can be understood as a fully distributed system without a central node; in this case, different nodes can communicate with neighboring nodes to participate in the learning model update. For another example, different nodes can be nodes in a centralized learning model, and the system composed of different nodes can be understood as a distributed system with a central node (or control node); in this case, different nodes can communicate with the central node to participate in the learning model update.

[0208] It should be noted that the first local data and the second local data can be the same data, that is, in the same node, the first local data and the second local data can be the same data. Alternatively, the first local data and the second local data can be different data, for example, the first local data can be query set data, and the second local data can be support set data. Among them, the above-mentioned learning model can be a learning model in a meta-learning scenario. Generally, the node in the meta-learning scenario can include support set data for training the model to an intermediate version, and can also include query set data for calculating the loss based on the intermediate version. In other words, the first local data used to process the model to obtain the loss and / or gradient in the node participating in the update of the learning model can be the query set data, and the second local data used to process the model can be the support set data.

[0209] It should be noted that the learning model may include at least two sub-models, such as a first sub-model and a second sub-model. Optionally, the learning model may also include other models in addition to the first sub-model and the second sub-model, which are not limited here. Among them, the learning model can be used to perform an AI task, and the first sub-model and the second sub-model can be used to perform sub-tasks of the AI ​​task, that is, the AI ​​task performed by the first sub-model and the AI ​​task performed by the second sub-model can be understood as part of the AI ​​task performed by the learning model.

[0210] For example, the AI ​​task performed by the learning model could be an autonomous driving task. Generally, image recognition is a necessary component of autonomous driving tasks. For example, image recognition tasks such as human image recognition, license plate image recognition, and obstacle image recognition can be subtasks of the autonomous driving task. Accordingly, the AI ​​task performed by the first or second submodel can be any of these image recognition tasks.

[0211] For another example, the AI ​​task performed by the learning model can be an image classification task (or N-classification task). Generally, image classification can include image classification between different animals (for example, classifying images of cats and dogs), image classification between different plants (for example, classifying images of herbaceous plants and woody plants), etc. In other words, tasks such as image classification between different animals and image classification between different plants can be subtasks of the image classification task. Accordingly, the AI ​​task performed by the first sub-model or the second sub-model can be any of the image classification tasks between different animals and the image classification tasks between different plants.

[0212] For example, the AI ​​task performed by the learning model can be a machine translation task. Generally, word segmentation tasks, grammatical analysis tasks, sentence rewriting tasks, etc. can be subtasks of the machine translation task. Accordingly, the AI ​​task performed by the first sub-model or the second sub-model can be any of the word segmentation tasks, grammatical analysis tasks, and sentence rewriting tasks.

[0213] It should be understood that the updated model (including the updated first sub-model, the updated second sub-model, the updated learning model, etc.) can be understood as a model obtained after updating / iterating / improving the parameters in the model.

[0214] Optionally, the basis for determining the updated learning model may include only the parameters of the updated first sub-model. Alternatively, the basis for determining the updated learning model may include the parameters of the updated first sub-model, and may also include the parameters of other updated sub-models (e.g., the second sub-model). Taking the AI ​​task performed by the above-mentioned learning model as an example, where the learning model only needs to optimize a single sub-task (e.g., the single sub-task is a license plate image recognition task), the process of updating the learning model can be understood as updating the first sub-model in the learning model for performing the license plate image recognition task. Accordingly, the basis for determining the updated learning model may include only the parameters of the updated first sub-model. Where the learning model needs to optimize M (M is an integer greater than or equal to 2) sub-tasks, the process of updating the learning model can be understood as updating the M sub-models in the learning model for performing the M sub-tasks (the M sub-models include the first sub-model and other sub-models). Accordingly, the basis for determining the updated learning model may include the parameters of the updated first sub-model, and the parameters of other sub-models (e.g., the second sub-model).

[0215] Optionally, the local information (including local data and / or local capabilities, etc.) of the nodes participating in the learning model update may be different, and accordingly, the sub-models that different nodes focus on may be different. In the present application, the sub-model that the first node focuses on includes the first sub-model, which can be called the focus sub-model of the first node, the target sub-model, etc.; and the other sub-models in the learning model (such as the second sub-model) can be called the non-focus sub-model of the first node, other sub-models, etc. It can be understood that the other sub-models (such as the second sub-model) can be the focus sub-model, target sub-model, etc. of other nodes (such as the second node). Taking the AI ​​task performed by the above-mentioned learning model as an example, which can be an autonomous driving task, the local data of the first node can include data collected by a parking lot camera. For this reason, the sub-model that the first node focuses on can be a sub-model that performs a license plate image recognition task; and the local data of the second node can include data collected by a camera near the sidewalk. For this reason, the sub-model that the second node focuses on can be a sub-model that performs a human image recognition task.

[0216] In one possible implementation, in the method shown in FIG3 , the method further includes: the first node sending sixth information, where the sixth information is used to indicate the loss and / or gradient obtained by processing the second sub-model using the first local data of the first node. Specifically, the first node may further send sixth information indicating the loss and / or gradient obtained by processing the second sub-model using the first local data of the first node, so that a recipient of the sixth information can obtain the loss and / or gradient and subsequently update the second sub-model based on the loss and / or gradient.

[0217] In one possible implementation, in the method shown in FIG3 , the method further includes: the first node sending or receiving first downlink control information (DCI) and a second DCI, the first DCI being used to indicate a resource carrying the first information, and the second DCI being used to indicate a resource carrying the second information; wherein both the first DCI and the second DCI include a learning phase indication field, and the value of the learning phase indication field of the first DCI is different from the value of the learning phase indication field of the second DCI. Specifically, the resource carrying the first information and the resource carrying the second information can be scheduled through different DCIs, wherein the first node can serve as the scheduler of the resource or the first node can serve as the scheduled party of the resource, so that the first node implements resource scheduling by sending or receiving the first DCI and the second DCI.

