WO2024026846A1 - 一种人工智能模型处理方法及相关设备 - Google Patents

一种人工智能模型处理方法及相关设备 Download PDF

Info

Publication number
WO2024026846A1
WO2024026846A1 PCT/CN2022/110616 CN2022110616W WO2024026846A1 WO 2024026846 A1 WO2024026846 A1 WO 2024026846A1 CN 2022110616 W CN2022110616 W CN 2022110616W WO 2024026846 A1 WO2024026846 A1 WO 2024026846A1
Authority
WO
WIPO (PCT)
Prior art keywords
model
node
information
nodes
type
Prior art date
Application number
PCT/CN2022/110616
Other languages
English (en)
French (fr)
Inventor
王坚
李榕
张公正
童文
马江镭
Original Assignee
华为技术有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 华为技术有限公司 filed Critical 华为技术有限公司
Priority to PCT/CN2022/110616 priority Critical patent/WO2024026846A1/zh
Publication of WO2024026846A1 publication Critical patent/WO2024026846A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G06N20/20—Ensemble learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L27/00—Modulated-carrier systems

Definitions

  • This application relates to the field of communications, and in particular to an artificial intelligence (artificial intelligence, AI) model processing method and related equipment.
  • AI artificial intelligence
  • Federated learning is widely used as a typical AI data processing model.
  • a central node in the federated learning system for model fusion so it is generally a star structure, which has the problem of poor robustness. Once there is a problem with the central node (such as being attacked), it will cause the entire system to be paralyzed.
  • the decentralized AI data processing model is an improvement over the federated learning AI data processing model.
  • This model does not require a central node and can improve the robustness of the system.
  • each node uses local data and local targets to train a local AI model, it interacts with its neighbor nodes with reachable communication and further processes (such as training, fusion, etc.) the local AI model based on the interactive AI model.
  • AI model After each node uses local data and local targets to train a local AI model, it interacts with its neighbor nodes with reachable communication and further processes (such as training, fusion, etc.) the local AI model based on the interactive AI model. AI model.
  • This application provides an AI model processing method and related equipment to improve the performance of the AI model.
  • the first aspect of this application provides an AI model processing method.
  • the method is executed by a first node, or the method is executed by some components (such as a processor, a chip or a chip system, etc.) in the first node, or the method It can also be implemented by a logic module or software that can realize all or part of the functions of the first node.
  • the method is described by taking an example that the method is executed by a first node.
  • the first node may be a terminal device or a network device.
  • a first node determines a first AI model; the first node sends first information indicating model information of the first AI model and auxiliary information of the first AI model.
  • the first node after the first node determines the first AI model, the first node sends first information indicating the model information of the first AI model and the auxiliary information of the first AI model.
  • the first information can also indicate the auxiliary information of the first AI model, so that the first The recipient of the information can perform AI model processing (such as training, fusion, etc.) on the model information of the first AI model based on the auxiliary information of the first AI model, and the recipient of the improved information can perform AI model processing based on the first AI model. Performance of the resulting AI model.
  • the auxiliary information of the first AI model includes at least one of the following: type information of the first AI model, identification information of the first node, Receive the identification information of the node, the version information of the first AI model, generate the time information of the first AI model, generate the geographical location information of the first AI model, and generate the distribution information of local data of the first node.
  • the auxiliary information of the first AI model indicated by the first information includes at least one of the above items, so that the recipient of the first information can perform AI on the model information of the first AI model based on the at least one item of information above.
  • Model processing improves the performance of the AI model obtained by the receiver of the first information based on the first AI model.
  • the model information of the first AI model and the auxiliary information of the first AI model are respectively carried in different transmission resources, and the transmission resources include time domain resources and/or frequency domain. resource.
  • the two can be separated carried on different transmission resources.
  • the transmission resources are preconfigured resources.
  • the first AI model is obtained based on the node type of the first node.
  • the first AI model indicated by the first information sent by the first node may be a model obtained by the first node based on the node type of the first node.
  • the first node can perform different AI model processing processes based on different node types, solving the problem of a single node function and improving flexibility. sex.
  • the first AI model is obtained based on a second AI model and the node type of the first node; the second AI model is obtained based on local data; or, the second AI model The model is obtained based on K information.
  • K information indicates the model information of the AI model of other nodes and the auxiliary information of the AI model of the other node.
  • K is a positive integer; or, the second AI model is based on the Local data and the K information are obtained.
  • the first AI model indicated by the first information sent by the first node can be a model understandable by other nodes, so as to facilitate Other nodes perform further model processing after receiving the first AI model.
  • the second AI model used to obtain the first AI model can be implemented as any of the above to improve the flexibility of solution implementation.
  • the node type includes any of the following: a node type for local training based on local data, a node type for fusion processing based on AI models of other nodes, a node type for fusion processing based on the local data.
  • the node type used to obtain the first AI model can be implemented as any of the above to improve the flexibility of solution implementation.
  • the method further includes: the first node sending indication information indicating a node type of the first node.
  • the first node can also send indication information indicating the node type of the first node, so that other nodes can clarify the node type of the first node based on the indication information, and subsequent nodes can be based on the node type of the first node. Type interacts with this first node.
  • the method further includes: the first node determines the node type of the first node based on capability information and/or demand information; or, the first node receives an indication of the first node. An indication of the node's node type.
  • the first node can determine the node type of the first node based on its own capability information and/or demand information, or can also determine the node type of the first node based on instructions from other nodes to improve the solution. Implement flexibility.
  • the first AI model is a model understandable by M nodes in the system where the first node is located, and M is an integer greater than or equal to 2.
  • the first AI model is a model that can be understood by M nodes in the system where the first node is located. It can be expressed as the first AI model is a model common to M nodes in the system where the first node is located.
  • the first AI model The model structure of the model is a model structure understandable by M nodes in the system where the first node is located.
  • the model structure of the first AI model is a model structure common to M nodes in the system where the first node is located.
  • the first AI model It is a public model of the system where the first node is located, or the model structure of the first AI model is a public model structure of the system where the first node is located.
  • the first AI model indicated by the first information sent by the first node is a model understandable by M nodes in the system where the first node is located, so that there are multiple different models among the M nodes.
  • other nodes can understand the first AI model and further perform model processing after receiving the first information.
  • the second aspect of this application provides an AI model processing method.
  • the method is executed by a second node, or the method is executed by some components (such as a processor, a chip or a chip system, etc.) in the second node, or the method It can also be implemented by a logic module or software that can realize all or part of the functions of the second node.
  • the method is described by taking the example that the method is executed by a second node.
  • the second node may be a terminal device or a network device.
  • the second node receives N pieces of first information from N first nodes, each of the N pieces of first information indicating model information of the first AI model and Auxiliary information, N is a positive integer; the second node performs model processing based on the N pieces of first information to obtain the target AI model.
  • each first piece of information indicates the model information of the first AI model and the auxiliary information of the first AI model.
  • the second node based on the Nth Perform model processing on one piece of information to obtain the target AI model.
  • each first information can also indicate the auxiliary information of the first AI model, Enable the second node to perform AI model processing (such as training, fusion, etc.) on the model information of the first AI model based on the auxiliary information of the first AI model, and improve the processing of the second node based on the first AI model to obtain The performance of the AI model.
  • the target AI model is used to complete the AI task of the second node, or the target AI model is a local model of the second node.
  • the auxiliary information of the first AI model includes at least one of the following: type information of the first AI model, identification information of the first node, reception of the first AI model The identification information of the node, the version information of the first AI model, the time information of generating the first AI model, the geographical location information of generating the first AI model, and the distribution information of local data of the first node.
  • the auxiliary information of the first AI model indicated by the first information includes at least one of the above items, so that the second node can perform AI model processing on the model information of the first AI model based on the at least one item of information above, Improve the performance of the AI model processed by the second node based on the first AI model.
  • the model information of the first AI model and the auxiliary information of the first AI model are respectively carried in different transmission resources, and the transmission resources include time domain resources and/or frequency domain. resource.
  • the two can be separated carried on different transmission resources.
  • the transmission resources are preconfigured resources.
  • the second node performs model processing based on the N pieces of first information
  • obtaining the target AI model includes: the second node based on the N pieces of first information and the second node Perform model processing on the node type to obtain the target AI model.
  • the target AI model obtained by the second node through model processing can be a model obtained by the second node based on at least the node type of the second node.
  • the second node can perform different AI model processing processes based on different node types, solving the problem of single node function to improve flexibility.
  • the second node performs model processing based on the N pieces of first information and the node type of the second node, and obtaining the target AI model includes: at the node of the second node When the type is a node type for fusion processing based on the AI models of other nodes, the second node performs model fusion on the N first AI models based on the N pieces of first information to obtain the target AI model.
  • the node type of the second node is a node type for fusion processing based on the AI models of other nodes
  • the second node receives N first information and determines N first AI models
  • the second node perform model fusion on the N first AI models to obtain the target AI model.
  • the second node performs model processing based on the N pieces of first information and the node type of the second node
  • obtaining the target AI model includes: at the node of the second node
  • the type is a node type that performs local training based on the local data and performs fusion processing based on the AI models of other nodes
  • the second node models the N first AI models and the second AI model based on the N pieces of first information.
  • the target AI model is obtained, wherein the second AI model is trained based on local data.
  • the second node determines N pieces of information after receiving N pieces of first information. After the first AI model, the second node needs to train the second AI model based on local data, and perform model fusion on the N first AI models and the second AI model to obtain the target AI model.
  • the method further includes: the second node receiving indication information indicating a node type of the first node.
  • the second node can also receive indication information indicating the node type of the first node, so that the second node can clarify the node type of the first node based on the indication information. Subsequently, the second node can based on The node type of the first node interacts with the first node.
  • the method further includes: the second node sending indication information indicating a node type of the second node.
  • the second node can also send indication information indicating the node type of the second node, so that other nodes can clarify the node type of the second node based on the indication information, and subsequent nodes can be based on the node type of the second node. Type interacts with this second node.
  • the method further includes: the second node determines the node type of the second node based on capability information and/or demand information; or, the second node receives an indication that the second node An indication of the node's node type.
  • the second node can determine the node type of the second node based on its own capability information and/or demand information, or can also determine the node type of the second node based on instructions from other nodes to improve the solution. Implement flexibility.
  • the first AI model is a model understandable by M nodes in the system where the second node is located, and M is an integer greater than or equal to 2.
  • the first AI model is a model that can be understood by M nodes in the system where the first node is located. It can be expressed as the first AI model is a model common to M nodes in the system where the first node is located.
  • the first AI model The model structure of the model is a model structure understandable by M nodes in the system where the first node is located.
  • the model structure of the first AI model is a model structure common to M nodes in the system where the first node is located.
  • the first AI model It is a public model of the system where the first node is located, or the model structure of the first AI model is a public model structure of the system where the first node is located.
  • the first AI model indicated by each first information is a model understandable by M nodes in the system where the second node is located, so as to facilitate When there are multiple different node types among the M nodes, each node can understand the first AI model and further perform model processing after receiving the first information.
  • the third aspect of this application provides an AI model processing method, which is executed by the first node, or the method is executed by some components (such as a processor, a chip or a chip system, etc.) in the first node, or the method It can also be implemented by a logic module or software that can realize all or part of the functions of the first node.
  • the method is described by taking the example that the method is executed by a first node.
  • the first node may be a terminal device or a network device.
  • the first node obtains a local AI model, and the local AI model is used to complete the AI task of the first node; the first node determines a public AI model based on the local AI model, and the public AI model is the first node.
  • the first node obtains the local AI model used to complete the AI task of the first node.
  • the first node determines the public AI model based on the local AI model.
  • the first node sends first information indicating model information of the public AI model.
  • the public AI model is a model understandable by M nodes in the system where the first node is located.
  • the public AI model indicated by the first information sent by the first node is a model understandable by M nodes in the system where the first node is located, so as to facilitate the situation where there are multiple different node types among the M nodes. Under this condition, other nodes can understand the public AI model and further perform model processing after receiving the first information.
  • the public AI model is a model that can be understood by M nodes in the system where the first node is located. It can be expressed as the public AI model is a model common to M nodes in the system where the first node is located.
  • the model of the public AI model The structure is a model structure understandable by M nodes in the system where the first node is located, and the model structure of the public AI model is a model structure common to M nodes in the system where the first node is located.
  • the first node determining the public AI model based on the local AI model includes: the first node determining the public AI model based on the local AI model and the node type of the first node.
  • the first node can determine the public AI model based on the local AI model and the node type of the first node, so that the first node can perform different AI model processing processes based on different node types, solving Node has a single function to improve flexibility.
  • the node type includes any of the following: a node type for local training based on local data, a node type for fusion processing based on AI models of other nodes, a node type for fusion processing based on the local data.
  • the node type used to obtain the public AI model can be implemented as any of the above to improve the flexibility of solution implementation.
  • the method further includes: the first node sending indication information indicating a node type of the first node.
  • the first node can also send indication information indicating the node type of the first node, so that other nodes can clarify the node type of the first node based on the indication information, and subsequent nodes can be based on the node type of the first node. Type interacts with this first node.
  • the method further includes: the first node determines the node type of the first node based on capability information and/or demand information; or, the first node receives an indication of the first node. An indication of the node's node type.
  • the first node can determine the node type of the first node based on its own capability information and/or demand information, or can also determine the node type of the first node based on instructions from other nodes to improve the solution. Implement flexibility.
  • the local AI model is obtained based on local data; or,
  • the local AI model is obtained based on K pieces of information.
  • Each piece of information in the K pieces of information indicates the model information of the AI model of other nodes and the auxiliary information of the AI model of the other nodes.
  • K is a positive integer; or,
  • the local AI model is obtained based on the local data and the K pieces of information.
  • the local AI model used to obtain the public AI model can be implemented as any of the above to improve the flexibility of solution implementation.
  • the first information also indicates auxiliary information of the public AI model.
  • the first information sent by the first node also indicates the auxiliary information of the public AI model.
  • the first information can also indicate the auxiliary information of the public AI model, so that the first information
  • the receiver can perform AI model processing (such as training, fusion, etc.) on the model information of the public AI model based on the auxiliary information of the public AI model, and improve the AI model obtained by the receiver of the first information based on the public AI model. performance.
  • the auxiliary information of the public AI model includes at least one of the following: type information of the public AI model, identification information of the first node, and information of the receiving node of the public AI model. Identification information, version information of the public AI model, time information for generating the public AI model, geographical location information for generating the public AI model, and distribution information of local data of the first node.
  • the auxiliary information of the public AI model indicated by the first information includes at least one of the above, so that the recipient of the first information can perform AI model processing on the model information of the public AI model based on the at least one of the above information. , improving the performance of the AI model obtained by the receiver of the first information based on the public AI model.
  • the model information of the public AI model and the auxiliary information of the public AI model are respectively carried in different transmission resources, and the transmission resources include time domain resources and/or frequency domain resources.
  • the two can be carried separately on on different transmission resources.
  • the fourth aspect of this application provides an AI model processing method.
  • the method is executed by a second node, or the method is executed by some components (such as a processor, a chip or a chip system, etc.) in the second node, or the method It can also be implemented by a logic module or software that can realize all or part of the functions of the second node.
  • the method is described by taking the example that the method is executed by a second node.
  • the second node may be a terminal device or a network device.
  • the second node receives N pieces of first information, and each of the N pieces of first information is used to indicate model information of a public AI model.
  • the public AI model is in the system where the second node is located. Models understandable by M nodes, where M is an integer greater than or equal to 2; the second node updates the local AI model based on the N first information to obtain an updated local AI model, which is used to complete the first Two-node AI tasks.
  • each first information among the N pieces of first information received by the second node is used to indicate the model information of the public AI model, and thereafter, the second node updates the local AI model based on the N pieces of first information, Get the updated local AI model.
  • the public AI model is a model understandable by M nodes in the system where the second node is located.
  • the public AI model indicated by the first information received by the second node is a model understandable by M nodes in the system where the second node is located, so as to facilitate the situation where there are multiple different node types among the M nodes. Under this condition, other nodes can understand the public AI model and further perform model processing after receiving the first information.
  • the public AI model is a model that can be understood by M nodes in the system where the second node is located. It can be expressed as the public AI model is a model common to M nodes in the system where the first node is located.
  • the model of the public AI model The structure is a model structure understandable by M nodes in the system where the first node is located, and the model structure of the public AI model is a model structure common to M nodes in the system where the first node is located.
  • the second node updates the local AI model based on the N first pieces of information
  • the process of obtaining the updated local AI model includes: the second node based on the N first pieces of information
  • the information and the node type of the second node are used to update the local AI model to obtain an updated local AI model.
  • the second node can determine the public AI model based on the N pieces of first information and the node type of the second node, so that the second node can perform different AI model processing processes based on different node types. , solve the problem of single node function to improve flexibility.
  • the node type includes any of the following: a node type for fusion processing based on the AI models of other nodes, local training based on the local data and training based on the AI models of other nodes.
  • the node type for fusion processing includes any of the following: a node type for fusion processing based on the AI models of other nodes, local training based on the local data and training based on the AI models of other nodes.
  • the node type for fusion processing includes any of the following: a node type for fusion processing based on the AI models of other nodes, local training based on the local data and training based on the AI models of other nodes.
  • the node type used to obtain the updated local AI model can be implemented as any of the above to improve the flexibility of solution implementation.
  • the method further includes: the second node sending indication information indicating a node type of the second node.