[0218] In addition, the first DCI and the second DCI may include a field for indicating a learning phase, and different resources may be indicated by different values ​​of the field.

[0219] In one possible implementation, the fusion parameter includes a fused gradient and / or a fused loss. Specifically, the fusion parameter obtained by processing the first sub-model using the first local data of the N second nodes may be obtained by fusing the gradients and / or losses obtained by processing the first sub-model using the first local data of the N second nodes, thereby improving the generalization capability of the updated first sub-model and the updated learning model subsequently obtained based on the fusion parameter.

[0220] It should be understood that the first information can be used to determine the fusion parameters obtained by processing the first sub-model by the first local data of N second nodes respectively. Generally, the fused gradient contained in the fusion parameters can refer to processing the parameters of a sub-model to obtain the gradient, and the fused loss contained in the fusion parameters can refer to processing the parameters of a learning model to determine the loss (or loss function).

[0221] Optionally, when the fusion parameter includes the fused gradient, for the first node, the first node may locally process the first sub-model based on the second local data of the first node to obtain an intermediate version of the first sub-model, and the first node may process the intermediate version of the first sub-model based on the fused gradient to obtain an updated first sub-model. Where the fusion parameter includes the fused loss, for the first node, the first node may locally process the learning model based on the fused loss to obtain an intermediate version of the learning model, and then the first node may process the first sub-model in the intermediate version of the learning model based on the second local data of the first node to obtain an updated first sub-model.

[0222] Based on the technical solution shown in FIG3 , after receiving the first information, the first node can determine the fusion parameters obtained by processing the first sub-model respectively by the first local data of the N second nodes based on the first information; thereafter, the first node can process the first sub-model based on the fusion parameters and the second local data of the first node to obtain the updated first sub-model. In other words, the first node can implement the model update of the first sub-model in the learning model based on the fusion parameters obtained by processing the first sub-model respectively by the first local data of other nodes (i.e., the N second nodes) and the second local data of the first node. Thus, using the first local data of other nodes and the second local data of the first node as the basis for updating the first sub-model can improve the model generalization performance of the first sub-model in the learning model while also meeting the personalized needs of the first node for the first sub-model in the learning model.

[0223] Furthermore, in the above technical solution, the first node may further transmit second information indicating updated parameters of the first sub-model. The updated parameters of the first sub-model are used to determine an updated learning model. Thus, during the learning model update process, the updated learning model determined based on the updated first sub-model can also improve model generalization performance and meet the personalized needs of the node.

[0224] In the method shown in FIG3 , the first node and the N second nodes may be located in a variety of possible architectures. Accordingly, the first node may receive the first information in step S301 in a variety of ways, which will be described below with reference to the accompanying drawings.

[0225] In implementation method 1, as shown in FIG4 , where N is greater than 2, a central node may exist in the architecture where the first node and the N second nodes reside. The interaction process between the nodes can be implemented with reference to the architecture shown in FIG2 d . Accordingly, in implementation method 1, the first information received by the first node in step S301 includes the fusion parameter.

[0226] It should be understood that in FIG4 , the central node may be any one of the N second nodes, or the central node may be different from the first node and the N second nodes, which is not limited here.

[0227] As an implementation example, the following will take the example that the central node can be any one of the N second nodes, and describe the implementation process in conjunction with Figure 5. As shown in Figure 5, the following steps are included.

[0228] S501. The second node sends parameters of a learning model, and the first node receives the parameters of the learning model accordingly. After the first node receives the parameters of the learning model, the first node may locally generate or construct the learning model based on the parameters of the learning model.

[0229] It should be noted that, compared with the updated learning model mentioned in other steps of this application, the learning model generated or constructed by the parameters of the interaction in step S501 here can be called the initial learning model.

[0230] Optionally, in addition to the central node's instruction method, the first node and the second node may obtain the parameters of the initial learning model through other methods, such as predefining the parameters of the initial learning model through a protocol / standard, or obtaining the parameters of the initial learning model through instructions from a core network element or an access network element, which is not limited here. Therefore, step S501 is optional.

[0231] As an implementation example, after step S501, the first node can obtain the updated parameters of the first sub-model based on the second local data. The first node is referred to as node k, and the second local data of the first node is referred to as the support set. For example, the attention sub-model of the first node (i.e., the sub-model that the second node does not pay attention to) can be recorded as sub-model S, and the attention sub-model of the second node (i.e., the sub-model that the first node does not pay attention to) can be recorded as sub-model O. The first node uses the support set The data in the training update first sub-model is obtained:

[0232] in, represents the parameters of the first sub-model before updating, represents the updated parameters of the first sub-model, γ represents the learning rate, Represents the first submodel and other sub-models (i.e. sub-models not followed by the first node) The learning model composed of The loss function on the training data, Represents the gradient of the loss function for the parameters of the first sub-model. The training and updating of the target sub-model can be performed based on the above formula in one or more steps (T steps are taken as an example here, where T is greater than or equal to 1) to obtain the updated target sub-model parameters (T>=1) Note that in this step, the parameters of other sub-models (ie, sub-models not concerned by the first node) are not updated.

[0233] S502. The first node sends sixth information, and correspondingly, the second node receives the sixth information, where the sixth information is used to indicate the loss and / or gradient obtained by processing the second sub-model with the first local data of the first node.

[0234] It should be noted that in step S502, the second node, serving as the central node, may also receive losses and / or gradients obtained by other nodes processing non-interested sub-models of the other nodes. Wherein, at least one non-interested sub-model of the other nodes is the first sub-model, so that the second node can determine the first information based on the losses and / or gradients obtained by processing the non-interested sub-models of the other nodes.

[0235] As an implementation example, before step S502, the first node can process the second sub-model based on the first local data to obtain the loss and / or gradient. For example, the first node uses the query set Calculate the query set The loss function of the learning model for the data in And / or, the first node can calculate the loss function for other sub-models The gradient of the parameters of the second sub-model (e.g. To obtain the loss and / or gradient indicated by the sixth information.