  • the second node can also send indication information indicating the node type of the second node, so that other nodes can clarify the node type of the second node based on the indication information, and subsequent nodes can be based on the node type of the second node. Type interacts with this second node.
  • the method further includes: the second node determines the node type of the second node based on capability information and/or demand information; or, the second node receives an indication of the second node. An indication of the node's node type.
  • the second node can determine the node type of the second node based on its own capability information and/or demand information, or can also determine the node type of the second node based on instructions from other nodes to improve the solution. Implement flexibility.
  • the local AI model is obtained based on P pieces of information, and each piece of information in the P pieces of information Indicates the model information of the AI model of other nodes and the auxiliary information of the AI model of the other node, and the P is a positive integer.
  • the local AI model is obtained based on the local data, or , the local AI model is obtained based on the local data and the P pieces of information.
  • Each piece of information in the P pieces of information indicates the model information of the AI model of other nodes and the auxiliary information of the AI model of the other nodes.
  • P is a positive integer. .
  • the local AI model can be implemented differently as mentioned above to improve the flexibility of solution implementation.
  • the first information is also used to indicate auxiliary information of the public AI model.
  • each first information received by the second node also indicates the auxiliary information of the public AI model.
  • the first information indicates the model information of the public AI model
  • the first information can also indicate the auxiliary information of the public AI model, so that the second node can
  • the local AI model is updated based on the auxiliary information of the public AI model to obtain an updated local AI model, thereby improving the performance of the AI model processed by the second node based on the public AI model.
  • the auxiliary information of the public AI model includes at least one of the following: type information of the public AI model, identification information of the first node, and information of the receiving node of the public AI model. Identification information, version information of the public AI model, time information for generating the public AI model, geographical location information for generating the public AI model, and distribution information of local data of the first node.
  • the auxiliary information of the public AI model indicated by the first information includes at least one of the above, so that the second node can perform AI model processing on the model information of the public AI model based on the at least one of the above information to improve the The performance of the AI model obtained by processing the second node based on the public AI model.
  • the model information of the public AI model and the auxiliary information of the public AI model are respectively carried in different transmission resources, and the transmission resources include time domain resources and/or frequency domain resources.
  • the two can be carried separately on on different transmission resources.
  • the fifth aspect of this application provides an AI model processing method.
  • the method is executed by the first node, or the method is executed by some components (such as a processor, a chip or a chip system, etc.) in the first node, or the method It can also be implemented by a logic module or software that can realize all or part of the functions of the first node.
  • the method is described by taking an example that the method is executed by a first node.
  • the first node may be a terminal device or a network device.
  • a first node determines a node type of the first node.
  • the first node can determine the node type of the first node, and subsequently the first node can perform AI model processing based on the node type. This allows the first node to perform different AI model processing processes based on different node types, solving the problem of single node function and improving flexibility.
  • the first node sends indication information indicating a node type of the first node.
  • the first node can also send indication information indicating the node type of the first node, so that other nodes can clarify the node type of the first node based on the indication information, and subsequent nodes can be based on the node type of the first node. Type interacts with this first node.
  • the method further includes: the first node sending first information, the first information being used to indicate model information of the first AI model, wherein the first AI model is Obtained based on the node type of the first node.
  • the first AI model indicated by the first information sent by the first node may be a model obtained by the first node based on the node type of the first node.
  • the first AI model is obtained based on at least the node type of the first node, the first node can perform different AI model processing processes based on different node types, solving the problem of a single node function and improving flexibility.
  • the first AI model is obtained based on the second AI model and the node type of the first node.
  • the first AI model indicated by the first information sent by the first node can be a model understandable by other nodes, so as to facilitate Other nodes perform further model processing after receiving the first AI model.
  • the second AI model is obtained based on local data; or,
  • the second AI model is obtained based on K information, each information in the K information indicates the model information of the AI model of other nodes and the auxiliary information of the AI model of the other node, K is a positive integer; or,
  • the second AI model is obtained based on the local data and the K pieces of information.
  • the second AI model used to obtain the first AI model can be implemented as any of the above to improve the flexibility of solution implementation.
  • the node type includes any of the following: a node type for local training based on local data, a node type for fusion processing based on AI models of other nodes, a node type for fusion processing based on the local data.
  • the node type used to obtain the first AI model can be implemented as any of the above to improve the flexibility of solution implementation.
  • the first information is also used to indicate auxiliary information of the first AI model.
  • the first information sent by the first node also indicates the auxiliary information of the first AI model.
  • the first information can also indicate the auxiliary information of the first AI model, so that the first The recipient of the information can perform AI model processing (such as training, fusion, etc.) on the model information of the first AI model based on the auxiliary information of the first AI model, and the recipient of the improved information can perform AI model processing based on the first AI model. Performance of the resulting AI model.
  • the auxiliary information of the first AI model includes at least one of the following: type information of the first AI model, identification information of the first node, The identification information of the receiving node of the model, the version information of the first AI model, the time information of the generated first AI model, the geographical location information of the generated first AI model, and the distribution information of the local data of the first node.
  • the auxiliary information of the first AI model indicated by the first information includes at least one of the above items, so that the recipient of the first information can perform AI on the model information of the first AI model based on the at least one item of information above.
  • Model processing improves the performance of the AI model obtained by the receiver of the first information based on the first AI model.
  • the model information of the first AI model and the auxiliary information of the first AI model are respectively carried in different transmission resources, and the transmission resources include time domain resources and/or frequency domain. resource.
  • the two can be carried separately on on different transmission resources.
  • the first AI model is a model understandable by M nodes in the system where the first node is located, and M is an integer greater than or equal to 2.
  • the first AI model indicated by the first information sent by the first node is a model understandable by M nodes in the system where the first node is located, so that there are multiple different models among the M nodes.
  • other nodes can understand the first AI model and further perform model processing after receiving the first information.
  • the method further includes: the first node determines the node type of the first node based on capability information and/or demand information; or, the first node receives an indication of the first node. An indication of the node's node type.
  • the first node can determine the node type of the first node based on its own capability information and/or demand information, or can also determine the node type of the first node based on instructions from other nodes to improve the solution. Implement flexibility.
  • the sixth aspect of this application provides an AI model processing method.
  • the method is executed by a second node, or the method is executed by some components (such as a processor, a chip or a chip system, etc.) in the second node, or the method It can also be implemented by a logic module or software that can realize all or part of the functions of the second node.
  • the method is described by taking the example that the method is executed by a second node.
  • the second node may be a terminal device or a network device.
  • the second node receives indication information indicating the node type of the first node; and/or the second node receives first information, the first information is used to indicate model information of the first AI model, wherein , the first AI model is obtained based on the node type of the first node.
  • the second node when the second node receives the indication information indicating the node type of the first node, the second node clarifies the node type of the first node based on the indication information, and subsequently the second node can determine the node type of the first node based on the indication information.
  • the node type of the first node interacts with the first node.
  • the first AI model is obtained based on the node type of the first node.
  • the first node can perform different AI model processing processes based on different node types, solving the problem of a single node function and improving flexibility.
  • the first AI model is obtained based on the node type of the first node and the second AI model.
  • the first AI model received by the second node can be a model understandable by other nodes other than the first node, so as to facilitate other nodes.
  • the second node subsequently performs further model processing after receiving the target AI model.
  • the second AI model is obtained based on local data; or,
  • the second AI model is obtained based on K information, each information in the K information indicates the model information of the AI model of other nodes and the auxiliary information of the AI model of the other node, K is a positive integer; or,
  • the second AI model is obtained based on the local data and the K pieces of information.
  • the second AI model used to obtain the first AI model can be implemented as any of the above to improve the flexibility of solution implementation.
  • the node type of the first node includes any of the following: a node type for local training based on local data, a node type for fusion processing based on AI models of other nodes, The node type that performs local training on local data and performs fusion processing based on the AI models of other nodes.
  • the node type used to obtain the first AI model can be implemented as any of the above to improve the flexibility of solution implementation.
  • the node type of the second node includes any of the following: a node type that performs fusion processing based on the AI model of other nodes, a node type that performs local training based on the local data and performs fusion processing based on the AI model of other nodes. .
  • the first information is also used to indicate auxiliary information of the first AI model.
  • the first information received by the second node also indicates the auxiliary information of the first AI model.
  • the first information can also indicate the auxiliary information of the first AI model, so that the second The node can perform AI model processing (such as training, fusion, etc.) on the model information of the first AI model based on the auxiliary information of the first AI model, and improve the AI model obtained by processing the second node based on the first AI model. performance.
  • AI model processing such as training, fusion, etc.
  • the auxiliary information of the first AI model includes at least one of the following: type information of the first AI model, identification information of the first node, Receive the identification information of the node, the version information of the first AI model, generate the time information of the first AI model, generate the geographical location information of the first AI model, and generate the distribution information of local data of the first node.
  • the auxiliary information of the first AI model indicated by the first information includes at least one of the above items, so that the recipient of the first information can perform AI on the model information of the first AI model based on the at least one item of information above.
  • Model processing improves the performance of the AI model obtained by the receiver of the first information based on the first AI model.
  • the model information of the first AI model and the auxiliary information of the first AI model are respectively carried in different transmission resources, and the transmission resources include time domain resources and/or frequency domain. resource.
  • the two can be carried separately on on different transmission resources.
  • the first AI model is a model understandable by M nodes in the system where the first node is located, and M is an integer greater than or equal to 2.
  • the first AI model indicated by the first information received by the second node is a model understandable by M nodes in the system where the second node is located, so that there are multiple different models among the M nodes.
  • other nodes can understand the first AI model and further perform model processing after receiving the first information.
  • the method further includes: the second node determines the node type of the second node based on capability information and/or demand information; or, the first node receives an indication of the second node An indication of the node's node type.
  • the second node can determine the node type of the second node based on its own capability information and/or demand information, or can also determine the node type of the second node based on instructions from other nodes to improve the solution. Implement flexibility.
  • the seventh aspect of the present application provides an AI model processing device.
  • the device is a first node, or the device is a partial component of the first node (such as a processor, a chip or a chip system, etc.), or the device is also It can be a logic module or software that can realize all or part of the functions of the first node.
  • the communication device is described by taking an example of executing the communication device for a first node.
  • the first node may be a terminal device or a network device.
  • the device includes a processing unit and a transceiver unit; the processing unit is used to determine the first AI model; the transceiver unit is used to send first information, the first information indicates the model information of the first AI model and the Supplementary information.
  • the auxiliary information of the first AI model includes at least one of the following:
  • the type information of the first AI model, the identification information of the first node, the identification information of the receiving node of the first AI model, the version information of the first AI model, the time information of generating the first AI model, and generating the The geographical location information of the first AI model and the distribution information of local data of the first node.
  • the model information of the first AI model and the auxiliary information of the first AI model are respectively carried in different transmission resources, and the transmission resources include time domain resources and/or frequency domain. resource.
  • the first AI model is obtained based on the node type of the first node.
  • the first AI model is obtained based on the second AI model and the node type of the first node.
  • the second AI model is obtained based on local data; or, the second AI model is obtained based on K pieces of information, and each information in the K pieces of information indicates the AI of other nodes.
  • the model information of the model and the auxiliary information of the AI model of other nodes, K is a positive integer; or, the second AI model is obtained based on the local data and the K pieces of information.
  • the node type includes any of the following: a node type for local training based on local data, a node type for fusion processing based on AI models of other nodes, a node type for performing local training based on the local data.
  • the transceiver unit is further configured to send indication information indicating the node type of the first node.
  • the processing unit is further configured to determine the node type of the first node based on capability information and/or demand information; or, the transceiver unit is further configured to receive an indication of the first node Instructions for the node type.
  • the first AI model is a model understandable by M nodes in the system where the first node is located, and M is an integer greater than or equal to 2.
  • the component modules of the communication device can also be used to perform the steps performed in each possible implementation manner of the first aspect, and achieve corresponding technical effects.
  • the first aspect here No longer.
  • the eighth aspect of this application provides an AI model processing device.
  • the device is a second node, or the device is a partial component of the second node (such as a processor, a chip or a chip system, etc.), or the device is also It can be a logic module or software that can realize all or part of the functions of the second node.
  • the communication device is described using an example of executing the communication device for a second node.
  • the second node may be a terminal device or a network device.
  • the device includes a processing unit and a transceiver unit; the transceiver unit is used to receive N pieces of first information, each of the N pieces of first information indicating model information of a first AI model and assistance of the first AI model. Information, N is a positive integer; the processing unit is used to perform model processing based on the N first information to obtain the target AI model.
  • the target AI model is used to complete the AI task of the second node, or the target AI model is a local model of the second node.
  • the auxiliary information of the first AI model includes at least one of the following:
  • the model information of the first AI model and the auxiliary information of the first AI model are respectively carried in different transmission resources, and the transmission resources include time domain resources and/or frequency domain. resource.
  • the processing unit is specifically configured to perform model processing based on the N pieces of first information and the node type of the second node to obtain the target AI model.
  • the processing unit is specifically configured to pair The N first AI models are model fused to obtain the target AI model.
  • the processing unit is specifically configured to Model fusion is performed on the N first AI models and the second AI model based on the N pieces of first information to obtain the target AI model, where the second AI model is trained based on local data.
  • the transceiver unit is further configured to receive indication information indicating a node type of the first node.
  • the transceiver unit is further configured to send indication information indicating the node type of the second node.
  • the processing unit is further configured to determine the node type of the second node based on capability information and/or demand information; or, the transceiver unit is further configured to receive an indication of the second node Instructions for the node type.
  • the first AI model is a model understandable by M nodes in the system where the second node is located, and M is an integer greater than or equal to 2.
  • the component modules of the communication device can also be used to perform the steps performed in each possible implementation manner of the second aspect, and achieve corresponding technical effects.
  • the second aspect here No longer.
  • the ninth aspect of this application provides an AI model processing device.
  • the device is a first node, or the device is a part of the components in the first node (such as a processor, a chip or a chip system, etc.), or the device is also It can be a logic module or software that can realize all or part of the functions of the first node.
  • the communication device is described as being executed for a first node.
  • the first node may be a terminal device or a network device.
  • the device includes a processing unit and a transceiver unit; the processing unit is used to obtain a local AI model, and the local AI model is used to complete the AI task of the first node; the processing unit is also used to determine a public AI model based on the local AI model,
  • the public AI model is a model understandable by M nodes in the system where the first node is located, M is an integer greater than or equal to 2; the transceiver unit is used to send first information, and the first information is used to indicate the public AI Model information for the model.
  • the processing unit is specifically configured to determine a public AI model based on the local AI model and the node type of the first node.
  • the node type includes any of the following: a node type for local training based on local data, a node type for fusion processing based on AI models of other nodes, a node type for performing local training based on the local data.
  • the transceiver unit is further configured to send indication information indicating the node type of the first node.
  • the processing unit is further configured to determine the node type of the first node based on capability information and/or demand information; or, the transceiver unit is further configured to receive an indication of the first node Instructions for the node type.
  • the local AI model is obtained based on local data; or,
  • the local AI model is obtained based on K pieces of information.
  • Each piece of information in the K pieces of information indicates the model information of the AI model of other nodes and the auxiliary information of the AI model of the other nodes.
  • K is a positive integer; or,
  • the local AI model is obtained based on the local data and the K pieces of information.
  • the first information also indicates auxiliary information of the public AI model.
  • the auxiliary information of the public AI model includes at least one of the following:
  • the model information of the public AI model and the auxiliary information of the public AI model are respectively carried in different transmission resources, and the transmission resources include time domain resources and/or frequency domain resources.
  • the component modules of the communication device can also be used to perform the steps performed in each possible implementation manner of the third aspect, and achieve corresponding technical effects.
  • the third aspect here No longer.
  • the tenth aspect of this application provides an AI model processing device.
  • the device is a second node, or the device is some components of the second node (such as a processor, a chip or a chip system, etc.), or the device is also It can be a logic module or software that can realize all or part of the functions of the second node.
  • the communication device is described as being executed for a second node.
  • the second node may be a terminal device or a network device.
  • the device includes a processing unit and a transceiver unit; the transceiver unit is used to receive N first information, each of the N first information is used to indicate model information of a public AI model, and the public AI model is the Models understandable by M nodes in the system where the two nodes are located, M is an integer greater than or equal to 2; the processing unit is used to update the local AI model based on the N first information to obtain the updated local AI model, which The AI model is used to complete the AI task of the second node.
  • the processing unit is specifically configured to update the local AI model based on the N pieces of first information and the node type of the second node, to obtain an updated local AI model.
  • the node type includes any of the following: a node type for fusion processing based on the AI models of other nodes, local training based on the local data and training based on the AI models of other nodes.
  • the node type for fusion processing includes any of the following: a node type for fusion processing based on the AI models of other nodes, local training based on the local data and training based on the AI models of other nodes.
  • the node type for fusion processing includes any of the following: a node type for fusion processing based on the AI models of other nodes, local training based on the local data and training based on the AI models of other nodes.
  • the transceiver unit is further configured to send indication information indicating the node type of the second node.
  • the processing unit is further configured to determine the node type of the second node based on the capability information and/or the demand information; or, the transceiver unit is further configured to receive an instruction indicating the second node Instructions for the node type.
  • the local AI model is obtained based on P pieces of information, and each piece of information in the P pieces of information Indicates the model information of the AI model of other nodes and the auxiliary information of the AI model of the other node, and the P is a positive integer.
  • the local AI model is obtained based on the local data, or , the local AI model is obtained based on the local data and the P pieces of information.
  • Each piece of information in the P pieces of information indicates the model information of the AI model of other nodes and the auxiliary information of the AI model of the other nodes.
  • P is a positive integer. .
  • the first information is also used to indicate auxiliary information of the public AI model.
  • the auxiliary information of the public AI model includes at least one of the following: type information of the public AI model, identification information of the first node, and information of the receiving node of the public AI model. Identification information, version information of the public AI model, time information for generating the public AI model, geographical location information for generating the public AI model, and distribution information of local data of the first node.
  • the model information of the public AI model and the auxiliary information of the public AI model are respectively carried in different transmission resources, and the transmission resources include time domain resources and/or frequency domain resources.
  • the component modules of the communication device can also be used to perform the steps performed in each possible implementation manner of the fourth aspect, and achieve corresponding technical effects.
  • the fourth aspect here No longer.
  • the eleventh aspect of this application provides an AI model processing device.
  • the device is a first node, or the device is a partial component of the first node (such as a processor, a chip or a chip system, etc.), or the device It may also be a logic module or software capable of realizing all or part of the functions of the first node.
  • the communication device is described as an example when it is executed as a first node.
  • the first node may be a terminal device or a network device.
  • the device includes a processing unit; the processing unit is used by the first node to determine the node type of the first node.
  • the device further includes a transceiver unit configured to send indication information indicating a node type of the first node.
  • the device further includes a transceiver unit, the transceiver unit being configured to send first information, the first information being used to indicate model information of the first AI model, wherein, the The first AI model is obtained based on the node type of the first node.
  • the first AI model is obtained based on the second AI model and the node type of the first node.
  • the second AI model is obtained based on local data; or,
  • the second AI model is obtained based on K information, each information in the K information indicates the model information of the AI model of other nodes and the auxiliary information of the AI model of the other node, K is a positive integer; or,
  • the second AI model is obtained based on the local data and the K pieces of information.
  • the node type includes any of the following: a node type for local training based on local data, a node type for fusion processing based on AI models of other nodes, a node type based on the local data A node type that performs local training and performs fusion processing based on the AI models of other nodes.
  • the first information is also used to indicate auxiliary information of the first AI model
  • the auxiliary information of the first AI model includes at least one of the following: the first Type information of the AI model, identification information of the first node, identification information of the receiving node of the first AI model, version information of the first AI model, time information for generating the first AI model, and generating the first AI The geographical location information of the model and the distribution information of the local data of the first node.
  • the model information of the first AI model and the auxiliary information of the first AI model are respectively carried in different transmission resources, and the transmission resources include time domain resources and/or frequency resources. domain resources.
  • the first AI model is a model understandable by M nodes in the system where the first node is located, and M is an integer greater than or equal to 2.
  • the processing unit is further configured to determine the node type of the first node based on capability information and/or demand information; or, the transceiver unit is further configured to receive an indication of the first node. An indication of the node's node type.
  • the component modules of the communication device can also be used to perform the steps performed in each possible implementation manner of the fifth aspect, and achieve corresponding technical effects.
  • the fifth aspect please refer to the fifth aspect. No further details will be given.
  • a twelfth aspect of the present application provides an AI model processing device.
  • the device is a second node, or the device is some components of the second node (such as a processor, a chip or a chip system, etc.), or the device It can also be a logic module or software that can realize all or part of the functions of the second node.
  • the communication device is described as being executed for a second node.
  • the second node may be a terminal device or a network device.
  • the device includes a transceiver unit
  • the transceiver unit is used by the second node to receive indication information indicating the node type of the first node;
  • the transceiver unit is configured to receive first information, where the first information is used to indicate model information of the first AI model, where the first AI model is obtained based on the node type of the first node.
  • the first AI model is obtained based on the second AI model and the node type of the first node.
  • the second AI model is obtained based on local data; or,
  • the second AI model is obtained based on K information, each information in the K information indicates the model information of the AI model of other nodes and the auxiliary information of the AI model of the other node, K is a positive integer; or,
  • the second AI model is obtained based on the local data and the K pieces of information.
  • the node type includes any of the following: a node type for local training based on local data, a node type for fusion processing based on AI models of other nodes, a node type based on the local data A node type that performs local training and performs fusion processing based on the AI models of other nodes.
  • the first information is also used to indicate auxiliary information of the first AI model
  • the auxiliary information of the first AI model includes at least one of the following: the first Type information of the AI model, identification information of the first node, identification information of the receiving node of the first AI model, version information of the first AI model, time information for generating the first AI model, and generating the first AI
  • the first Type information of the AI model includes at least one of the following: the first Type information of the AI model, identification information of the first node, identification information of the receiving node of the first AI model, version information of the first AI model, time information for generating the first AI model, and generating the first AI
  • the model information of the first AI model and the auxiliary information of the first AI model are respectively carried in different transmission resources, and the transmission resources include time domain resources and/or frequency resources. domain resources.
  • the first AI model is a model understandable by M nodes in the system where the first node is located, and M is an integer greater than or equal to 2.
  • the device further includes a processing unit configured to determine the node type of the second node based on capability information and/or demand information; or, the transceiver unit also uses and receiving indication information indicating a node type of the second node.
  • the component modules of the communication device can also be used to perform the steps performed in each possible implementation manner of the sixth aspect, and achieve corresponding technical effects.
  • the sixth aspect please refer to the sixth aspect. No further details will be given.
  • a thirteenth aspect of the embodiment of the present application provides a communication device, including at least one processor, where the at least one processor is coupled to a memory;
  • This 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 foregoing first aspect or the method described in any possible implementation manner of the first aspect, or so that the device implements the foregoing second aspect or the method described in any possible implementation manner of the first aspect.
  • the method described in any possible implementation manner of the second aspect, or to enable the device to implement the aforementioned third aspect or the method described in any possible implementation manner of the third aspect, or to enable the device to implement the aforementioned third aspect The method described in the fourth aspect or any possible implementation manner of the fourth aspect, or to enable the device to implement the method described in the aforementioned fifth aspect or any possible implementation manner of the fifth aspect, or to enable the device to implement the method described in any possible implementation manner of the fifth aspect or the fifth aspect.
  • the device implements the method described in the foregoing sixth aspect or any possible implementation manner of the sixth aspect.
  • the fourteenth aspect of the embodiment of the present application provides a communication device, including at least one logic circuit and an input and output interface;
  • the logic circuit is used to perform the method described in the aforementioned first aspect or any possible implementation of the first aspect, or the logic circuit is used to perform the method described in the aforementioned second aspect or any possible implementation of the second aspect.
  • the method described in the above manner, or the logic circuit is used to perform the method described in the aforementioned third aspect or any possible implementation manner of the third aspect, or the logic circuit is used to perform the method described in the aforementioned fourth aspect or the third aspect.
  • the method described in any one possible implementation of the fourth aspect, or the logic circuit is used to perform the method described in the fifth aspect or any one possible implementation of the fifth aspect, or the logic circuit is used Execute the method described in the foregoing sixth aspect or any possible implementation manner of the sixth aspect.
  • the fifteenth aspect of the embodiment of the present application provides a computer-readable storage medium that stores one or more computer-executable instructions.
  • the processor executes the above-mentioned first aspect or any of the first aspects.
  • the processor performs the method described in the above second aspect or any possible implementation manner of the second aspect, or the processor performs the method described in the above third aspect or The method described in any possible implementation manner of the third aspect, or the processor executes the method described in the above fourth aspect or any possible implementation manner of the fourth aspect, or the processor executes the method described above The method described in the fifth aspect or any possible implementation manner of the fifth aspect, or the processor executes the method described in the above-mentioned sixth aspect or any possible implementation manner of the sixth aspect.
  • a sixteenth aspect of the embodiment of the present application provides a computer program product (or computer program) that stores one or more computers.
  • the processor executes the above first aspect or the first aspect.
  • the method of any possible implementation of the aspect, or the processor executes the above second aspect or the method of any possible implementation of the second aspect, or the processor executes the above third aspect or any one of the third aspect
  • a method of possible implementation, or the processor performs the fourth aspect or any of the possible implementation methods of the fourth aspect, or the processor performs the fifth aspect or any of the possible implementation methods of the fifth aspect.
  • method, or the processor executes the method of the sixth aspect or any of the possible implementation methods of the sixth aspect.
  • the seventeenth aspect of the embodiment of the present application provides a chip system.
  • the chip system includes at least one processor and is used to support the communication device to implement the functions involved in the above-mentioned first aspect or any possible implementation manner of the first aspect. , or, used to support the communication device to implement the functions involved in the above-mentioned second aspect or any one of the possible implementations of the second aspect, or, used to support the communication device to implement the above-mentioned third aspect or any one of the possible implementations of the third aspect.
  • the chip system may also include a memory for storing necessary program instructions and data of the first communication device.
  • the chip system may be composed of chips, or may include chips and other discrete devices.
  • the chip system further includes an interface circuit that provides program instructions and/or data to the at least one processor.
  • An eighteenth aspect of the embodiments 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, and/or the communication system includes the communication device of the ninth aspect and the communication device of the tenth aspect.
  • communication device, and/or the communication system includes the communication device of the eleventh aspect and the communication device of the twelfth aspect, and/or the communication system includes the communication device of the thirteenth aspect, and/or the The communication system includes the communication device of the fourteenth aspect described above.
  • the first node after the first node determines the first AI model, the first node sends first information indicating model information of the first AI model and auxiliary information of the first AI model.
  • the first information can also indicate the auxiliary information of the first AI model, so that the first The recipient of the information can perform AI model processing (such as training, fusion, etc.) on the model information of the first AI model based on the auxiliary information of the first AI model, and the recipient of the improved information can perform AI model processing based on the first AI model. Performance of the resulting AI model.
  • the first node obtains a local AI model used to complete the AI task of the first node, the first node determines a public AI model based on the local AI model, and the first node sends a message indicating the public AI model.
  • the first information of the model information of the AI model is a model understandable by M nodes in the system where the first node is located.
  • the public AI model indicated by the first information sent by the first node is a model understandable by M nodes in the system where the first node is located, so as to facilitate the situation where there are multiple different node types among the M nodes. Under this condition, other nodes can understand the first AI model and further perform model processing after receiving the first information.
  • the first node may determine the node type of the first node, and subsequently the first node may perform AI model processing based on the node type. This allows the first node to perform different AI model processing processes based on different node types, solving the problem of a single node function and improving flexibility.
  • Figure 1a is a schematic diagram of the communication system provided by this application.
  • FIG 1b is another schematic diagram of the communication system provided by this application.
  • Figure 1c is another schematic diagram of the communication system provided by this application.
  • FIG. 2a is a schematic diagram of the AI processing process involved in this application.
  • FIG. 2b is another schematic diagram of the AI processing process involved in this application.
  • FIG. 2c is another schematic diagram of the AI processing process involved in this application.
  • FIG. 2d is another schematic diagram of the AI processing process involved in this application.
  • FIG. 2e is another schematic diagram of the AI processing process involved in this application.
  • Figure 3a is a schematic diagram of the AI processing process based on federated learning
  • Figure 3b is a schematic diagram of the AI processing process based on distributed learning
  • Figure 4 is an interactive schematic diagram of the AI model processing method provided by this application.
  • Figure 5a is a schematic diagram of the AI model processing method provided by this application.
  • Figure 5b is another schematic diagram of the AI model processing method provided by this application.
  • Figure 5c is another schematic diagram of the AI model processing method provided by this application.
  • Figure 5d is another schematic diagram of the AI model processing method provided by this application.
  • Figure 6 is another interactive schematic diagram of the AI model processing method provided by this application.
  • Figure 7 is another interactive schematic diagram of the AI model processing method provided by this application.
  • FIG. 8 is a schematic diagram of the communication device provided by this application.
  • FIG. 9 is another schematic diagram of the communication device provided by this application.
  • FIG. 10 is another schematic diagram of the communication device provided by this application.
  • FIG 11 is another schematic diagram of the communication device provided by this application.
  • 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 users, or a handheld device with wireless connection function, or Other processing equipment connected to the wireless modem.
  • the terminal device can communicate with one or more core networks or the Internet via the RAN.
  • the terminal device can be a mobile terminal device, such as a mobile phone (also known as a "cellular" phone), a computer and a data card, such as , which may be portable, pocket-sized, handheld, computer-built-in, or vehicle-mounted mobile devices that exchange voice and/or data with the wireless access network.
  • a mobile terminal device such as a mobile phone (also known as a "cellular" phone), a computer and a data card, such as , which may be portable, pocket-sized, handheld, computer-built-in, or vehicle-mounted mobile devices that exchange voice and/or data with the wireless access network.
  • PCS personal communication service
  • SIP Session Initiation Protocol
  • WLL wireless local loop
  • PDA personal digital assistants
  • Tablets tablets Computers
  • Wireless terminal equipment can also be called a system, subscriber unit, subscriber station, mobile station, mobile station (MS), remote station, access point ( access point, AP), remote terminal equipment (remote terminal), access terminal equipment (access terminal), user terminal equipment (user terminal), user agent (user agent), subscriber station (subscriber station, SS), client equipment (customer premises equipment, CPE), terminal (terminal), user equipment (user equipment, UE), mobile terminal (mobile terminal, MT), etc.
  • a system subscriber unit, subscriber station, mobile station, mobile station (MS), remote station, access point ( access point, AP), remote terminal equipment (remote terminal), access terminal equipment (access terminal), user terminal equipment (user terminal), user agent (user agent), subscriber station (subscriber station, SS), client equipment (customer premises equipment, CPE), terminal (terminal), user equipment (user equipment, UE), mobile terminal (mobile terminal, MT), etc.
  • the terminal device may also be a wearable device.
  • Wearable devices can also be called wearable smart devices or smart wearable devices. It is a general term for applying wearable technology to intelligently design daily wear and develop wearable devices, such as glasses, gloves, watches, clothing and shoes. wait.
  • a wearable device is a portable device that is worn directly on the body or integrated into the user's clothing or accessories. Wearable devices are not just hardware devices, but also achieve powerful functions through software support, data interaction, and cloud interaction.
  • wearable smart devices include full-featured, large-sized devices that can achieve complete or partial functions without relying on smartphones, such as smart watches or smart glasses, and those that only focus on a certain type of application function and need to cooperate with other devices such as smartphones. Used, such as various smart bracelets, smart helmets, smart jewelry, etc. for physical sign monitoring.
  • the terminal can also be a drone, a robot, a terminal in device-to-device (D2D) communication, a terminal in vehicle to everything (V2X), or a virtual reality (VR) terminal.
  • equipment augmented reality (AR) terminal equipment, wireless terminals in industrial control, wireless terminals in self-driving, wireless terminals in remote medical, smart grids ( Wireless terminals in smart grid, wireless terminals in transportation safety, wireless terminals in smart city, wireless terminals in smart home, etc.
  • the terminal device may also be a terminal device in a communication system evolved after the fifth generation (5th generation, 5G) communication system (such as a sixth generation (6th generation, 6G) communication system, etc.) or a public land mobile network that will evolve in the future.
  • Terminal equipment in public land mobile network, PLMN
  • 6G networks can further expand the form and function of 5G communication terminals.
  • 6G terminals include but are not limited to cars, cellular network terminals (integrated with satellite terminal functions), drones, and Internet of Things (IoT) devices.
  • the above-mentioned terminal device can also obtain AI services provided by network devices.
  • the terminal device may also have AI processing capabilities.
  • Network device It may be a device in a wireless network.
  • the network device may be a RAN node (or device) that connects the terminal device to the wireless network, and may also be called a base station.
  • some examples of RAN equipment are: base station gNB (gNodeB), transmission reception point (TRP), evolved node B (evolved Node B, eNB), radio network controller (radio network) in the 5G communication system controller, RNC), Node B (Node B, NB), home base station (e.g., home evolved Node B, or home Node B, HNB), base band unit (base band unit, BBU), or wireless fidelity, Wi-Fi) access point AP, etc.
  • the network device may include a centralized unit (CU) node, a distributed unit (DU) node, or a RAN device including a CU node and a DU node.
  • CU centralized unit
  • DU distributed unit
  • RAN device including a CU node and
  • the network device may be other devices that provide wireless communication functions for terminal devices.
  • the embodiments of this application do not limit the specific technology and specific equipment form used by the network equipment. For convenience of description, the embodiments of this application are not limited.
  • the network equipment may also include core network equipment.
  • the core network equipment may include, for example, a mobility management entity (MME), a home subscriber server (HSS), and a service in a fourth generation (4G) network.
  • MME mobility management entity
  • HSS home subscriber server
  • 4G fourth generation
  • Gateway serving gateway, S-GW
  • policy and charging rules function PCRF
  • PDN gateway public data network gateway
  • P-GW access and charging in 5G networks
  • Network elements such as access and mobility management function (AMF), user plane function (UPF) or session management function (SMF).
  • the core network equipment may also include other core network equipment in the 5G network and the next generation network of the 5G network.
  • the above-mentioned network device may also be a network node with AI capabilities, and may provide AI services for terminals or other network devices.
  • it may be an AI node or a computing power node on the network side (access network or core network).
  • RAN nodes with AI capabilities may be an AI node or a computing power node on the network side (access network or core network).
  • core network elements with AI capabilities may be an AI node or a computing power node on the network side (access network or core network).