[0236] In addition, after step S502, the second node can determine the fusion parameters obtained by processing the second sub-model by the first local data of one or more nodes (including the first node) based on the sixth information, and send indication information to the nodes that focus on the second sub-model (including other nodes among the N second nodes except the central node), and the indication information is used to indicate the fusion parameters obtained by processing the second sub-model by the first local data of the one or more nodes.

[0237] As an implementation example, in the process of determining the fusion parameter at the second node, the second node can determine the fusion parameter based on loss function fusion or gradient fusion. Take the example of the second node determining the fusion parameter obtained by processing the first local data of N second nodes on the first sub-model. After receiving the sixth information, the second node as the central node can perform loss function fusion or gradient fusion to obtain the fusion loss function. or fused gradients The superscript m indicates that the corresponding S and O are the parameters of the sub-models S and O on node m. is the query set of node m, Based on the query set The loss function of calculation is, is the gradient of the loss function with respect to the parameters of the sub-model S, P m is the number of training update steps of the second sub-model O on node m, which may be the same as or different from T above. M is the number of nodes that use the first sub-model as other sub-models (or unattended sub-models).

[0238] It should be noted that, as mentioned above, the second sub-model may not be updated. Therefore, step S502 is an optional step.

[0239] S503: The second node sends first information, and correspondingly, the first node receives the first information, wherein the first information is used to determine fusion parameters obtained by processing the first sub-model using the first local data of the N second nodes.

[0240] Specifically, in step S503, the first information received by the first node may include the fusion parameter, so that the first node can directly obtain the fusion parameter based on the received first information, and subsequently the first node can determine the updated first sub-model based on the fusion parameter.

[0241] As an implementation example, the fusion parameters received by the first node include the fusion loss (denoted as ), the first node can calculate the fusion gradient based on the fusion loss (denoted as ); or the fusion parameter received by the first node includes the fusion gradient Thereafter, the first node can be based on the fusion gradient Update the first submodel

[0242] It should be noted that the rest of the implementation process of step S503 can refer to the implementation process of step S301 above, and will not be described in detail here.

[0243] S504. The first node sends second information, and the second node receives the second information. The second information indicates updated parameters of the first sub-model, which are obtained by processing the first sub-model based on the fusion parameters and the second local data of the first node. The updated parameters of the first sub-model are used to determine an updated learning model.

[0244] It should be noted that the implementation process of step S504 can refer to the implementation process of step S302 above, and will not be repeated here.

[0245] Optionally, as described above, the basis for determining the updated learning model may include the updated parameters of the first sub-model and the parameters of other sub-models (e.g., the second sub-model). Accordingly, in addition to obtaining the updated parameters of the first sub-model in step S504, the second node may also interact with other nodes to obtain updated parameters of other sub-models (e.g., the second sub-model).

[0246] S505: The second node sends third information, and correspondingly, the first node receives the third information, wherein the third information is used to indicate the parameters of the updated learning model.

[0247] Thus, the first node can obtain the parameters of the updated learning model based on the third information, and the first node can determine the updated learning model based on the parameters to enable the first node to update the learning model containing the first sub-model and the second sub-model.

[0248] Optionally, the node that sends the first information in step S503 and the node that sends the third information in step S505 can be the same node. For example, in the scenario shown in Figure 4, the same node can be the central node of the first node and the N second nodes; that is, the node used in Figure 5 to perform the sending actions in steps S503 and S505 can be the central node in Figure 4. For another example, the same node can be a neighbor node of the first node (the neighbor node can be one of the N second nodes), and the neighbor node has the ability to process the first local data of the N second nodes, respectively, on the first sub-model to obtain parameters, and then fuse the obtained parameters to obtain fused parameters.

[0249] Optionally, the learning model update process involved in steps S501 to S505 may be performed in one or more rounds. That is, after each node executes steps S501 to S505 in a certain round, each node may re-execute the learning model involved in steps S501 to S505 based on the result of step S505 until a preset condition is met. The preset condition may be that a node (e.g., a central node, a first node, or a second node, etc.) determines that the performance of the updated learning model meets the requirements, or that the updated learning model converges to a predetermined condition.

[0250] Optionally, in the implementation process shown in FIG5 , in each round of training, the interaction between each node can be divided into two phases. With the central node as the network device and the other distributed nodes as terminal devices, each phase includes one uplink transmission (i.e., from the distributed node to the central node) and one downlink transmission (i.e., from the central node to the distributed node). The content of each transmission is different, as shown in Table 1 below.

[0251] Table 1

[0252] To this end, when a network device schedules the above four-step transmission, two bits can be added to the DCI. These two bits correspond to four values, respectively, indicating which phase and direction the node's currently scheduled DCI resources are used for transmission. Alternatively, when the direction of scheduled transmission is distinguished by different DCI formats, only one bit can be added to the DCI. This one bit corresponds to two values, respectively, indicating which phase the node's currently scheduled DCI resources are used for transmission.

[0253] In the second implementation, as shown in Figure 6 , the architecture of the first node and the N second nodes may not include a central node. The interaction between the nodes can be implemented with reference to the architecture shown in Figure 2e . Accordingly, in this second implementation, the first information received by the first node in step S301 includes N sub-information, each of which is from the N second nodes. The N sub-information indicates parameters obtained by processing the first local data of the N second nodes for the first sub-model.

[0254] As an implementation example, the implementation process will be described below in conjunction with the implementation process shown in Figure 7. As shown in Figure 7, the following steps are included.

[0255] S701a. The second node sends first information, and correspondingly, the first node receives the first information, wherein the first information is used to determine fusion parameters obtained by processing the first sub-model using the first local data of the N second nodes.

[0256] S701b: The first node sends sixth information, and correspondingly, the second node receives the sixth information, wherein the sixth information is used to indicate the loss and / or gradient obtained by processing the second sub-model with the first local data of the first node.