  • the device used to implement the function of the network device may be a network device, or may be a device that can support the network device to implement the function, such as a chip system, and the device may be installed in the network device.
  • the technical solution provided by the embodiment of the present application the technical solution provided by the embodiment of the present application is described by taking the device for realizing the functions of the network device being a network device as an example.
  • Configuration and preconfiguration In this application, configuration and preconfiguration will be used at the same time.
  • Configuration means that the network device/server sends the configuration information or parameter values of some parameters to the terminal through messages or signaling, so that the terminal can determine communication parameters or transmission resources based on these values or information.
  • Preconfiguration is similar to configuration. It can be parameter information or parameter values that the network device/server has negotiated with the terminal device in advance, or it can be parameter information or parameter values adopted by the base station/network device or terminal device specified in the standard protocol, or it can be Parameter information or parameter values stored in the base station/server or terminal equipment in advance. This application does not limit this.
  • system and “network” in the embodiments of this application can be used interchangeably.
  • "Plural” means two or more.
  • “And/or” describes the relationship between associated objects, indicating that there can be three relationships, for example, A and/or B, which can mean: A alone exists, A and B exist simultaneously, and B alone exists, where A, B can be singular or plural.
  • the character “/” generally indicates that the related objects are in an "or” relationship.
  • “At least one of the following” or similar expressions thereof refers to any combination of these items, including any combination of a single item (items) or a plurality of items (items).
  • At least one of A, B, and C includes A, B, C, AB, AC, BC, or ABC.
  • the ordinal numbers such as “first” and “second” mentioned in the embodiments of this application are used to distinguish multiple objects and are not used to limit the order, timing, priority or importance of multiple objects. degree.
  • the communication system includes at least one network device and/or at least one terminal device.
  • FIG. 1a is a schematic diagram of the communication system in this application.
  • a network device 101 and six terminal devices are shown as an example.
  • the six terminal devices are terminal device 1, terminal device 2, terminal device 3, terminal device 4, terminal device 5, terminal device 6, etc. .
  • 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
  • terminal device 6 is The printer is used as an example.
  • the AI configuration information sending entity may be a network device.
  • the AI configuration information receiving entity can be terminal device 1-terminal device 6.
  • the network device and terminal device 1-terminal device 6 form a communication system.
  • terminal device 1-terminal device 6 can send data to Network equipment needs to receive data sent by terminal equipment 1-terminal equipment 6.
  • the network device can send configuration information to terminal device 1-terminal device 6.
  • the AI configuration information may include indication information of the node type mentioned later; the data may include model information of the AI model mentioned later and/or auxiliary information of the AI model.
  • terminal equipment 4-terminal equipment 6 can also form a communication system.
  • terminal device 5 serves as a network device, that is, the AI configuration information sending entity
  • terminal device 4 and terminal device 6 serve as terminal devices, that is, the AI configuration information receiving entity.
  • terminal device 5 sends AI configuration information to terminal device 4 and terminal device 6 respectively, and receives data sent by terminal device 4 and terminal device 6; correspondingly, terminal device 4 and terminal device 6 receive terminal device 5 Send the AI configuration information and send data to the terminal device 5.
  • AI-related business for example, as shown in Figure 1b, taking a network device as a base station, 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.
  • Related Business taking the terminal device including a TV and a mobile phone as an example, communication-related services and AI-related services can also be executed between the TV and the mobile phone.
  • MLP multilayer perceptron
  • an MLP consists of an input layer (left), an output layer (right), and multiple hidden layers (middle).
  • each layer of MLP contains several nodes, called neurons.
  • neurons in two adjacent layers are connected in pairs.
  • the output h of the neuron in the next layer is the weighted sum of all the neurons x in the previous layer connected to it and passes through the activation function, which can be expressed as:
  • w is the weight matrix
  • b is the bias vector
  • f is the activation function
  • the output of the neural network can be expressed recursively as:
  • a neural network can be understood as a mapping relationship from an input data set to an output data set.
  • neural networks are initialized randomly, and the process of using existing data to obtain this mapping relationship from random w and b is called neural network training.
  • the specific method of training is to use a loss function to evaluate the output results of the neural network.
  • the error can be back-propagated, and the neural network parameters (including w and b) can be iteratively optimized through the gradient descent method until the loss function reaches the minimum value, which is the "optimal point" in Figure 2b.
  • the neural network parameters corresponding to the "optimal point" in Figure 2b can be used as neural network parameters in the trained AI model information.
  • the gradient descent process can be expressed as:
  • ⁇ is the parameter to be optimized (including w and b)
  • L is the loss function
  • eta is the learning rate, which controls the step size of gradient descent
  • the backpropagation process utilizes the chain rule of partial derivatives.
  • the gradient of the previous layer parameters can be calculated recursively from the gradient of the subsequent layer parameters, which can be expressed as:
  • w ij is the weight of node j connecting node i
  • s i is the weighted sum of inputs on node i.
  • GAN Generative Adversarial Network
  • Generative models are a rapidly developing research direction in the field of computer vision in recent years.
  • Ian Goodfellow proposed a generative model based on probability distribution for generating simulated data, named Generative adversarial network (GAN). Because of their ability to approximate complex probability density functions, generative adversarial networks have proven useful in a variety of machine learning tasks including image generation, video generation, and natural language processing.
  • a generative adversarial network usually consists of at least one generator and at least one discriminator.
  • the generator takes random samples from a potential space as input, outputs false samples after processing, and requires the output false samples to be as realistic as possible to the real samples in the training set.
  • the input of the discriminator is a real sample or a false sample of the generator, and it outputs a probability value that the input sample belongs to the real sample.
  • the discriminator aims to distinguish the false samples generated by the generator from the real samples, while the generator needs to deceive the discriminator as much as possible so that it cannot distinguish the authenticity of the input samples. By constantly confronting each other and adjusting their respective model parameters, the generator and the discriminator finally reach the point where the discriminator cannot judge whether the output samples of the generator are real samples.
  • the autoencoder is an artificial neural network structure, and its schematic diagram is shown in Figure 2e. It is generally used to represent a set of data (also called encoding), usually for dimensionality reduction, and contains encoder and decoder parts.
  • the encoder encodes the input data through one or more layers of neural networks to obtain dimensionally reduced codewords.
  • the decoder reconstructs the code words into output data through one or more layers of neural networks.
  • the output data needs to be as identical as possible to the encoder input data.
  • the encoder and decoder are trained simultaneously to achieve the above requirements.
  • AI technology has made significant progress in fields such as machine vision and natural language processing, and has gradually become popular in real life. It is foreseeable that AI will be ubiquitous in various connected devices (such as terminals and edges).
  • communication systems can become platforms for large-scale machine learning and AI services.
  • the terminal device can not only enjoy AI inference services or AI model training services from the network, but also participate in the data collection required for network model training, and even participate in distributed model training.
  • FL federated learning
  • the concept of FL effectively solves the dilemma faced by the current development of artificial intelligence.
  • the FL architecture is the most extensive training architecture in the current FL field.
  • the FedAvg algorithm is the basic algorithm of FL. Taking the central node as the central end and the distributed nodes as the clients as an example, the algorithm process is roughly as follows:
  • the central terminal initializes the model to be trained and broadcast it to all client devices.
  • the client k ⁇ [1,K] is based on the local data set to the received global model Perform training for E epochs to obtain local training results Report it to the central node.
  • the central node aggregates and collects local training results from all (or part) clients. Assume that the set of clients uploading local models in round t is The central end will use the number of samples corresponding to the client as the weight to perform weighted averaging to obtain a new global model. The specific update rule is: Afterwards, the central end will transfer the latest version of the global model to Broadcasts are sent to all client devices for a new round of training.
  • convergence generally means that the performance of the trained model meets the preset requirements. For example, using FL to train a model for image classification tasks, the preset classification accuracy is 95%, then as the training progresses, Evaluate the accuracy performance of the model through the test set. When it is found that the accuracy reaches 95% or above, you can confirm that the model has converged and stop training. At the same time, due to poor model structure design, parameter selection, training methods and other factors, it may never be possible to preset performance requirements. In this case, it is necessary to consider setting an upper limit for the number of training rounds. When the upper limit is reached, even if the model If the preset performance is not reached, training should be stopped.
  • the trained local gradients can also be After reporting, the central node averages the local gradient and updates the global model based on the direction of this average gradient.
  • the data set exists at the distributed node, that is, the distributed node collects the local data set, conducts local training, and reports the local results (model or gradient) obtained by the training to the central node.
  • the central node itself does not have a data set, and is only responsible for fusion processing of the training results of distributed nodes, obtaining the global model, and issuing it to distributed nodes.
  • the decentralized AI data processing model is an improvement on the federated learning AI data processing model. This model does not require the establishment of a central node and can improve the robustness of the system.
  • each node uses local data and local targets to calculate a local AI model, it interacts with its neighbor nodes that have reachable communication with their respective AI models, and further processes (such as training, fusion, etc.) based on the interactive AI model locally.
  • AI model After each node uses local data and local targets to calculate a local AI model, it interacts with its neighbor nodes that have reachable communication with their respective AI models, and further processes (such as training, fusion, etc.) based on the interactive AI model locally.
  • AI model is an improvement on the federated learning AI data processing model.
  • the decentralized learning process in the decentralized AI data processing model is shown in Figure 3b, which is a completely distributed system without a central node.
  • the design goal f(x) of the system is generally the mean value of the goal f i (x) of each node, that is Where n is the number of distributed nodes, and x is the parameter to be optimized.
  • x is the parameter of the machine learning (such as neural network) model.
  • Each node uses local data and local target f i (x) to calculate the local gradient. It is then sent to reachable neighbor nodes. After any node receives the gradient information from its neighbors, it can update the parameters x of the local AI model according to the following formula:
  • N i is the set of neighbor nodes of node i
  • N i represents the number of elements in the set of neighbor nodes of node i, that is, the number of neighbor nodes of node i
  • the superscript k represents the kth (k is a positive integer) round of training.
  • ⁇ k is the training step size used in the k-th round of training.
  • Figure 4 is an interactive schematic diagram of the AI model processing method provided in this application.
  • the method includes the following steps.
  • the first node determines the first AI model.
  • the first node sends first information, which is used to indicate model information of the first AI model and auxiliary information of the first AI model.
  • the first node sends the first information in step S402.
  • the second node receives N pieces of first information sent from N first nodes in step S402, where N is a positive integer.
  • the first information is used to indicate model information of the first AI model and auxiliary information of the first AI model.
  • model information of the first AI model is used to construct the first AI model.
  • the model information may include at least one item such as parameter information of the model and structural information of the model.
  • the structural information of the model may include at least one item such as the number of model layers, the number of neurons in each layer of the model, and the connection relationship between layers in the model.
  • the first AI model determined by the first node in step S401 is a model understandable by M nodes in the system where the first node is located, where M is an integer greater than or equal to 2.
  • the first AI model indicated by the first information sent by the first node in step S402 is a model understandable by M nodes in the system where the first node is located, so that there are multiple In the case of different node types, other nodes can understand the first AI model and further perform model processing after receiving the first information.
  • the first AI model determined by the first node in step S401 may not be a model understandable by some or all of the M nodes in the system where the first node is located.
  • some or all of the M nodes can send the model information of the first AI model to other nodes (such as servers, controllers, network devices, centers node, etc.), so that other nodes perform model processing based on the model information of the first AI model to obtain and send understandable model information to some or all of the M nodes, so that among the M nodes Some or all of the nodes perform subsequent model processing based on the model information of the understandable model.
  • the first AI model determined by the first node in step S401 may not be a model understandable by some or all of the M nodes in the system where the first node is located.
  • some or all of the M nodes may discard the incomprehensible model.
  • the first AI model is a model that can be understood by M nodes in the system where the first node is located. It can be expressed as the first AI model is a model common to M nodes in the system where the first node is located.
  • the first AI model The model structure of the model is a model structure understandable by M nodes in the system where the first node is located.
  • the model structure of the first AI model is a model structure common to M nodes in the system where the first node is located.
  • the first AI model It is a public model of the system where the first node is located, or the model structure of the first AI model is a public model structure of the system where the first node is located.
  • the AI model processed locally by each of the M nodes in the system where the first node is located may not be accessible to other nodes.
  • the model of understanding can be called a local AI model. The following is an exemplary description of the acquisition process of the public AI model and the local AI model for any node among the M nodes.
  • the local AI model of each node is used for the learning task of this node.
  • the local AI model structure and parameter amounts of different nodes may be different.
  • each of the M nodes in the system where the first node is located can select an appropriate local AI based on capability information (i.e., the device capabilities of the device itself) and/or demand information (i.e., the requirements of the local task). Model.
  • the capability information ie, the device capability of the device itself
  • the computing power is generally determined by the computing module in the device.
  • the computing module can include a central processing unit (central processing unit, CPU), a graphics processing unit (graphics processing unit, GPU), a neural processing unit At least one item such as neural processing unit (NPU) or tensor processing unit (TPU). For example, in terms of floating-point operations per second (flops), it describes how quickly the device can perform model training and inference.
  • the storage capacity is generally determined by the storage module (such as video memory/cache/memory, etc.) in the device, and is measured in Bytes (or KB/MB/GB based on bytes). ), which describes how large a model the device can hold.
  • the storage module such as video memory/cache/memory, etc.
  • Bytes or KB/MB/GB based on bytes.
  • demand information (that is, the requirements of local tasks) generally refers to the performance requirements of nodes when using a certain model to complete specific local tasks, such as positioning accuracy in positioning tasks, classification accuracy in classification tasks, and information reconstruction tasks. Mean square error, compression ratio in compression tasks, etc.
  • each node among the M nodes in the system where the first node is located configures a local AI model, it can select the model that meets the local task requirements and has the smallest computing power and storage capacity requirements as the local AI model. For example, when multiple When all models can meet the local task requirements, select the model with the smallest number of model parameters and the smallest amount of model training/inference calculations; you can also select the model with the best performance among the models whose computing power and storage capacity requirements are less than or equal to the node computing power and storage capacity. Good models serve as local AI models.
  • each of the M nodes in the system where the first node is located can also configure a local AI model in a pre-configured manner.
  • the system or standard can provide a set of optional model libraries for a given task. The respective node selects a local AI model from it.
  • the configuration of the local AI model can be performed by the management node in the system where the first node is located, and then distributed or installed to each node, or it can be performed by each node itself. If it is the former, each node can also report its own computing capabilities, storage capabilities, and task requirements to the management node.
  • the public AI model is used for knowledge transfer between nodes.
  • Different nodes of the system use public AI models with the same structure and parameter amounts.
  • This isomorphic public AI model design is conducive to simplifying the mechanism design of model transmission between nodes (for the implementation of model transmission, please refer to the aforementioned Figure 3b and related descriptions).
  • public AI models can be implemented in a configured or pre-configured manner for other nodes, for example, pre-given by management nodes or standards, with different learning tasks (such as channel state information (CSI) compression, beam management, Positioning, etc.) can use different public AI models.
  • CSI channel state information
  • the configuration of the public AI model needs to consider the computing power and/or storage capacity of each node in the system where the first node is located. For example, when the management node determines the public AI model, the management node can collect the computing power and storage capacity information of each node, and determine the public AI model based on the computing power and storage capacity of the node with the weakest capabilities. For another example, when a public AI model is given in advance through preconfiguration, the standard should also specify the computing power requirements and storage capacity requirements corresponding to the public AI model. Before each node joins the system, it should ensure that its own corresponding capabilities are not sufficient. Less than the capacity requirements given by the standard.
  • the local AI model configuration and the public AI model configuration are generally performed at the beginning of the system construction.
  • the system status such as the system environment, the capabilities and needs of each node
  • Local AI models and public AI models are configured or reset.
  • the above-mentioned local AI model and public AI model can be oriented to the same function.
  • the functions of the local AI model and the public AI model can be the same, for example, both are used for tasks such as positioning, classification, reconstruction, and compression. At this time, the only differences between the two are in model structure, parameter quantity, and performance.
  • the above-mentioned local AI model and public AI model can be oriented to different functions: that is, the functions of the local AI model and the public AI model can be different.
  • each node generates data based on the GAN structure.
  • the local AI model can be the generator model in the GAN structure
  • the public AI model can be the discriminator model in the GAN structure, that is, each node generates data based on local data and neighbors.
  • the discriminator model is trained to update the local generator model and discriminator model, and the updated discriminator model is sent to the neighbors.
  • the discriminator model contains local data-related information of each node, and the interactive discriminator model realizes the interaction of knowledge between nodes.
  • the local AI model and the public AI model are two parts of a completed model.
  • the local AI model is the encoder part of the autoencoder model
  • the public AI model is the decoder part of the autoencoder model.
  • Each node can use local data to train different encoders (local AI models) plus the same decoder (public AI model), and then send the decoder (public AI model) to neighboring nodes.
  • the decoder part also carries the local data information of each node, thus realizing the interaction of knowledge between nodes.
  • knowledge sharing refers to obtaining (or updating) public AI models based on local AI models.
  • knowledge sharing can be achieved through the existing technologies of knowledge distillation, pruning, and expansion; when the local AI model and the public AI model are oriented to different functions, then the local AI model can be used to share knowledge. Joint training with public AI models enables knowledge sharing.
  • knowledge absorption refers to updating the local AI model based on the public AI model.
  • knowledge absorption can be achieved through existing technologies of knowledge distillation, pruning, and expansion; when the local AI model and the public AI model are oriented to different functions, then through the local Joint training of AI models and public AI models enables knowledge absorption.