[0257] Through the interaction of steps S701a and S701b, each node can exchange its own first local data to process the parameters obtained by the sub-model that the node is not paying attention to, so that each node can obtain the parameters obtained by other nodes processing the sub-model that the node is paying attention to (for example, the first node can obtain the relevant parameters of the first sub-model that the first node is paying attention to, and the second node can obtain the relevant parameters of the second sub-model that the second node is paying attention to). Thereafter, each node can process the sub-model that it is paying attention to based on the obtained parameters and the second local data of each node to obtain an updated sub-model (for example, the first node can obtain the updated first sub-model, and the second node can obtain the updated second sub-model).

[0258] S702a. The first node sends second information, and the second node receives the second information. The second information indicates updated parameters of the first sub-model, which are obtained by processing the first sub-model based on the fusion parameters and the second local data of the first node. The updated parameters of the first sub-model are used to determine an updated learning model.

[0259] S702b: The second node sends fourth information, and correspondingly, the first node receives the fourth information, wherein the fourth information is used to indicate updated parameters of the second sub-model, and the updated parameters of the second sub-model are used to determine the updated learning model.

[0260] Through the interaction of step S702a and step S702b, each node can exchange the parameters of the updated sub-model that it is paying attention to, so that each node can obtain the relevant parameters after other nodes update the sub-model that the node is not paying attention to (for example, the first node can send the relevant parameters of the updated first sub-model that the first node is paying attention to, and the second node can send the relevant parameters of the updated second sub-model that the second node is paying attention to). Thereafter, each node can perform model fusion based on the obtained parameters and the updated sub-model that each node is paying attention to locally to obtain a fused learning model (for example, the first node can determine the fused learning model based on the updated first sub-model and the parameters of the updated second sub-model, and the first node can determine the fused learning model based on the updated second sub-model and the parameters of the updated first sub-model) to realize the updating process of the learning model at multiple nodes.

[0261] It should be understood that the implementation process of each information in the steps shown in FIG. 7 may refer to the implementation process of each information in the steps shown in FIG. 5 .

[0262] It should be noted that in step S702b, K pieces of fourth information may be sent via K nodes, and accordingly, the first node may receive K pieces of fourth information, where K is an integer greater than or equal to 1. It should be understood that FIG7 provides an implementation schematic diagram in which K is 1, i.e., FIG7 is illustrative using the example of the number of second nodes being 1.

[0263] Optionally, the first node may receive the N sub-information by communicating with the N second nodes respectively through communication links with the N second nodes. Alternatively, the first node may receive the N sub-information by communicating with K nodes (for example, the K nodes may be one or more of the N second nodes. Alternatively, the K nodes may be relay nodes between the first node and the N second nodes, etc.) through communication links with the K nodes, which is not limited here.

[0264] Optionally, K and N have the same value, or K is smaller than N.

[0265] Optionally, in the method shown in FIG7 , K is an integer greater than 1, and the method further includes: step S703: the first node sends fifth information, where the fifth information is used to indicate a fusion result obtained based on the K fourth information. Specifically, when the number of fourth information received by the first node is greater than 1, the first node may perform a fusion process on the parameters of the K updated second sub-models respectively indicated by the K information, and indicate the fusion result of the fusion process through the fifth information, so that the recipient of the fifth information can subsequently determine the updated learning model based on the fusion result of the K updated second sub-models.

[0266] Referring to Figure 8, an embodiment of the present application provides a communication device 800. The communication device 800 can implement the functions of the first node (the first node is a terminal device or a network device) in the above method embodiment, and thus can also achieve the beneficial effects of the above method embodiment. In the embodiment of the present application, the communication device 800 can be the first node, or it can be an integrated circuit or component within the first node, such as a chip. The following embodiments are described using the communication device 800 as the first node as an example.

[0267] In one possible implementation, when the device 800 is used to execute the method executed by the first node in any of the aforementioned embodiments, the device 800 includes a processing unit 801 and a transceiver unit 802; the transceiver unit 802 is used to receive first information, and the first information is used to determine the fusion parameters obtained by processing the first local data of N second nodes on the first sub-model respectively, where N is a positive integer; wherein the first sub-model is included in the learning model, and the learning model also includes a second sub-model; the processing unit 801 is used to determine second information, and the second information is used to indicate the parameters of the updated first sub-model, and the updated first sub-model is obtained by processing the first sub-model based on the fusion parameters and the second local data of the first node; wherein the parameters of the updated first sub-model are used to determine the updated learning model; the transceiver unit 802 is also used to send the second information.

[0268] In a possible implementation manner, the first information includes the fusion parameter.

[0269] In a possible implementation, the transceiver unit 802 is further configured to receive third information, where the third information is configured to indicate parameters of the updated learning model.

[0270] In one possible implementation, the first information includes N sub-information, and the N sub-information respectively comes from the N second nodes; wherein the N sub-information respectively indicates the parameters obtained by processing the first sub-model using the first local data of the N second nodes.

[0271] In one possible implementation, the transceiver unit 802 is also used to receive K fourth information, where the fourth information is used to indicate the parameters of the updated second sub-model, where K is an integer greater than or equal to 1; the parameters of the updated second sub-model are used to determine the updated learning model.

[0272] In a possible implementation, K is an integer greater than 1, and the transceiver unit 802 is further configured to send fifth information, where the fifth information is configured to indicate a fusion result obtained based on the K fourth information.

[0273] In a possible implementation, the transceiver unit 802 is further configured to send sixth information, where the sixth information is configured to indicate the loss and / or gradient obtained by processing the second sub-model using the first local data of the first node.

[0274] In a possible implementation, the transceiver unit 802 is further configured to send or receive first downlink control information DCI and second DCI, where the first DCI is used to indicate resources carrying the first information, and the second DCI is used to indicate resources carrying the second information;

[0275] The first DCI and the second DCI both include a learning phase indication field, and a value of the learning phase indication field of the first DCI is different from a value of the learning phase indication field of the second DCI.

[0276] In a possible implementation, the first local data is query set data, and the second local data is support set data.

[0277] In a possible implementation, the fusion parameter includes a fused gradient and / or a fused loss.

[0278] In a possible implementation, the updated parameters of the first sub-model include model parameters and / or gradient parameters.

[0279] In one possible implementation, when the device 800 is used to execute the method executed by the second node in any of the aforementioned embodiments, the device 800 includes a processing unit 801 and a transceiver unit 802; the processing unit 801 is used to determine the first information, and the first information is used to determine the fusion parameters obtained by processing the first local data of N second nodes on the first sub-model respectively, where N is a positive integer; wherein the first sub-model is included in the learning model, and the learning model also includes the second sub-model; the transceiver unit 802 is used to send the first information; the transceiver unit 802 is also used to receive the second information, and the second information is used to indicate the parameters of the updated first sub-model, and the updated first sub-model is obtained by processing the first sub-model based on the fusion parameters and the second local data of the first node; wherein the parameters of the updated first sub-model are used to determine the updated learning model.

[0280] In a possible implementation manner, the first information includes the fusion parameter.

[0281] In a possible implementation, the transceiver unit 802 is further configured to send third information, where the third information is configured to indicate parameters of the updated learning model.

[0282] In a possible implementation, the first information includes one or more sub-information among N sub-information; wherein the N sub-information are respectively used to indicate parameters obtained by processing the first sub-model using the first local data of the N second nodes.

[0283] In a possible implementation, the transceiver unit 802 is further used to send fourth information, where the fourth information is used to indicate the parameters of the updated second sub-model; the parameters of the updated second sub-model are used to determine the updated learning model.

[0284] In a possible implementation, the transceiver unit 802 is further configured to receive fifth information, where the fifth information is used to indicate a fusion result obtained based on the K fourth information, where K is an integer greater than 1.

[0285] In a possible implementation, the transceiver unit 802 is further configured to receive sixth information, where the sixth information is configured to indicate a loss and / or gradient obtained by processing the second sub-model using the first local data of the first node.

[0286] In one possible implementation, the transceiver unit 802 is also used to send or receive first downlink control information DCI and second DCI, the first DCI is used to indicate the resources carrying the first information, and the second DCI is used to indicate the resources carrying the second information; wherein, both the first DCI and the second DCI include a learning phase indication field, and the value of the learning phase indication field of the first DCI is different from the value of the learning phase indication field of the second DCI.

[0287] In a possible implementation, the first local data is query set data, and the second local data is support set data.

[0288] In a possible implementation, the fusion parameter includes a fused gradient and / or a fused loss.

[0289] In a possible implementation, the updated parameters of the first sub-model include model parameters and / or gradient parameters.

[0290] It should be noted that, for details on the information execution process of the units of the above-mentioned communication device 800, please refer to the description in the method embodiment shown above in this application, and no further details will be given here.

[0291] Please refer to Fig. 9, which is another schematic structural diagram of a communication device 900 provided in this application. The communication device 900 at least includes an input and output interface 902. The communication device 900 may be a chip or an integrated circuit.

[0292] Optionally, the communication device further includes a logic circuit 901 .

[0293] The transceiver unit 802 shown in FIG8 may be a communication interface, which may be the input / output interface 902 in FIG9 , which may include an input interface and an output interface. Alternatively, the communication interface may be a transceiver circuit, which may include an input interface circuit and an output interface circuit.

[0294] Optionally, the input-output interface 902 is used to receive first information, which is used to determine the fusion parameters obtained by processing the first local data of N second nodes on the first sub-model respectively, where N is a positive integer; wherein the first sub-model is included in the learning model, and the learning model also includes the second sub-model; the logic circuit 901 is used to determine the second information, which is used to indicate the parameters of the updated first sub-model, and the updated first sub-model is obtained by processing the first sub-model based on the fusion parameters and the second local data of the first node; wherein the parameters of the updated first sub-model are used to determine the updated learning model; the transceiver unit is also used to send the second information.

[0295] Optionally, the logic circuit 901 is used to determine the first information, which is used to determine the fusion parameters obtained by processing the first local data of N second nodes on the first sub-model respectively, where N is a positive integer; wherein the first sub-model is included in the learning model, and the learning model also includes the second sub-model; the input-output interface 902 is used to send the first information; the input-output interface 902 is used to receive the second information, which is used to indicate the parameters of the updated first sub-model, and the updated first sub-model is obtained by processing the first sub-model based on the fusion parameters and the second local data of the first node; wherein the parameters of the updated first sub-model are used to determine the updated learning model.

[0296] The logic circuit 901 and the input / output interface 902 may also execute other steps executed by the first node or the second node in any embodiment and achieve corresponding beneficial effects, which will not be described in detail here.

[0297] In a possible implementation, the processing unit 801 shown in FIG. 8 may be the logic circuit 901 in FIG. 9 .

[0298] Optionally, the logic circuit 901 may be a processing device, and the functions of the processing device may be partially or entirely implemented by software. The functions of the processing device may be partially or entirely implemented by software.

[0299] Optionally, the processing device may include a memory and a processor, wherein the memory is used to store a computer program, and the processor reads and executes the computer program stored in the memory to perform corresponding processing and / or steps in any one of the method embodiments.

[0300] Alternatively, the processing device may include only a processor. A memory for storing the computer program is located outside the processing device, and the processor is connected to the memory via circuits / wires to read and execute the computer program stored in the memory. The memory and processor may be integrated or physically separate.

[0301] Optionally, the processing device may be one or more chips, or one or more integrated circuits. For example, the processing device may be one or more field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), system-on-chips (SoCs), central processor units (CPUs), network processors (NPs), digital signal processors (DSPs), microcontroller units (MCUs), programmable logic devices (PLDs), or other integrated chips, or any combination of the above chips or processors.

[0302] Please refer to Figure 10, which shows the communication device 1000 involved in the above-mentioned embodiments provided in an embodiment of the present application. The communication device 1000 can specifically be a communication device serving as a terminal device in the above-mentioned embodiments. The example shown in Figure 10 is that the terminal device is implemented through the terminal device (or a component in the terminal device).