  • the first AI model determined by the first node in step S401 is obtained based on the node type of the first node.
  • each node uses the same learning model (all parameters are Average method, and the functions of each node are the same.
  • the functional distinction of nodes is not considered in the above system.
  • the first AI model indicated by the first information sent by the first node may be a model obtained by the first node based on the node type of the first node.
  • the first node can perform different AI model processing processes based on different node types, solving the problem of single node function to improve flexibility. .
  • the first AI model determined by the first node in step S401 is obtained based on the node type of the first node and the second AI model.
  • the first AI model indicated by the first information sent by the first node may be a model understandable by other nodes, so that other nodes can After receiving the first AI model, further model processing is performed.
  • the node type includes any of the following: a node type for local training based on local data (for convenience of subsequent description, denoted as L node), a node type for fusion processing based on AI models of other nodes (for convenience of subsequent description) (described later, denoted as A node), the type of node that performs local training based on the local data and performs fusion processing based on the AI model of other nodes (denoted as H node for the convenience of subsequent description).
  • the node type used to obtain the first AI model can be implemented as any of the above to improve the flexibility of solution implementation.
  • the method further includes: the first node sending indication information indicating the node type of the first node.
  • the first node may also send indication information indicating the node type of the first node, so that other nodes can clarify the node type of the first node based on the indication information.
  • the node type of the first node may be determined based on the node type and the node type of the first node. The first node to interact with.
  • the first node determines the node type of the first node based on capability information and/or demand information; or, the first node receives an instruction indicating that the first node Instructions for the node type.
  • the first node can determine the node type of the first node based on its own capability information and/or demand information, or can also determine the node type of the first node based on instructions from other nodes to improve the implementation of the solution. flexibility.
  • the L node For the L node, it can be called a local training node. This type of node has the ability and/or demand for knowledge sharing but no ability and/or demand for knowledge absorption. At the same time, there is local data on the node and can be used for local training.
  • a node it can be called a model fusion node.
  • This type of node does not have the ability/need for knowledge sharing, but has the ability and/or demand for knowledge absorption. At the same time, the node may not have local data for local training.
  • H nodes they can be called mixed-function nodes.
  • This type of node has both the ability/need for knowledge sharing and the ability and/or demand for knowledge absorption.
  • there is local data on the node which can be used for local training.
  • the type of node can be determined by the node according to its own capabilities, needs and local data conditions, or can be specified by the management node, or can be pre-configured. Sure.
  • the node type after the node type is determined, it needs to inform the neighboring nodes of its own node type (ie, role information), that is, role interaction. At the same time, the node can also inform its neighbors of the learning task type information and node type that the node participates in. For example, after each node receives the node type sent from the neighbor, the implementation of constructing the neighbor information table can be as shown in Table 1 below.
  • the node ID is the unique identifier of the neighbor node in the decentralized learning system, which can be uniformly assigned when the neighbor node joins the learning system; the neighbor role is assigned by the neighbor node.
  • the type of node sent is determined; the task type indicates the type of learning task that the neighbor participates in, such as CSI compression, beam management, positioning, etc.
  • the task type can also be determined by the type of node sent by the neighbor; the link quality indicator describes The quality of the communication link between a given neighbor and this node can include indicators such as the signal-to-noise ratio (SNR), throughput, delay, and packet loss rate of the communication link.
  • SNR signal-to-noise ratio
  • the role of each node is not static.
  • the node role may change when the communication network topology, node requirements, environmental status, etc. change. As shown in Figure 5b, some examples of node role changes are given.
  • the H node cannot obtain new local data, or based on privacy or other considerations, the knowledge sharing function is turned off and transformed into an A node.
  • the H node determines that local data is sufficient to train and obtain a local AI model with better performance, turns off the knowledge absorption function, and transforms into an L node.
  • Another example is that the L node cannot obtain new local data, so it turns off the local training function, turns on the knowledge absorption function, and transforms into an A node.
  • the L node determines that a local AI model with satisfactory performance cannot be trained based on local data, turns on the knowledge absorption function, and transforms into an H node.
  • a node can obtain local data, enable the local training function, and transform into an H node.
  • a node can obtain local data, and it is judged that the local data is sufficient to train and obtain a local AI model with better performance.
  • the knowledge absorption function is turned off and transformed into an L node.
  • the neighbors can be notified of the change so that the neighbors can update the neighbor information tables they maintain.
  • each node after completing the construction of the neighbor information table, each node starts to run relevant operations according to its own role and perform distributed learning.
  • the learning process proceeds cyclically, and each cycle is divided into two stages: model update and model interaction.
  • Table 2 below gives the specific operations of different nodes in each stage.
  • the local training update of the model can be implemented using gradient backpropagation training similar to that used in the fully connected neural network described in Figure 2a, or can also be implemented through other AI training processes, which are not limited here.
  • the method of obtaining the public AI model based on the local AI model and updating the local AI model based on the public AI model can refer to the introduction in the aforementioned Figure 5a and related descriptions, and will not be described again here.
  • the second AI model when the first AI model indicated by the first information sent by the first node is obtained based on the second AI model, the second AI model may be obtained by the first node based on local data; or , the second AI model may be obtained by the first node based on K pieces of information.
  • Each of the K pieces of information indicates the model information of the AI model of other nodes and the auxiliary information of the AI model of the other nodes.
  • K is a positive integer.
  • the second AI model may be obtained by the first node based on the local data and the K pieces of information.
  • the second AI model used to obtain the first AI model can be implemented by any of the above to improve the flexibility of solution implementation.
  • the value of K is 1 less than the value of M, or the value of K is less than M-1.
  • the auxiliary information of the first AI model includes at least one of the following: type information of the first AI model, identification information of the first node, and identification of the receiving node of the first AI model.
  • the auxiliary information of the first AI model indicated by the first information includes at least one of the above items, so that the recipient of the first information can perform AI model processing on the model information of the first AI model based on the at least one item of information above. , improving the performance of the AI model obtained by the receiver of the first information based on the first AI model.
  • the model information of the first AI model and the auxiliary information of the first AI model are respectively carried in different transmission resources, and the transmission resources include time domain resources and/or frequency domain resources.
  • the transmission resources include time domain resources and/or frequency domain resources.
  • the transmission resources are preconfigured resources.
  • M nodes in the system where the first node is located send the updated public AI model of this node to neighboring nodes (for example, the first information sent by the first node to the second node in step S402 carries There is model information for the first AI model). Since the structure and parameters of the public AI model are preset and unified, a communication mechanism with lower complexity can be designed. The implementation process of the first information sent by the first node in step S402 will be described below with more implementation examples.
  • the first AI model indicated by the first information may be a public AI model.
  • the first node may package the first information into a data packet in the manner shown in Figure 5c, in which the model information of the public AI model is carried in the payload, and the auxiliary information of the public AI model is carried in the header part.
  • the header part includes at least one piece of information such as payload type, node ID, version number, timestamp, geographical location, and data distribution.
  • the payload type is used to indicate the type of public AI model information carried in the payload part.
  • This type may be model parameters, model gradients, (intermediate) inference results, etc.
  • the bit mapping table can be designed to implicitly indicate this information, for example Use 00 to represent model parameters, 11 to represent model gradients, and 01 or 10 to represent (intermediate) inference results.
  • the node ID can include a source node ID and a destination node ID.
  • the node ID is the same as the node ID used in the neighbor information table.
  • the source node ID is used to identify the node that generates the public AI model information contained in the payload.
  • the destination node The ID is used to identify the destination node of the public AI model information contained in the payload.
  • the version number is used to indicate the version of the public AI model included in the payload.
  • the design of the version number will be discussed in detail in Embodiment 2.
  • the timestamp is the time point when the update of the public AI model contained in the payload is completed.
  • the geographical location identifies the geographical location information of the node when the public AI model contained in the payload is updated.
  • the data distribution is the distribution information of the local data of the source node.
  • this field can be empty (or filled with 0).
  • Information such as version number, timestamp, geographical location, data distribution, etc. will be used for model fusion, so it is collectively referred to as model fusion auxiliary information.
  • the first node may send the data packet to the adjacent node in step S402 after completing the data packet packaging in the above manner.
  • the communication link between the first node and the second node is a sidelink, that is, the sidelink transmits data packets.
  • the header and payload part of the data packet can be sent separately.
  • the header is sent on the control channel and the payload is sent on the data channel.
  • the control channel where the header is located also needs to send instructions to send the payload corresponding to the header.
  • the location of the transmission resources (time-frequency and space resources) of the data channel used. It should be noted that since the structure and parameter amount of the public AI model are preset and unified, the amount of transmission resources used by it is also predictable and fixed. Therefore, resources dedicated to the transmission of the public AI model can be allocated in advance, thus The design of the communication mechanism is simplified (that is, there is no need to determine transmission resource allocation and scheduling strategies based on different data packet lengths).
  • the second node performs model processing based on the N pieces of first information to obtain the target AI model.
  • the second node after the second node receives N pieces of first information in step S402, the second node performs model processing based on the N pieces of first information in step S403 to obtain the target AI model.
  • the target AI model is used to complete the AI task of the second node, or the target AI model is a local model of the second node.
  • the second node performs model processing based on the N pieces of first information
  • obtaining the target AI model includes: the second node performs model processing based on the N pieces of first information and the node type of the second node.
  • Model processing to obtain the target AI model Specifically, when the second AI model is obtained based on at least the node type of the second node, the second node can perform different AI model processing processes based on different node types to solve the problem of single node function and improve flexibility. sex.
  • the value of K is 1 less than the value of M, or the value of K is less than M-1.
  • step S403 after the second node receives the data packet containing the public AI model information sent by N neighboring nodes, it can unpack it and obtain the packet header and payload part.
  • the node fuses the public AI model based on model fusion auxiliary information (version number, timestamp, geographical location, data distribution, etc.).
  • the public AI model information sent from N neighbors can be grouped and fused using multi-level fusion. For example, group the public AI model information in data packages with similar timestamps into one group, or group the public AI model information in data packages with similar geographical location information into one group, or group the public AI model information in data packages with different data distribution. Public AI model information is grouped into one group. Model fusion can be implemented using methods such as weighted average and distillation, which will not be described in detail here.
  • the version number can be designed as AxLyHz, A, L, H are fixed fields, x, y, z are integers.
  • the version number update rules are as follows: Node L performs a training update on the public AI model based on the local AI model, and the y field in the model version number is incremented by 1; Node A performs a training update on the public AI model based on the received neighbor node. After fusion, an updated public AI model is obtained.
  • the x field in the model version number is incremented by 1, and the x field is incremented by 1 each time, and the y and z fields are set to 0; the H node performs the public AI model based on the local AI model and the public AI model sent from the neighbor.
  • the AI model is updated once, the x, y, and z fields in the model version number are incremented by 1, the x field is incremented by 1 each time, and the y and z fields are set to 0.
  • a circle represents an AI model, and the version number identifying a certain model is AxLyHz.
  • the value of x can be an integer such as 1, 2, 3 or 4; the value of y can be an integer such as 0, 1, y, u; the value of z can be 0, x, Integers such as v, w or z.
  • public AI models with similar version numbers can be divided into a group, and the intra-group fusion is performed first, and then grouping and intra-group fusion are performed until the fusion process is completed. .
  • public AI models with the same x field are divided into a group. Each time the intra-group fusion is completed, the x field is incremented by 1, and the y and z fields are reset to 0 until a fused public AI is obtained. Model.
  • version numbers of the public models finally obtained by fusion of different nodes in the system where the second node is located are similar, and may even be the same.
  • the purpose of the entire system is for each node to obtain a model for the completion of the node's local AI tasks. Therefore, the existence of the version number is only to allow more similar models to be integrated first. It does not necessarily mean that the performance of the model with a higher version will be better. The better.
  • the second node can also perform fusion (distillation, pruning, expansion) based on the auxiliary information of other AI models (such as timestamps, geographical locations, etc.).
  • fusion distillation, pruning, expansion
  • auxiliary information of other AI models such as timestamps, geographical locations, etc.
  • the second node performs model processing based on the N pieces of first information and the node type of the second node in step S403.
  • the process of obtaining the target AI model includes: When the node type of is a node type for fusion processing based on the AI models of other nodes, the second node performs model fusion on the N first AI models based on the N first information to obtain the target AI model. Specifically, when the node type of the second node is a node type for fusion processing based on the AI models of other nodes, after the second node receives N first information and determines N first AI models, the second node Model fusion is performed on the N first AI models to obtain the target AI model.
  • the second node performs model processing based on the N pieces of first information and the node type of the second node in step S403.
  • the process of obtaining the target AI model includes: When the node type is a node type that performs local training based on the local data and performs fusion processing based on the AI models of other nodes, the second node compares the N first AI models and the second AI model based on the N pieces of first information. Model fusion is performed to obtain the target AI model, where the second AI model is trained based on local data.
  • the second node determines the N first pieces of information after receiving N pieces of first information. After the AI model, the second node needs to train the second AI model based on local data, and perform model fusion on the N first AI models and the second AI model to obtain the target AI model.
  • the second AI model used by the second node can refer to the related description of the second AI model used by the first node and achieve corresponding technical effects, which will not be described again here.
  • the method further includes: the second node receiving indication information indicating a node type of the first node.
  • the second node may also receive indication information indicating the node type of the first node, so that the second node can clarify the node type of the first node based on the indication information, and subsequently the second node may determine the node type based on the first node.
  • a node's node type interacts with the first node.
  • the method further includes: the second node sending indication information indicating a node type of the second node.
  • the second node may also send indication information indicating the node type of the second node, so that other nodes can clarify the node type of the second node based on the indication information.
  • the node type of the second node may be determined based on the node type and the node type of the second node. This second node interacts with.
  • the method further includes: the second node determines a node type of the second node based on capability information and/or demand information; or, the second node receives an indication of the node type of the second node. instructions.
  • the second node can determine the node type of the second node based on its own capability information and/or demand information, or can also determine the node type of the second node based on instructions from other nodes to improve the implementation of the solution. flexibility.
  • the relevant implementation process of the node type of the second node can refer to the relevant description of the node type of the first node and achieve the corresponding technical effects, which will not be done here. Repeat.
  • the first node After the first node determines the first AI model, the first node sends first information indicating the model information of the first AI model and the auxiliary information of the first AI model.
  • the first information can also indicate the auxiliary information of the first AI model, so that the first The recipient of the information can perform AI model processing (such as training, fusion, etc.) on the model information of the first AI model based on the auxiliary information of the first AI model, and the recipient of the improved information can perform AI model processing based on the first AI model. Performance of the resulting AI model.
  • FIG 6 is another interactive schematic diagram of the AI model processing method provided by this application.
  • the method includes the following steps.
  • the first node obtains the local AI model.
  • the first node acquires a local AI model in step S601, and the local AI model is used to complete the AI task of the first node.
  • the local AI model obtained by the first node in step S601 is obtained based on local data; or, the local AI model obtained by the first node in step S601 is obtained based on K pieces of information, and among the K pieces of information Each information indicates the model information of the AI model of other nodes and the auxiliary information of the AI model of the other node, and K is a positive integer; or, the local AI model obtained by the first node in step S601 is based on the local data and the K pieces of information get.
  • the local AI model used to obtain the public AI model can be implemented as any of the above to improve the flexibility of solution implementation.
  • the relevant implementation process of the local AI model obtained by the first node in step S601 can refer to the aforementioned Figure 4 and the description of the related embodiments, and implement the corresponding The technical effects will not be described here.
  • the first node determines the public AI model based on the local AI model.
  • the first node determines a public AI model based on the local AI model in step S402, where the public AI model is in the system where the first node is located.
  • a model that can be understood by M nodes, where M is an integer greater than or equal to 2.
  • the public AI model determined by the first node in step S602 is a model understandable by M nodes in the system where the first node is located, and can be expressed as the public AI model is common to M nodes in the system where the first node is located.
  • the model structure of the public AI model is a model structure understandable by M nodes in the system where the first node is located, and the model structure of the public AI model is a model structure common to M nodes in the system where the first node is located.
  • the process of the first node determining the public AI model based on the local AI model includes: the first node determines the public AI based on the local AI model and the node type of the first node. Model. Specifically, the first node can determine the public AI model based on the local AI model and the node type of the first node, so that the first node can perform different AI model processing processes based on different node types to solve node functions. Single question to increase flexibility.
  • the relevant implementation process of the node type of the first node can refer to the aforementioned FIG. 4 and the description of the related embodiments, and achieve corresponding technical effects, and will not be described again here. .
  • the node type includes any of the following: a node type for local training based on local data, a node type for fusion processing based on the AI model of other nodes, a node type for local training based on the local data and based on the AI model of other nodes.
  • the node type for fusion processing Specifically, the node type used to obtain the public AI model can be implemented as any of the above to improve the flexibility of solution implementation.
  • the method further includes: the first node sending indication information indicating the node type of the first node.
  • the first node may also send indication information indicating the node type of the first node, so that other nodes can clarify the node type of the first node based on the indication information. Subsequently, based on the node type of the first node and The first node to interact with.
  • the method further includes: the first node determines the node type of the first node based on capability information and/or demand information; or, the first node receives an instruction indicating that the first node Instructions for the node type.
  • the first node can determine the node type of the first node based on its own capability information and/or demand information, or can also determine the node type of the first node based on instructions from other nodes to improve the implementation of the solution. flexibility.
  • the first node sends the first information, which is used to indicate the model information of the public AI model.
  • the first node sends the first information indicating the model information of the public AI model in step S603.
  • the second node receives data from the Nth node in step S603.
  • N first messages sent by a node, N is a positive integer.
  • the relevant implementation process of the first information can refer to the aforementioned FIG. 4 and the description of the related embodiments, and achieve the corresponding technical effects, which will not be described here. Repeat.
  • the second node updates the local AI model based on the N pieces of first information, and obtains the updated local AI model information.
  • the second node after the second node receives N pieces of first information in step S603, the second node updates the local AI model based on the N pieces of first information in step S604 to obtain updated local AI model information.
  • the first information sent by the first node in step S603 also indicates the auxiliary information of the public AI model. Specifically, the first information sent by the first node also indicates the auxiliary information of the public AI model. Compared with the way in which different nodes only interact with their own AI models, since the first information indicates the model information of the public AI model, the first information can also indicate the auxiliary information of the public AI model, so that the first information The receiver can perform AI model processing (such as training, fusion, etc.) on the model information of the public AI model based on the auxiliary information of the first AI model, and improve the AI obtained by the receiver of the first information based on the public AI model. model performance.
  • AI model processing such as training, fusion, etc.
  • the auxiliary information of the public AI model includes at least one of the following: type information of the public AI model, identification information of the first node, identification information of the receiving node of the public AI model, and version of the public AI model.
  • Information generating time information of the public AI model, generating geographical location information of the public AI model, and distribution information of local data of the public node.
  • the auxiliary information of the public AI model indicated by the first information includes at least one of the above items, so that the recipient of the first information can perform AI model processing on the model information of the public AI model based on the at least one of the above information, thereby improving The performance of the AI model obtained by the recipient of the first information based on the public AI model.
  • the model information of the public AI model and the auxiliary information of the public AI model are respectively carried in different transmission resources, and the transmission resources include time domain resources and/or frequency domain resources.
  • the transmission resources include time domain resources and/or frequency domain resources.
  • the relevant implementation process of the first information (such as the receiving process of the first information, the AI model processing process of the second node based on the first information, etc.) can refer to the aforementioned Figure 4 and the description of the related embodiments, and implement the corresponding technology. The effect will not be described in detail here.
  • the second node updates the local AI model based on the N pieces of first information.
  • the process of obtaining the updated local AI model includes: the second node updates the local AI model based on the N pieces of first information.
  • the first information and the node type of the second node are used to update the local AI model to obtain an updated local AI model.
  • the second node can determine the public AI model based on the N pieces of first information and the node type of the second node, so that the second node can perform different AI model processing processes based on different node types, solving Node has a single function to improve flexibility.
  • the node type includes any of the following: a node type that performs fusion processing based on the AI models of other nodes, a node type that performs local training based on the local data and performs fusion processing based on the AI models of other nodes.
  • the node type used to obtain the updated local AI model can be implemented as any of the above to improve the flexibility of solution implementation.
  • the local AI model is obtained based on P pieces of information, and each piece of information in the P pieces of information indicates the model information of the AI models of other nodes. and the auxiliary information of the AI model of other nodes, where P is a positive integer.
  • the local AI model is obtained based on the local data (for example, during the first iteration process), or , the local AI model is obtained based on the local data and the P pieces of information (for example, in other iterative processes other than the first iterative process), and each of the P pieces of information indicates the model information of the AI models of other nodes. and the auxiliary information of the AI model of other nodes, where P is a positive integer.
  • the local AI model can be implemented differently as mentioned above to improve the flexibility of solution implementation.
  • the method further includes: the second node sending indication information indicating a node type of the second node.
  • the second node may also send indication information indicating the node type of the second node, so that other nodes can clarify the node type of the second node based on the indication information.
  • the node type of the second node may be determined based on the node type and the node type of the second node. This second node interacts with.
  • the method further includes: the second node determining a node type of the second node based on capability information and/or demand information; or, the second node receiving indication information indicating a node type of the second node.
  • the second node can determine the node type of the second node based on its own capability information and/or demand information, or can also determine the node type of the second node based on instructions from other nodes to improve the implementation of the solution. flexibility.
  • the first node obtains the local AI model used to complete the AI task of the first node, the first node determines the public AI model based on the local AI model, and the first node sends a message indicating the The first information of the model information of the public AI model.
  • the public AI model is a model understandable by M nodes in the system where the first node is located.
  • the public AI model indicated by the first information sent by the first node is a model understandable by M nodes in the system where the first node is located, so as to facilitate the situation where there are multiple different node types among the M nodes. Under this condition, other nodes can understand the first AI model and further perform model processing after receiving the first information.
  • FIG 7 is another interactive schematic diagram of the AI model processing method provided by this application.
  • the method includes the following steps.
  • the first node determines the node type of the first AI node.
  • the first node determines the node type of the first node in step S701, and the first node performs at least one of steps S702 and S703 after step S701.
  • step S702 is executed first and step S703 is executed later, or step S703 is executed first and step S702 is executed later.
  • the node type determined by the first node in step S701 includes any of the following: a node type for local training based on local data, a node type for fusion processing based on AI models of other nodes, The node type that performs local training on local data and performs fusion processing based on the AI models of other nodes.
  • the node type used to obtain the first AI model can be implemented as any of the above to improve the flexibility of solution implementation.
  • the method further includes: the first node determining a node type of the first node based on capability information and/or demand information; or, the first node receiving indication information indicating a node type of the first node.
  • the first node can determine the node type of the first node based on its own capability information and/or demand information, or can also determine the node type of the first node based on instructions from other nodes to improve the implementation of the solution. flexibility.
  • the first node sends indication information indicating the node type of the first node.
  • the first node determines the node type of the first node in step S701
  • the first node sends indication information indicating the node type of the first node in step S702.
  • the second node receives indication information indicating the node type of the first node in step S702.
  • the first node may also send indication information indicating the node type of the first node, so that other nodes can clarify the node type of the first node based on the indication information. Subsequently, based on the node type of the first node and The first node to interact with.
  • the relevant implementation process of the node type of the first node can refer to the aforementioned FIG. 4 and the description of the related embodiments, and achieve corresponding technical effects, and will not be described again here. .
  • the first node sends the first information, and the first information is used to indicate the model information of the first AI model.
  • the first node determines the node type of the first node in step S701
  • the first node determines the first AI model based on the node type of the first node, and the first node sends First information indicating model information of the first AI model.
  • the second node receives N pieces of first information sent by N pieces of first nodes in step S703.
  • the first AI model indicated by the first information sent by the first node may be a model obtained by the first node based on the node type of the first node.
  • the first AI model is obtained based on at least the node type of the first node, the first node can perform different AI model processing processes based on different node types, solving the problem of a single node function and improving flexibility.
  • the first AI model indicated by the first information sent by the first node may be a model obtained by the first node based on the node type of the first node and the second AI model.
  • the first AI model indicated by the first information sent by the first node may be a model understandable by other nodes, so that other nodes can After receiving the first AI model, further model processing is performed.
  • the second AI model is obtained based on local data; or, the second AI model is obtained based on K pieces of information, each of the K pieces of information indicates the model information of the AI model of other nodes and the model information of the other nodes.
  • K is a positive integer; or, the second AI model is obtained based on the local data and the K pieces of information.
  • the second AI model used to obtain the first AI model can be implemented by any of the above to improve the flexibility of solution implementation.
  • the first information is also used to indicate auxiliary information of the first AI model.
  • the first information sent by the first node also indicates the auxiliary information of the first AI model.
  • the first information can also indicate the auxiliary information of the first AI model, so that the first The recipient of the information can perform AI model processing (such as training, fusion, etc.) on the model information of the first AI model based on the auxiliary information of the first AI model, and the recipient of the improved information can perform AI model processing based on the first AI model. Performance of the resulting AI model.
  • the auxiliary information of the first AI model includes at least one of the following: type information of the first AI model, identification information of the first node, identification information of the receiving node of the first AI model, the first The version information of the AI model, the time information of the first AI model, the geographical location information of the first AI model, and the distribution information of local data of the first node.
  • the auxiliary information of the first AI model indicated by the first information includes at least one of the above items, so that the recipient of the first information can perform AI model processing on the model information of the first AI model based on the at least one item of information above. , improving the performance of the AI model obtained by the receiver of the first information based on the first AI model.
  • the model information of the first AI model and the auxiliary information of the first AI model are respectively carried in different transmission resources, and the transmission resources include time domain resources and/or frequency domain resources.
  • the transmission resources include time domain resources and/or frequency domain resources.
  • the first AI model is a model understandable by M nodes in the system where the first node is located, and M is an integer greater than or equal to 2.
  • the first AI model indicated by the first information sent by the first node is a model understandable by M nodes in the system where the first node is located, so that there are multiple different node types among the M nodes. In this case, other nodes can understand the first AI model and further perform model processing after receiving the first information.
  • the first node can determine the node type of the first node, and subsequently the first node can perform AI model processing based on the node type. This allows the first node to perform different AI model processing processes based on different node types, solving the problem of a single node function and improving flexibility.
  • 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 therefore can also implement The above method embodiments have beneficial effects.
  • the communication device 800 may be the first node, or may be an integrated circuit or component within the first node, such as a chip. The following embodiments will be described taking the communication device 800 as the first node as an example.
  • the device 800 when the device 800 is used to execute the method executed by the first node in any of the foregoing embodiments, the device 800 includes a processing unit 801 and a transceiver unit 802; the processing unit 801 is used to determine First AI model; the transceiver unit 802 is used to send first information, where the first information indicates model information of the first AI model and auxiliary information of the first AI model.
  • the auxiliary information of the first AI model includes at least one of the following:
  • the type information of the first AI model, the identification information of the first node, the identification information of the receiving node of the first AI model, the version information of the first AI model, the time information of generating the first AI model, and generating the The geographical location information of the first AI model and the distribution information of local data of the first node.
  • the model information of the first AI model and the auxiliary information of the first AI model are respectively carried in different transmission resources, and the transmission resources include time domain resources and/or frequency domain resources.
  • the first AI model is obtained based on the node type of the first node.
  • the first AI model is obtained based on the second AI model and the node type of the first node.
  • the second AI model is obtained based on local data; or, the second AI model is obtained based on K pieces of information, and each information in the K pieces of information indicates model information of the AI models of other nodes. and the auxiliary information of the AI model of other nodes, K is a positive integer; or, the second AI model is obtained based on the local data and the K pieces of information.
  • the node type includes any of the following: a node type for local training based on local data, a node type for fusion processing based on AI models of other nodes, local training based on the local data and The node type for fusion processing of AI models of other nodes.
  • the transceiver unit 802 is also configured to send indication information indicating the node type of the first node.
  • the processing unit 801 is also configured to determine the node type of the first node based on capability information and/or demand information; or, the transceiver unit 802 is also configured to receive an indication of the node type of the first node. Type instructions.
  • the first AI model is a model understandable by M nodes in the system where the first node is located, and M is an integer greater than or equal to 2.
  • the device 800 when the device 800 is used to execute the method executed by the second node in any of the foregoing embodiments, the device 800 includes a processing unit 801 and a transceiver unit 802; the transceiver unit 802 is used to receive N first information, each of the N first information indicates the model information of the first AI model and the auxiliary information of the first AI model, N is a positive integer; the processing unit 801 is configured to based on the The N first information is subjected to model processing to obtain the target AI model.
  • the target AI model is used to complete the AI task of the second node, or the target AI model is a local model of the second node.
  • the auxiliary information of the first AI model includes at least one of the following: type information of the first AI model, identification information of the first node, and identification information of the receiving node of the first AI model. , the version information of the first AI model, the time information of the first AI model, the geographical location information of the first AI model, and the distribution information of the local data of the first node.
  • the model information of the first AI model and the auxiliary information of the first AI model are respectively carried in different transmission resources, and the transmission resources include time domain resources and/or frequency domain resources.
  • the processing unit 801 is specifically configured to perform model processing based on the N pieces of first information and the node type of the second node to obtain the target AI model.
  • the processing unit 801 is specifically configured to perform fusion processing on the N pieces of information based on the N pieces of first information.
  • the first AI model performs model fusion to obtain the target AI model.
  • the processing unit 801 is specifically configured to perform local training based on the N
  • the first information is model-fused to the N first AI models and the second AI model to obtain the target AI model, where the second AI model is trained based on local data.
  • the transceiver unit 802 is also configured to receive indication information indicating the node type of the first node.
  • the transceiver unit 802 is also configured to send indication information indicating the node type of the second node.
  • the processing unit 801 is also configured to determine the node type of the second node based on capability information and/or demand information; or, the transceiver unit 802 is also configured to receive an indication of the node type of the second node. Type instructions.
  • the first AI model is a model understandable by M nodes in the system where the second node is located, and M is an integer greater than or equal to 2.
  • the device 800 when the device 800 is used to execute the method executed by the first node in any of the foregoing embodiments, the device 800 includes a processing unit 801 and a transceiver unit 802; the processing unit 801 is used to obtain A local AI model, the local AI model is used to complete the AI task of the first node; the processing unit 801 is also used to determine a public AI model based on the local AI model, the public AI model is M in the system where the first node is located A model that can be understood by nodes, M is an integer greater than or equal to 2; the transceiver unit 802 is used to send first information, and the first information is used to indicate model information of the public AI model.
  • the processing unit 801 is specifically configured to determine a public AI model based on the local AI model and the node type of the first node.
  • the node type includes any of the following: a node type for local training based on local data, a node type for fusion processing based on AI models of other nodes, local training based on the local data and The node type for fusion processing of AI models of other nodes.
  • the transceiver unit 802 is also configured to send indication information indicating the node type of the first node.
  • the processing unit 802 is also configured to determine the node type of the first node based on capability information and/or demand information; or, the transceiver unit 802 is also configured to receive an indication of the node type of the first node. Type instructions.
  • the local AI model is obtained based on local data; or,
  • the local AI model is obtained based on K pieces of information.
  • Each piece of information in the K pieces of information indicates the model information of the AI model of other nodes and the auxiliary information of the AI model of the other nodes.
  • K is a positive integer; or,
  • the local AI model is obtained based on the local data and the K pieces of information.
  • the first information also indicates auxiliary information of the public AI model.
  • the auxiliary information of the public AI model includes at least one of the following:
  • the model information of the public AI model and the auxiliary information of the public AI model are respectively carried in different transmission resources, and the transmission resources include time domain resources and/or frequency domain resources.
  • the device 800 when the device 800 is used to execute the method executed by the second node in any of the foregoing embodiments, the device 800 includes a processing unit 801 and a transceiver unit 802; the transceiver unit 802 is used to receive N first information, each of the N first information is used to indicate model information of a public AI model.
  • the public AI model is a model understandable by M nodes in the system where the first node is located, M is an integer greater than or equal to 2; the processing unit 801 is used to update the local AI model based on the N pieces of first information to obtain an updated local AI model.
  • the local AI model is used to complete the AI task of the second node.
  • the processing unit 801 is specifically configured to update the local AI model based on the N pieces of first information and the node type of the second node, to obtain an updated local AI model.
  • the node type includes any of the following: a node type that performs fusion processing based on the AI models of other nodes, a node that performs local training based on the local data and performs fusion processing based on the AI models of other nodes. type.
  • the transceiver unit 802 is also configured to send indication information indicating the node type of the second node.
  • the processing unit 801 is also configured to determine the node type of the second node based on capability information and/or demand information; or, the transceiver unit 802 is also configured to receive an indication of the node type of the second node. Type instructions.
  • the local AI model is obtained based on P pieces of information, and each piece of information in the P pieces of information indicates the data of other nodes.
  • the model information of the AI model and the auxiliary information of the AI model of other nodes, P is a positive integer.
  • the local AI model is obtained based on the local data, or, the local AI The model is obtained based on the local data and the P pieces of information.
  • Each of the P pieces of information indicates the model information of the AI model of other nodes and the auxiliary information of the AI model of the other nodes. P is a positive integer.
  • the first information is also used to indicate auxiliary information of the public AI model.
  • the auxiliary information of the public AI model includes at least one of the following: type information of the public AI model, identification information of the first node, identification information of the receiving node of the public AI model, the The version information of the public AI model, the time information of the public AI model, the geographical location information of the public AI model, and the distribution information of the local data of the first node.
  • the model information of the public AI model and the auxiliary information of the public AI model are respectively carried in different transmission resources, and the transmission resources include time domain resources and/or frequency domain resources.
  • the device 800 when the device 800 is used to execute the method executed by the first node in any of the foregoing embodiments, the device 800 includes a processing unit 801 and a transceiver unit 802; the processing unit 801 is used for the first node.
  • a node determines the node type of the first node.
  • the device further includes a transceiver unit, the transceiver unit 802 being configured to send indication information indicating the node type of the first node.
  • the device further includes a transceiver unit 802, the transceiver unit 802 is configured to send first information, the first information is used to indicate the model information of the first AI model, wherein the first AI The model is obtained based on the node type of the first node.
  • the first AI model is obtained based on the second AI model and the node type of the first node.
  • the second AI model is obtained based on local data; or,
  • the second AI model is obtained based on K information, each information in the K information indicates the model information of the AI model of other nodes and the auxiliary information of the AI model of the other node, K is a positive integer; or,
  • the second AI model is obtained based on the local data and the K pieces of information.
  • the node type includes any of the following: a node type for local training based on local data, a node type for fusion processing based on AI models of other nodes, local training based on the local data and The node type for fusion processing of AI models of other nodes.