[0303] Herein, a possible logical structure diagram of the communication device 1000 is shown. The communication device 1000 may include but is not limited to at least one processor 1001 and a communication port 1002 .

[0304] Further optionally, the device may also include at least one of a memory 1003 and a bus 1004. In an embodiment of the present application, the at least one processor 1001 is used to control and process the actions of the communication device 1000.

[0305] In addition, the processor 1001 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, a transistor logic device, a hardware component, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. The processor can also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a digital signal processor and a microprocessor, and so on. Those skilled in the art will clearly understand that for the convenience and brevity of description, the specific working processes of the systems, devices, and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0306] It should be noted that the communication device 1000 shown in Figure 10 can be specifically used to implement the steps implemented by the terminal device in the aforementioned method embodiment and achieve the corresponding technical effects of the terminal device. The specific implementation methods of the communication device shown in Figure 10 can refer to the description in the aforementioned method embodiment and will not be repeated here.

[0307] Please refer to Figure 11, which is a structural diagram of the communication device 1100 involved in the above-mentioned embodiments provided in an embodiment of the present application. The communication device 1100 can specifically be a communication device as a network device in the above-mentioned embodiments. The example shown in Figure 11 is that the network device is implemented through the network device (or a component in the network device), wherein the structure of the communication device can refer to the structure shown in Figure 11.

[0308] The communication device 1100 includes at least one processor 1111 and at least one network interface 1114. Further optionally, the communication device also includes at least one memory 1112, at least one transceiver 1113 and one or more antennas 1115. The processor 1111, the memory 1112, the transceiver 1113 and the network interface 1114 are connected, for example, via a bus. In an embodiment of the present application, the connection may include various interfaces, transmission lines or buses, etc., which are not limited in this embodiment. The antenna 1115 is connected to the transceiver 1113. The network interface 1114 is used to enable the communication device to communicate with other communication devices through a communication link. For example, the network interface 1114 may include a network interface between the communication device and the core network device, such as an S1 interface, and the network interface may include a network interface between the communication device and other communication devices (such as other network devices or core network devices), such as an X2 or Xn interface.

[0309] Processor 1111 is primarily used to process communication protocols and communication data, control the entire communication device, execute software programs, and process software program data, for example, to support the communication device in performing the actions described in the embodiments. The communication device may include a baseband processor and a central processing unit. The baseband processor is primarily used to process communication protocols and communication data, while the central processing unit is primarily used to control the entire terminal device, execute software programs, and process software program data. Processor 1111 in Figure 11 may integrate the functions of both a baseband processor and a central processing unit. Those skilled in the art will appreciate that the baseband processor and the central processing unit may also be independent processors interconnected via a bus or other technology. Those skilled in the art will appreciate that a terminal device may include multiple baseband processors to accommodate different network standards, multiple central processing units to enhance its processing capabilities, and various components of the terminal device may be connected via various buses. The baseband processor may also be referred to as a baseband processing circuit or a baseband processing chip. The central processing unit may also be referred to as a central processing circuit or a central processing chip. The functionality for processing communication protocols and communication data may be built into the processor or stored in memory as a software program, which is executed by the processor to implement the baseband processing functionality.

[0310] The memory is primarily used to store software programs and data. Memory 1112 can exist independently and be connected to processor 1111. Alternatively, memory 1112 can be integrated with processor 1111, for example, within a single chip. Memory 1112 can store program code for executing the technical solutions of the embodiments of the present application, and execution is controlled by processor 1111. The various computer program codes executed can also be considered drivers for processor 1111.

[0311] Figure 11 shows only one memory and one processor. In an actual terminal device, there may be multiple processors and multiple memories. The memory may also be referred to as a storage medium or a storage device. The memory may 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 present embodiment.

[0312] The transceiver 1113 can be used to support the reception or transmission of radio frequency signals between the communication device and the terminal. The transceiver 1113 can be connected to the antenna 1115. The transceiver 1113 includes a transmitter Tx and a receiver Rx. Specifically, one or more antennas 1115 can receive radio frequency signals. The receiver Rx of the transceiver 1113 is used to receive the radio frequency signal from the antenna, convert the radio frequency signal into a digital baseband signal or a digital intermediate frequency signal, and provide the digital baseband signal or digital intermediate frequency signal to the processor 1111 so that the processor 1111 can further process the digital baseband signal or digital intermediate frequency signal, such as demodulation and decoding. In addition, the transmitter Tx in the transceiver 1113 is also used to receive a modulated digital baseband signal or digital intermediate frequency signal from the processor 1111, convert the modulated digital baseband signal or digital intermediate frequency signal into a radio frequency signal, and transmit the radio frequency signal through one or more antennas 1115. Specifically, the receiver Rx can selectively perform one or more stages of down-mixing and analog-to-digital conversion on the RF signal to obtain a digital baseband signal or a digital intermediate frequency signal. The order of the down-mixing and analog-to-digital conversion processes is adjustable. The transmitter Tx can selectively perform one or more stages of up-mixing and digital-to-analog conversion on the modulated digital baseband signal or digital intermediate frequency signal to obtain a RF signal. The order of the up-mixing and digital-to-analog conversion processes is adjustable. The digital baseband signal and the digital intermediate frequency signal may be collectively referred to as digital signals.

[0313] The transceiver 1113 may also be referred to as a transceiver unit, a transceiver, a transceiver device, etc. Optionally, a device in the transceiver unit that implements a receiving function may be referred to as a receiving unit, and a device in the transceiver unit that implements a transmitting function may be referred to as a transmitting unit. That is, the transceiver unit includes a receiving unit and a transmitting unit. The receiving unit may also be referred to as a receiver, an input port, a receiving circuit, etc., and the transmitting unit may be referred to as a transmitter, a transmitter, or a transmitting circuit, etc.

[0314] It should be noted that the communication device 1100 shown in Figure 11 can be specifically used to implement the steps implemented by the network device in the aforementioned method embodiment, and to achieve the corresponding technical effects of the network device. The specific implementation methods of the communication device 1100 shown in Figure 11 can refer to the description in the aforementioned method embodiment, and will not be repeated here one by one.

[0315] An embodiment of the present application also provides a computer-readable storage medium, which is used to store 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 of the first node or the second node in the above embodiment.

[0316] An embodiment of the present application also provides 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 of the above-mentioned first node or second node.

[0317] An embodiment of the present application also provides a chip system, which includes at least one processor for supporting a communication device to implement the functions involved in the possible implementation methods of the above-mentioned communication device. Optionally, the chip system also includes an interface circuit, which provides program instructions and / or data to the at least one processor. In one possible design, the chip system may also include a memory for storing program instructions and data necessary for the communication device. The chip system can be composed of a chip, or it can include a chip and other discrete devices, wherein the communication device can specifically be the first node or the second node in the aforementioned method embodiment.

[0318] An embodiment of the present application also provides a communication system, wherein the network system architecture includes a first node and a second node in any of the above embodiments, wherein the first node may be a terminal device or a network device, and the second node may also be a terminal device or a network device.

[0319] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0320] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0321] In addition, the functional units in the various embodiments of the present application can be integrated into a processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of a software functional unit. If 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 solution of the present application is essentially or the contributing part or all or part of the technical solution can be embodied in the form of a software product, which 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 method described in the various embodiments of the present application. The aforementioned storage medium includes various media that can store program code, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

Claims

1. A communication method, characterized in that: include: Receive first information, where the first information is used to determine a fusion parameter obtained by respectively processing the first sub-model with the first local data of N second nodes, where N is a positive integer; wherein the first sub-model is included in a learning model, and the learning model also includes a second sub-model; Send second information, where the second information is used to indicate parameters of an updated first sub-model, where the updated first sub-model is obtained by processing the first sub-model based on the fusion parameters and the second local data of the first node; wherein the parameters of the updated first sub-model are used to determine an updated learning model.

2. The method according to claim 1, characterized in that The first information includes the fusion parameter.

3. The method according to claim 2, characterized in that The method further comprises: Receive third information, where the third information is used to indicate parameters of the updated learning model.

4. The method according to claim 1, characterized in that: The first information includes N sub-information, and the N sub-information are respectively from the N second nodes; wherein the N sub-information are respectively used to indicate parameters obtained by processing the first sub-model with the first local data of the N second nodes.

5. The method according to claim 4, characterized in that The method further comprises: Receive K fourth information, where the fourth information is used to indicate the parameters of the updated second sub-model, and K is an integer greater than or equal to 1; the parameters of the updated second sub-model are used to determine the updated learning model.

6. The method according to claim 5, characterized in that K is an integer greater than 1, and the method further comprises: Send fifth information, where the fifth information is used to indicate a fusion result obtained based on the K fourth information.

7. The method according to any one of claims 1 to 6, characterized in that: The method further comprises: Send sixth information, where the sixth information is used to indicate the loss and / or gradient obtained by processing the second sub-model with the first local data of the first node.

8. The method according to any one of claims 1 to 7, characterized in that: The method further comprises: Sending or receiving first downlink control information DCI and second DCI, where the first DCI is used to indicate resources carrying the first information, and the second DCI is used to indicate resources carrying the second information; The first DCI and the second DCI both include a learning phase indication field, and a value of the learning phase indication field of the first DCI is different from a value of the learning phase indication field of the second DCI.

9. The method according to any one of claims 1 to 8, characterized in that: The first local data is query set data, and the second local data is support set data.

10. The method according to any one of claims 1 to 9, characterized in that: The fusion parameters include fused gradients and / or fused losses.

11. The method according to any one of claims 1 to 10, characterized in that: The updated parameters of the first sub-model include model parameters and / or gradient parameters.

12. A communication method, characterized in that: include: Sending first information, where the first information is used to determine a fusion parameter obtained by respectively processing the first sub-model with the first local data of N second nodes, where N is a positive integer; wherein the first sub-model is included in a learning model, and the learning model also includes a second sub-model; Receive second information, where the second information is used to indicate parameters of an updated first sub-model, where the updated first sub-model is obtained by processing the first sub-model based on the fusion parameters and the second local data of the first node; wherein the parameters of the updated first sub-model are used to determine an updated learning model.

13. The method according to claim 12, characterized in that The first information includes the fusion parameter.

14. The method according to claim 13, characterized in that The method further comprises: Send third information, where the third information is used to indicate parameters of the updated learning model.

15. The method according to claim 12, characterized in that The first information includes one or more sub-information among N sub-information; wherein the N sub-information are respectively used to indicate parameters obtained by processing the first sub-model by the first local data of the N second nodes.

16. The method according to claim 15, characterized in that The method further comprises: Send fourth information, where the fourth information is used to indicate the updated parameters of the second sub-model; the updated parameters of the second sub-model are used to determine the updated learning model.

17. The method according to claim 16, characterized in that The method further comprises: Receive fifth information, where the fifth information is used to indicate a fusion result obtained based on the K fourth information, where K is an integer greater than 1.

18. The method according to any one of claims 12 to 17, characterized in that The method further comprises: Receive sixth information, where the sixth information is used to indicate the loss and / or gradient obtained by processing the second sub-model with the first local data of the first node.

19. The method according to any one of claims 12 to 18, characterized in that The method further comprises: Sending or receiving first downlink control information DCI and second DCI, where the first DCI is used to indicate resources carrying the first information, and the second DCI is used to indicate resources carrying the second information; The first DCI and the second DCI both include a learning phase indication field, and a value of the learning phase indication field of the first DCI is different from a value of the learning phase indication field of the second DCI.

20. The method according to any one of claims 12 to 19, characterized in that The first local data is query set data, and the second local data is support set data.

21. The method according to any one of claims 12 to 20, characterized in that The fusion parameters include fused gradients and / or fused losses.