  • the first information is also used to indicate auxiliary information of the first AI model, wherein the auxiliary information of the first AI model includes at least one of the following: type information of the first AI model , the identification information of the first node, the identification information of the receiving node of the first AI model, the version information of the first AI model, the time information of generating the first AI model, and the geographical location information of generating the first AI model. , the distribution information of the local data of the first node.
  • the model information of the first AI model and the auxiliary information of the first AI model are respectively carried in different transmission resources, and the transmission resources include time domain resources and/or frequency domain resources.
  • the first AI model is a model understandable by M nodes in the system where the first node is located, and M is an integer greater than or equal to 2.
  • the processing unit 801 is also configured to determine the node type of the first node based on capability information and/or demand information; or, the transceiver unit 802 is also configured to receive an indication of the node type of the first node. Type instructions.
  • the device 800 when the device 800 is used to perform the method performed by the second node in any of the foregoing embodiments, the device 800 includes a transceiver unit 802; the transceiver unit 802 is used for the second node to receive instructions. Indication information of the node type of the first node; and/or, the transceiver unit 802 is configured to receive first information, the first information is used to indicate model information of the first AI model, wherein the first AI model is Obtained based on the node type of the first node.
  • the first AI model is obtained based on the second AI model and the node type of the first node.
  • the second AI model is obtained based on local data; or,
  • the second AI model is obtained based on K information, each information in the K information indicates the model information of the AI model of other nodes and the auxiliary information of the AI model of the other node, K is a positive integer; or,
  • the second AI model is obtained based on the local data and the K pieces of information.
  • the node type includes any of the following: a node type for local training based on local data, a node type for fusion processing based on AI models of other nodes, local training based on the local data and The node type for fusion processing of AI models of other nodes.
  • the first information is also used to indicate auxiliary information of the first AI model, wherein the auxiliary information of the first AI model includes at least one of the following: type information of the first AI model , the identification information of the first node, the identification information of the receiving node of the first AI model, the version information of the first AI model, the time information of generating the first AI model, and the geographical location information of generating the first AI model. , the distribution information of the local data of the first node.
  • the model information of the first AI model and the auxiliary information of the first AI model are respectively carried in different transmission resources, and the transmission resources include time domain resources and/or frequency domain resources.
  • the first AI model is a model understandable by M nodes in the system where the first node is located, and M is an integer greater than or equal to 2.
  • the device further includes a processing unit 801, which is configured to determine the node type of the second node based on capability information and/or demand information; or, the transceiver unit is also configured to receive an indication. Indication information of the node type of the second node.
  • FIG. 9 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.
  • the communication device also includes a logic circuit 901.
  • the transceiver unit 802 shown in FIG. 8 may be a communication interface, and the communication interface may be the input-output interface 902 in FIG. 9 .
  • the input-output interface 902 may include an input interface and an output interface.
  • the communication interface may also be a transceiver circuit, and the transceiver circuit may include an input interface circuit and an output interface circuit.
  • the logic circuit 901 is used to determine the first AI model; the input and output interface 902 is used to send first information indicating the model information of the first AI model and the auxiliary information of the first AI model.
  • the input and output interface 902 is used to receive N pieces of first information, each of the N pieces of first information indicating model information of the first AI model and auxiliary information of the first AI model; logic circuit 901 is used to perform model processing based on the N pieces of first information to obtain the target AI model.
  • the logic circuit 901 is used to obtain a local AI model, which is used to complete the AI task of the first node; the logic circuit 901 is used to determine a public AI model based on the local AI model, and the public AI model is A model understandable by M nodes in the system where the first node is located, where M is an integer greater than or equal to 2; the input and output interface 902 is used to send first information, and the first information is used to indicate model information of the public AI model.
  • the input and output interface 902 is used to receive N pieces of first information.
  • Each of the N pieces of first information is used to indicate model information of a public AI model.
  • the public AI model is where the first node is located.
  • the local AI model is To complete the AI task of the second node.
  • the logic circuit 901 is used for determining the node type of the first node.
  • the input and output interface 902 is used by the second node to receive indication information indicating the node type of the first node;
  • the input and output interface 902 is used to receive first information, where the first information is used to indicate model information of the first AI model, where the first AI model is obtained based on the node type of the first node.
  • the logic circuit 901 and the input-output interface 902 can also perform other steps performed by the network device in any embodiment and achieve corresponding beneficial effects, which will not be described again here.
  • the processing unit 801 shown in FIG. 8 may be the logic circuit 901 in FIG. 9 .
  • the logic circuit 901 may be a processing device, and the functions of the processing device may be partially or fully implemented through software. Among them, the functions of the processing device can be partially or fully implemented through software.
  • 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 method embodiment. .
  • the processing means may comprise only a processor.
  • the memory for storing computer programs is located outside the processing device, and the processor is connected to the memory through circuits/wires to read and execute the computer programs stored in the memory.
  • the memory and processor can be integrated together, or they can also be physically independent of each other.
  • the processing device may be one or more chips, or one or more integrated circuits.
  • the processing device may be one or more field-programmable gate arrays (FPGA), application specific integrated circuit (ASIC), system on chip (SoC), central processing unit (central processor unit, CPU), network processor (network processor, NP), digital signal processing circuit (digital signal processor, DSP), microcontroller unit (micro controller unit, MCU), programmable logic device, PLD) or other integrated chips, or any combination of the above chips or processors, etc.
  • FPGA field-programmable gate arrays
  • ASIC application specific integrated circuit
  • SoC system on chip
  • central processing unit central processor unit, CPU
  • network processor network processor
  • NP network processor
  • DSP digital signal processing circuit
  • microcontroller unit microcontroller unit
  • microcontroller unit micro controller unit, MCU
  • PLD programmable logic device
  • FIG. 10 is a communication device 1000 involved in the above embodiment provided by an embodiment of the present application.
  • the communication device 1000 may specifically be a communication device serving as a terminal device in the above embodiment.
  • the example shown in FIG. 10 is a terminal.
  • the device is implemented through a terminal device (or a component in the terminal device).
  • the communication device 1000 may include but is not limited to at least one processor 1001 and a communication port 1002.
  • the device may also include at least one of a memory 1003 and a bus 1004.
  • the at least one processor 1001 is used to control the actions of the communication device 1000.
  • the processor 1001 may 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 may implement or execute the various illustrative logical blocks, modules, and circuits described in connection with this disclosure.
  • the processor may 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.
  • the communication device 1000 shown in Figure 10 can be specifically used to implement the steps implemented by the terminal device in the foregoing method embodiment, and to achieve the corresponding technical effects of the terminal device.
  • the specific implementation methods of the communication device shown in Figure 10 are all Reference may be made to the descriptions in the foregoing method embodiments, which will not be described again here.
  • FIG. 11 is a schematic structural diagram of a communication device 1100 involved in the above embodiment provided by an embodiment of the present application.
  • the communication device 1100 may specifically be a communication device serving as a network device in the above embodiment, as shown in FIG. 11
  • An example is that the network device is implemented by a network device (or a component in the network device), wherein the structure of the communication device may refer to the structure shown in FIG. 11 .
  • the communication device 1100 includes at least one processor 1111 and at least one network interface 1114. Further optionally, the communication device further 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, through a bus. In the embodiment of the present application, the connection may include various interfaces, transmission lines or buses, etc., which is not limited in this embodiment.
  • Antenna 1115 is connected to transceiver 1113.
  • the network interface 1114 is used to enable the communication device to communicate with other communication devices through communication links.
  • the network interface 1114 may include a network interface between a communication device and a core network device, such as an S1 interface, and the network interface may include a network interface between a communication device and other communication devices (such as other network devices or core network devices), such as an X2 Or Xn interface.
  • a network interface between a communication device and a core network device such as an S1 interface
  • the network interface may include a network interface between a communication device and other communication devices (such as other network devices or core network devices), such as an X2 Or Xn interface.
  • the processor 1111 is mainly used to process communication protocols and communication data, control the entire communication device, execute software programs, and process data of the software programs, for example, to support the communication device to perform actions described in the embodiments.
  • the communication device may include a baseband processor and a central processing unit.
  • the baseband processor is mainly used to process communication protocols and communication data.
  • the central processing unit is mainly used to control the entire terminal device, execute software programs, and process data of the software programs.
  • the processor 1111 in Figure 11 can integrate the functions of the baseband processor and the central processor. Those skilled in the art can understand that the baseband processor and the central processor can also be independent processors, interconnected through technologies such as buses.
  • the terminal device may include multiple baseband processors to adapt to different network standards, the terminal device may include multiple central processors to enhance its processing capabilities, and the various components of the terminal device may be connected through various buses.
  • the baseband processor can also be expressed as a baseband processing circuit or a baseband processing chip.
  • the central processing unit can also be expressed as a central processing circuit or a central processing chip.
  • the function of processing communication protocols and communication data can be built into the processor, or can be stored in the memory in the form of a software program, and the processor executes the software program to implement the baseband processing function.
  • Memory is mainly used to store software programs and data.
  • the memory 1112 may exist independently and be connected to the processor 1111.
  • the memory 1112 can be integrated with the processor 1111, for example, integrated into a chip.
  • the memory 1112 can store the program code for executing the technical solution of the embodiment of the present application, and the execution is controlled by the processor 1111.
  • the various computer program codes executed can also be regarded as the driver of the processor 1111.
  • Figure 11 shows only one memory and one processor. In an actual terminal device, there may be multiple processors and multiple memories. Memory can also be called storage media or storage devices.
  • the memory may be a storage element on the same chip as the processor, that is, an on-chip storage element, or an independent storage element, which is not limited in the embodiments of the present application.
  • the transceiver 1113 may be used to support the reception or transmission of radio frequency signals between the communication device and the terminal, and the transceiver 1113 may be connected to the antenna 1115.
  • Transceiver 1113 includes a transmitter Tx and a receiver Rx.
  • one or more antennas 1115 can receive radio frequency signals
  • the receiver Rx of the transceiver 1113 is used to receive the radio frequency signals from the antennas, convert the radio frequency signals into digital baseband signals or digital intermediate frequency signals, and convert the digital baseband signals into digital baseband signals.
  • the signal or digital intermediate frequency signal is provided to the processor 1111, so that the processor 1111 performs further processing on the digital baseband signal or digital intermediate frequency signal, such as demodulation processing and decoding processing.
  • the transmitter Tx in the transceiver 1113 is also used to receive a modulated digital baseband signal or a digital intermediate frequency signal from the processor 1111, and convert the modulated digital baseband signal or digital intermediate frequency signal into a radio frequency signal, and pass it through a Or multiple antennas 1115 transmit the radio frequency signal.
  • the receiver Rx can selectively perform one or more levels of down-mixing processing and analog-to-digital conversion processing on the radio frequency signal to obtain a digital baseband signal or a digital intermediate frequency signal.
  • the sequence of the down-mixing processing and the analog-to-digital conversion processing is The order is adjustable.
  • the transmitter Tx can selectively perform one or more levels of upmixing processing and digital-to-analog conversion processing on the modulated digital baseband signal or digital intermediate frequency signal to obtain a radio frequency signal.
  • the upmixing processing and digital-to-analog conversion processing are The order is adjustable.
  • Digital baseband signals and digital intermediate frequency signals can be collectively referred to as digital signals.
  • the transceiver 1113 may also be called a transceiver unit, a transceiver, a transceiver device, etc.
  • the devices used to implement the receiving function in the transceiver unit can be regarded as the receiving unit
  • the devices used in the transceiver unit used to implement the transmitting function can be regarded as the transmitting unit, that is, the transceiver unit includes a receiving unit and a transmitting unit, and the receiving unit also It can be called a receiver, input port, receiving circuit, etc.
  • the sending unit can be called a transmitter, transmitter, or transmitting circuit, etc.
  • the communication device 1100 shown in Figure 11 can be used to implement the steps implemented by the network equipment in the foregoing method embodiments, and to achieve the corresponding technical effects of the network equipment.
  • the specific implementation of the communication device 1100 shown in Figure 11 is: Reference may be made to the descriptions in the foregoing method embodiments, and details will not be repeated here.
  • Embodiments of the present application also provide a computer-readable storage medium that stores one or more computer-executable instructions.
  • the processor executes the possible implementations of the terminal device in the foregoing embodiments. Methods.
  • Embodiments of the present application also provide a computer-readable storage medium that stores one or more computer-executable instructions.
  • the processor executes the possible implementations of the network device in the foregoing embodiments. Methods.
  • Embodiments of the present application also provide a computer program product (or computer program) that stores one or more computers.
  • the processor executes the method of possible implementation of the above terminal device.
  • Embodiments of the present application also provide a computer program product that stores one or more computers.
  • the processor executes the method of the possible implementation of the network device.
  • Embodiments of the present application also provide a chip system, which includes at least one processor and is used to support the communication device in implementing the functions involved in the possible implementation manners of the communication device.
  • the chip system further includes an interface circuit that provides program instructions and/or data to the at least one processor.
  • the chip system may also include a memory for storing necessary program instructions and data of the communication device.
  • the chip system may be composed of chips, or may include chips and other discrete devices, where the communication device may specifically be the terminal equipment in the foregoing method embodiment.
  • Embodiments of the present application also provide a chip system, which includes at least one processor and is used to support the communication device in implementing the functions involved in the possible implementation manners of the communication device.
  • the chip system further includes an interface circuit that provides program instructions and/or data to the at least one processor.
  • the chip system may also include a memory, which is used to store necessary program instructions and data for the communication device.
  • the chip system may be composed of chips, or may include chips and other discrete devices.
  • the communication device may specifically be the network device in the aforementioned method embodiment.
  • the embodiment of the present application also provides a communication system.
  • the network system architecture includes the first node and the second node in any of the above embodiments.
  • the first node can be a terminal device or a network device, and the second node can also be For terminal equipment or network equipment.
  • the disclosed systems, devices and methods can be implemented in other ways.
  • the device embodiments described above are only illustrative.
  • the division of the units is only a logical function division. In actual implementation, there may be other division methods.
  • multiple units or components may be combined or can be integrated into another system, or some features can be ignored, or not implemented.
  • the coupling or direct coupling or communication connection between each other shown or discussed may be through some interfaces, and the indirect coupling or communication connection of the devices or units may be in electrical, mechanical or other forms.
  • the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place, or they may be distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.
  • each functional unit in each embodiment of the present application can be integrated into one processing unit, each unit can exist physically alone, or two or more units can be integrated into one unit.
  • the above integrated units can be implemented in the form of hardware or software functional units. If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it may be stored in a computer-readable storage medium.
  • the technical solution of the present application is essentially or contributes to the existing technology, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium , including several instructions to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of this application.
  • the aforementioned storage media include: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or optical disk and other media that can store program code. .