22. The method according to any one of claims 12 to 21, characterized in that The updated parameters of the first sub-model include model parameters and / or gradient parameters.

23. A communication device, characterized in that: including a transceiver unit and a processing unit; The transceiver unit is used to receive first information, and the first information is used to determine the fusion parameters obtained by respectively processing the first sub-model by the first local data of N second nodes, where N is a positive integer; wherein the first sub-model is included in the learning model, and the learning model also includes the second sub-model; The processing unit is used to determine second information, where the second information is used to indicate parameters of an updated first sub-model, where the updated first sub-model is obtained by processing the first sub-model based on the fusion parameter and the second local data of the first node; wherein the parameters of the updated first sub-model are used to determine an updated learning model; The transceiver unit is further configured to send the second information.

24. The device according to claim 23, characterized in that The first information includes the fusion parameter.

25. The device according to claim 24, characterized in that The transceiver unit is further used to receive third information, where the third information is used to indicate parameters of the updated learning model.

26. The device according to claim 23, characterized in that The first information includes N sub-information, and the N sub-information are respectively from the N second nodes; wherein the N sub-information are respectively used to indicate parameters obtained by processing the first sub-model with the first local data of the N second nodes.

27. The device according to claim 26, characterized in that The transceiver unit is also used to receive K fourth information, where the fourth information is used to indicate the parameters of the updated second sub-model, where K is an integer greater than or equal to 1; the parameters of the updated second sub-model are used to determine the updated learning model.

28. The device according to claim 27, characterized in that K is an integer greater than 1, and the transceiver unit is further used to send fifth information, where the fifth information is used to indicate a fusion result obtained based on the K fourth information.

29. The device according to any one of claims 23 to 28, characterized in that The transceiver unit is further used to send sixth information, where the sixth information is used to indicate the loss and / or gradient obtained by processing the second sub-model with the first local data of the first node.

30. The device according to any one of claims 23 to 29, characterized in that The transceiver unit is further used to send or receive first downlink control information DCI and second DCI, the first DCI is used to indicate resources carrying the first information, and the second DCI is used to indicate resources carrying the second information; The first DCI and the second DCI both include a learning phase indication field, and a value of the learning phase indication field of the first DCI is different from a value of the learning phase indication field of the second DCI.

31. The device according to any one of claims 23 to 30, characterized in that The first local data is query set data, and the second local data is support set data.

32. The device according to any one of claims 23 to 31, characterized in that The fusion parameters include fused gradients and / or fused losses.

33. The device according to any one of claims 23 to 32, characterized in that The updated parameters of the first sub-model include model parameters and / or gradient parameters.

34. A communication device, characterized in that: including a transceiver unit and a processing unit; The processing unit is used to determine first information, and the first information is used to determine a fusion parameter obtained by respectively processing the first sub-model by the first local data of N second nodes, where N is a positive integer; wherein the first sub-model is included in a learning model, and the learning model also includes a second sub-model; The transceiver unit is used to send the first information; The transceiver unit is also used to receive second information, where the second information is used to indicate parameters of an updated first sub-model, where the updated first sub-model is obtained by processing the first sub-model based on the fusion parameters and the second local data of the first node; wherein the parameters of the updated first sub-model are used to determine an updated learning model.

35. The device according to claim 34, characterized in that The first information includes the fusion parameter.

36. The device according to claim 35, characterized in that The transceiver unit is further used to send third information, where the third information is used to indicate the parameters of the updated learning model.

37. The device according to claim 34, characterized in that The first information includes one or more sub-information among N sub-information; wherein the N sub-information are respectively used to indicate parameters obtained by processing the first sub-model by the first local data of the N second nodes.

38. The device according to claim 37, characterized in that The transceiver unit is also used to send fourth information, where the fourth information is used to indicate the updated parameters of the second sub-model; the updated parameters of the second sub-model are used to determine the updated learning model.

39. The device according to claim 38, characterized in that The transceiver unit is further used to receive fifth information, where the fifth information is used to indicate a fusion result obtained based on the K fourth information, where K is an integer greater than 1.

40. The device according to any one of claims 34 to 39, characterized in that The transceiver unit is further used to receive sixth information, where the sixth information is used to indicate the loss and / or gradient obtained by processing the second sub-model with the first local data of the first node.

41. The device according to any one of claims 34 to 40, characterized in that The transceiver unit is further used to send or receive first downlink control information DCI and second DCI, the first DCI is used to indicate resources carrying the first information, and the second DCI is used to indicate resources carrying the second information; The first DCI and the second DCI both include a learning phase indication field, and a value of the learning phase indication field of the first DCI is different from a value of the learning phase indication field of the second DCI.

42. The device according to any one of claims 34 to 41, characterized in that The first local data is query set data, and the second local data is support set data.

43. The device according to any one of claims 34 to 42, characterized in that The fusion parameters include fused gradients and / or fused losses.

44. The device according to any one of claims 34 to 43, characterized in that The updated parameters of the first sub-model include model parameters and / or gradient parameters.

45. A communication device, characterized in that: The method comprises at least one processor coupled to a memory; the at least one processor is configured to execute the method according to any one of claims 1 to 11.

46. ​​The communication device according to claim 45, characterized in that The communication device is a chip or a chip system.

47. A communication device, characterized in that: comprising at least one processor coupled to a memory; The at least one processor is configured to execute the method of any one of claims 12 to 22.

48. The communication device according to claim 47, characterized in that The communication device is a chip or a chip system.

49. A communication system, characterized in that: The system comprises a communication device as claimed in any one of claims 23 to 33, and a communication device for executing any one of claims 34 to 44; or The system comprises a communication device for executing the communication device according to claim 45, and the communication device according to claim 46.

50. A computer-readable storage medium, characterized in that The medium stores instructions, and when the instructions are executed by a computer, the method according to any one of claims 1 to 22 is implemented.

51. A computer program product, characterized in that The method comprises instructions which, when executed on a computer, cause the computer to perform the method according to any one of claims 1 to 22.

Images