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Software Systems (AREA)
  • Computing Systems (AREA)
  • Artificial Intelligence (AREA)
  • Mathematical Physics (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Biomedical Technology (AREA)
  • Molecular Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Data Exchanges In Wide-Area Networks (AREA)

Abstract

本申请提供了一种AI模型处理方法及相关设备，用于提升AI模型的性能。在该方法中，第一节点确定第一AI模型；该第一节点发送第一信息，该第一信息指示该第一AI模型的模型信息和该第一AI模型的辅助信息。相比于不同节点之间仅交互各自的AI模型的方式，由于第一信息指示第一AI模型的模型信息之外，该第一信息还可以指示第一AI模型的辅助信息，使得该第一信息的接收方能够基于该第一AI模型的辅助信息对该第一AI模型的模型信息进行AI模型处理(例如训练、融合等)，提升该第一信息的接收方基于该第一AI模型进行处理得到的AI模型的性能。

Description

一种人工智能模型处理方法及相关设备 技术领域
本申请涉及通信领域，尤其涉及一种人工智能(artificial intelligence，AI)模型处理方法及相关设备。
背景技术
联邦学习(federated learning，FL)，作为一种典型的AI数据处理模式被广泛应用。其中，联邦学习系统中存在一个中心节点进行模型的融合，因而一般是星型结构，存在鲁棒性差的问题，一旦中心节点出现问题(如被攻击)，则会导致整个系统的瘫痪。
目前，去中心式的AI数据处理模式作为对联邦学习的AI数据处理模式的一种改进，该模式无需设置中心节点，可以提升系统的鲁棒性。在去中心式系统中，各个节点利用本地数据和本地目标训练本地AI模型之后，与通信可达的邻居节点交互各自的AI模型，并基于交互的AI模型进一步处理(例如训练、融合等)本地AI模型。
然而，在上述去中心式系统中，由于没有中心节点的统一调度，不同分布式节点在接收到其它节点的AI模型之后，如何实现AI模型处理，是一个亟待解决的技术问题。
发明内容
本申请提供了一种AI模型处理方法及相关设备，用于提升AI模型的性能。
本申请第一方面提供了一种AI模型处理方法，该方法由第一节点执行，或者，该方法由第一节点中的部分组件(例如处理器、芯片或芯片系统等)执行，或者该方法还可以由能实现全部或部分第一节点功能的逻辑模块或软件实现。在第一方面及其可能的实现方式中，以该方法由第一节点执行为例进行描述，该第一节点可以为终端设备或网络设备。在该方法中，第一节点确定第一AI模型；该第一节点发送第一信息，该第一信息指示该第一AI模型的模型信息和该第一AI模型的辅助信息。
基于上述技术方案，第一节点在确定第一AI模型之后，该第一节点发送指示该第一AI模型的模型信息和该第一AI模型的辅助信息的第一信息。相比于不同节点之间仅交互各自的AI模型的方式，由于第一信息指示第一AI模型的模型信息之外，该第一信息还可以指示第一AI模型的辅助信息，使得该第一信息的接收方能够基于该第一AI模型的辅助信息对该第一AI模型的模型信息进行AI模型处理(例如训练、融合等)，提升该第一信息的接收方基于该第一AI模型进行处理得到的AI模型的性能。
在第一方面的一种可能的实现方式中，该第一AI模型的辅助信息包括以下至少一项：该第一AI模型的类型信息，该第一节点的标识信息，该第一AI模型的接收节点的标识信息，该第一AI模型的版本信息，生成该第一AI模型的时间信息，生成该第一AI模型的地理位置信息，该第一节点的本地数据的分布信息。
基于上述技术方案，第一信息所指示的第一AI模型的辅助信息包括上述至少一项，以便于第一信息的接收方能够基于上述至少一项信息对该第一AI模型的模型信息进行AI模 型处理，提升该第一信息的接收方基于该第一AI模型进行处理得到的AI模型的性能。
在第一方面的一种可能的实现方式中，该第一AI模型的模型信息和该第一AI模型的辅助信息分别承载于不同的传输资源，该传输资源包括时域资源和/或频域资源。
基于上述技术方案，由于该第一AI模型的模型信息对应的数据量和该第一AI模型的辅助信息对应的数据量一般是不同的(例如前者一般多于后者)，使得两者可以分别承载于不同的传输资源上。
可选地，该传输资源为预配置的资源。
在第一方面的一种可能的实现方式中，该第一AI模型为基于该第一节点的节点类型得到。
基于上述技术方案，第一节点发送的第一信息所指示的第一AI模型可以为该第一节点基于该第一节点的节点类型得到的模型。其中，在该第一AI模型至少基于第一节点的节点类型得到的情况下，使得第一节点可以基于不同的节点类型可以执行不同的AI模型处理过程，解决节点功能单一的问题，以提升灵活性。
在第一方面的一种可能的实现方式中，该第一AI模型为基于第二AI模型和该第一节点的节点类型得到；该第二AI模型基于本地数据得到；或，该第二AI模型基于K个信息得到，该K个信息中的每个信息指示其它节点的AI模型的模型信息和该其它节点的AI模型的辅助信息，K为正整数；或，该第二AI模型基于该本地数据以及该K个信息得到。
基于上述技术方案，在该第一AI模型至少基于第二AI模型得到的情况下，该第一节点所发送的第一信息所指示的第一AI模型可以为其它节点可理解的模型，以便于其它节点在接收该第一AI模型之后进行进一步的模型处理。此外，用于得到该第一AI模型的第二AI模型可以为上述任一项实现，以提升方案实现的灵活性。
在第一方面的一种可能的实现方式中，该节点类型包括以下任一项：基于本地数据进行本地训练的节点类型，基于其它节点的AI模型进行融合处理的节点类型，基于该本地数据进行本地训练并基于其它节点的AI模型进行融合处理的节点类型。
基于上述技术方案，用于得到该第一AI模型的节点类型可以为上述任一项实现，以提升方案实现的灵活性。
在第一方面的一种可能的实现方式中，该方法还包括：该第一节点发送指示该第一节点的节点类型的指示信息。
基于上述技术方案，该第一节点还可以发送指示该第一节点的节点类型的指示信息，以便于其它节点基于该指示信息明确该第一节点的节点类型，后续可以基于该第一节点的节点类型与该第一节点进行交互。
在第一方面的一种可能的实现方式中，该方法还包括：该第一节点基于能力信息和/或需求信息确定该第一节点的节点类型；或，该第一节点接收指示该第一节点的节点类型的指示信息。
基于上述技术方案，该第一节点可以基于自身的能力信息和/或需求信息以明确该第一节点的节点类型，也可以基于其它节点的指示以明确该第一节点的节点类型，以提升方案实现的灵活性。
在第一方面的一种可能的实现方式中，第一AI模型为该第一节点所在的系统中M个节点可理解的模型，M为大于或等于2的整数。
应理解，第一AI模型为该第一节点所在的系统中M个节点可理解的模型，可以表述为第一AI模型为该第一节点所在的系统中M个节点通用的模型，第一AI模型的模型结构为该第一节点所在的系统中M个节点可理解的模型结构，第一AI模型的模型结构为该第一节点所在的系统中M个节点通用的模型结构，第一AI模型为该第一节点所在的系统的公共模型，或，第一AI模型的模型结构为该第一节点所在的系统的公共模型结构。
基于上述技术方案，第一节点发送的第一信息所指示的第一AI模型为该第一节点所在的系统中M个节点可理解的模型，以便于在该M个节点中存在多种不同的节点类型的情况下，其它节点在接收该第一信息之后能够理解该第一AI模型并进一步执行模型处理。
本申请第二方面提供了一种AI模型处理方法，该方法由第二节点执行，或者，该方法由第二节点中的部分组件(例如处理器、芯片或芯片系统等)执行，或者该方法还可以由能实现全部或部分第二节点功能的逻辑模块或软件实现。在第二方面及其可能的实现方式中，以该方法由第二节点执行为例进行描述，该第二节点可以为终端设备或网络设备。在该方法中，第二节点接收来自N个第一节点的N个第一信息，该N个第一信息中的每个第一信息指示第一AI模型的模型信息和该第一AI模型的辅助信息，N为正整数；该第二节点基于该N个第一信息进行模型处理，得到目标AI模型。
基于上述技术方案，第二节点接收的N个第一信息中，每个第一信息指示第一AI模型的模型信息和该第一AI模型的辅助信息，此后，第二节点基于该N个第一信息进行模型处理，得到目标AI模型。相比于不同节点之间仅交互各自的AI模型的方式，由于每个第一信息指示第一AI模型的模型信息之外，该每个第一信息还可以指示第一AI模型的辅助信息，使得该第二节点能够基于该第一AI模型的辅助信息对该第一AI模型的模型信息进行AI模型处理(例如训练、融合等)，提升该第二节点基于该第一AI模型进行处理得到的AI模型的性能。
可选地，该目标AI模型用于完成该第二节点的AI任务，或，该目标AI模型为该第二节点的本地模型。
在第二方面的一种可能的实现方式中，该第一AI模型的辅助信息包括以下至少一项：该第一AI模型的类型信息，第一节点的标识信息，该第一AI模型的接收节点的标识信息，该第一AI模型的版本信息，生成该第一AI模型的时间信息，生成该第一AI模型的地理位置信息，该第一节点的本地数据的分布信息。
基于上述技术方案，第一信息所指示的第一AI模型的辅助信息包括上述至少一项，以便于第二节点能够基于上述至少一项信息对该第一AI模型的模型信息进行AI模型处理，提升第二节点基于该第一AI模型进行处理得到的AI模型的性能。
在第二方面的一种可能的实现方式中，该第一AI模型的模型信息和该第一AI模型的辅助信息分别承载于不同的传输资源，该传输资源包括时域资源和/或频域资源。
基于上述技术方案，由于该第一AI模型的模型信息对应的数据量和该第一AI模型的 辅助信息对应的数据量一般是不同的(例如前者一般多于后者)，使得两者可以分别承载于不同的传输资源上。
可选地，该传输资源为预配置的资源。
在第二方面的一种可能的实现方式中，该第二节点基于该N个第一信息进行模型处理，得到目标AI模型包括：该第二节点基于该N个第一信息以及该第二节点的节点类型进行模型处理，得到该目标AI模型。
基于上述技术方案，第二节点经过模型处理得到的目标AI模型可以为该第二节点至少基于该第二节点的节点类型得到的模型。其中，在该目标AI模型至少基于第二节点的节点类型得到的情况下，使得第二节点可以基于不同的节点类型可以执行不同的AI模型处理过程，解决节点功能单一的问题，以提升灵活性。
在第二方面的一种可能的实现方式中，该第二节点基于该N个第一信息和该第二节点的节点类型进行模型处理，得到该目标AI模型包括：在该第二节点的节点类型为基于其它节点的AI模型进行融合处理的节点类型时，该第二节点基于该N个第一信息对该N个第一AI模型进行模型融合，得到该目标AI模型。
基于上述技术方案，在该第二节点的节点类型为基于其它节点的AI模型进行融合处理的节点类型时，该第二节点在接收N个第一信息确定N个第一AI模型之后，该第二节点对该N个第一AI模型进行模型融合，以得到该目标AI模型。
在第二方面的一种可能的实现方式中，该第二节点基于该N个第一信息和该第二节点的节点类型进行模型处理，得到该目标AI模型包括：在该第二节点的节点类型为基于该本地数据进行本地训练并基于其它节点的AI模型进行融合处理的节点类型时，该第二节点基于该N个第一信息对该N个第一AI模型以及第二AI模型进行模型融合，得到该目标AI模型，其中，该第二AI模型为基于本地数据训练得到。
基于上述技术方案，在该第二节点的节点类型为基于该本地数据进行本地训练并基于其它节点的AI模型进行融合处理的节点类型时，该第二节点在接收N个第一信息确定N个第一AI模型之后，该第二节点需要基于本地数据训练得到第二AI模型，并对该N个第一AI模型以及第二AI模型进行模型融合，以得到该目标AI模型。
在第二方面的一种可能的实现方式中，该方法还包括：该第二节点接收用于指示该第一节点的节点类型的指示信息。
基于上述技术方案，该第二节点还可以接收指示该第一节点的节点类型的指示信息，以便于该第二节点基于该指示信息明确该第一节点的节点类型，后续该第二节点可以基于该第一节点的节点类型与该第一节点进行交互。
在第二方面的一种可能的实现方式中，该方法还包括：该第二节点发送用于指示该第二节点的节点类型的指示信息。
基于上述技术方案，该第二节点还可以发送指示该第二节点的节点类型的指示信息，以便于其它节点基于该指示信息明确该第二节点的节点类型，后续可以基于该第二节点的节点类型与该第二节点进行交互。
在第二方面的一种可能的实现方式中，该方法还包括：该第二节点基于能力信息和/ 或需求信息确定该第二节点的节点类型；或，该第二节点接收指示该第二节点的节点类型的指示信息。
基于上述技术方案，该第二节点可以基于自身的能力信息和/或需求信息以明确该第二节点的节点类型，也可以基于其它节点的指示以明确该第二节点的节点类型，以提升方案实现的灵活性。
在第二方面的一种可能的实现方式中，第一AI模型为该第二节点所在的系统中M个节点可理解的模型，M为大于或等于2的整数。
应理解，第一AI模型为该第一节点所在的系统中M个节点可理解的模型，可以表述为第一AI模型为该第一节点所在的系统中M个节点通用的模型，第一AI模型的模型结构为该第一节点所在的系统中M个节点可理解的模型结构，第一AI模型的模型结构为该第一节点所在的系统中M个节点通用的模型结构，第一AI模型为该第一节点所在的系统的公共模型，或，第一AI模型的模型结构为该第一节点所在的系统的公共模型结构。
基于上述技术方案，在第二节点接收的N个第一信息中，每个第一信息所指示的第一AI模型为该第二节点所在的系统中M个节点可理解的模型，以便于在该M个节点中存在多种不同的节点类型的情况下，各个节点在接收该第一信息之后能够理解该第一AI模型并进一步执行模型处理。
本申请第三方面提供了一种AI模型处理方法，该方法由第一节点执行，或者，该方法由第一节点中的部分组件(例如处理器、芯片或芯片系统等)执行，或者该方法还可以由能实现全部或部分第一节点功能的逻辑模块或软件实现。在第三方面及其可能的实现方式中，以该方法由第一节点执行为例进行描述，该第一节点可以为终端设备或网络设备。在该方法中，第一节点获取本地AI模型，该本地AI模型用于完成该第一节点的AI任务；该第一节点基于该本地AI模型确定公共AI模型，该公共AI模型为该第一节点所在的系统中M个节点可理解的模型，M为大于或等于2的整数；该第一节点发送第一信息，该第一信息用于指示该公共AI模型的模型信息。
基于上述技术方案，第一节点获取用于完成该第一节点的AI任务的本地AI模型。该第一节点基于该本地AI模型确定公共AI模型。该第一节点发送用于指示该公共AI模型的模型信息的第一信息。其中，该公共AI模型为该第一节点所在的系统中M个节点可理解的模型。换言之，第一节点发送的第一信息所指示的公共AI模型为该第一节点所在的系统中M个节点可理解的模型，以便于在该M个节点中存在多种不同的节点类型的情况下，其它节点在接收该第一信息之后能够理解该公共AI模型并进一步执行模型处理。
应理解，公共AI模型为该第一节点所在的系统中M个节点可理解的模型，可以表述为公共AI模型为该第一节点所在的系统中M个节点通用的模型，公共AI模型的模型结构为该第一节点所在的系统中M个节点可理解的模型结构，公共AI模型的模型结构为该第一节点所在的系统中M个节点通用的模型结构。
在第三方面的一种可能的实现方式中，该第一节点基于该本地AI模型确定公共AI模型包括：该第一节点基于该本地AI模型和该第一节点的节点类型确定公共AI模型。
基于上述技术方案，该第一节点可以基于该本地AI模型和该第一节点的节点类型确定该公共AI模型，使得该第一节点可以基于不同的节点类型可以执行不同的AI模型处理过程，解决节点功能单一的问题，以提升灵活性。
在第三方面的一种可能的实现方式中，该节点类型包括以下任一项：基于本地数据进行本地训练的节点类型，基于其它节点的AI模型进行融合处理的节点类型，基于该本地数据进行本地训练并基于其它节点的AI模型进行融合处理的节点类型。
基于上述技术方案，用于得到该公共AI模型的节点类型可以为上述任一项实现，以提升方案实现的灵活性。
在第三方面的一种可能的实现方式中，该方法还包括：该第一节点发送指示该第一节点的节点类型的指示信息。
基于上述技术方案，该第一节点还可以发送指示该第一节点的节点类型的指示信息，以便于其它节点基于该指示信息明确该第一节点的节点类型，后续可以基于该第一节点的节点类型与该第一节点进行交互。
在第三方面的一种可能的实现方式中，该方法还包括：该第一节点基于能力信息和/或需求信息确定该第一节点的节点类型；或，该第一节点接收指示该第一节点的节点类型的指示信息。
基于上述技术方案，该第一节点可以基于自身的能力信息和/或需求信息以明确该第一节点的节点类型，也可以基于其它节点的指示以明确该第一节点的节点类型，以提升方案实现的灵活性。
在第三方面的一种可能的实现方式中，
该本地AI模型基于本地数据得到；或，
该本地AI模型基于K个信息得到，该K个信息中的每个信息指示其它节点的AI模型的模型信息和该其它节点的AI模型的辅助信息，K为正整数；或，
该本地AI模型基于该本地数据以及该K个信息得到。
基于上述技术方案，用于得到该公共AI模型的本地AI模型可以为上述任一项实现，以提升方案实现的灵活性。
在第三方面的一种可能的实现方式中，该第一信息还指示该公共AI模型的辅助信息。
基于上述技术方案，该第一节点发送的第一信息还指示该公共AI模型的辅助信息。相比于不同节点之间仅交互各自的AI模型的方式，由于第一信息指示公共AI模型的模型信息之外，该第一信息还可以指示公共AI模型的辅助信息，使得该第一信息的接收方能够基于该公共AI模型的辅助信息对该公共AI模型的模型信息进行AI模型处理(例如训练、融合等)，提升该第一信息的接收方基于该公共AI模型进行处理得到的AI模型的性能。
在第三方面的一种可能的实现方式中，该公共AI模型的辅助信息包括以下至少一项：该公共AI模型的类型信息，该第一节点的标识信息，该公共AI模型的接收节点的标识信息，该公共AI模型的版本信息，生成该公共AI模型的时间信息，生成该公共AI模型的地理位置信息，该第一节点的本地数据的分布信息。
基于上述技术方案，第一信息所指示的公共AI模型的辅助信息包括上述至少一项，以 便于第一信息的接收方能够基于上述至少一项信息对该公共AI模型的模型信息进行AI模型处理，提升该第一信息的接收方基于该公共AI模型进行处理得到的AI模型的性能。
在第三方面的一种可能的实现方式中，该公共AI模型的模型信息和该公共AI模型的辅助信息分别承载于不同的传输资源，该传输资源包括时域资源和/或频域资源。
基于上述技术方案，由于该公共AI模型的模型信息对应的数据量和该公共AI模型的辅助信息对应的数据量一般是不同的(例如前者一般多于后者)，使得两者可以分别承载于不同的传输资源上。
本申请第四方面提供了一种AI模型处理方法，该方法由第二节点执行，或者，该方法由第二节点中的部分组件(例如处理器、芯片或芯片系统等)执行，或者该方法还可以由能实现全部或部分第二节点功能的逻辑模块或软件实现。在第四方面及其可能的实现方式中，以该方法由第二节点执行为例进行描述，该第二节点可以为终端设备或网络设备。在该方法中，第二节点接收N个第一信息，该N个第一信息中的每个第一信息用于指示公共AI模型的模型信息，该公共AI模型为第二节点所在的系统中M个节点可理解的模型，M为大于或等于2的整数；该第二节点基于该N个第一信息更新本地AI模型，得到更新后的本地AI模型，该本地AI模型用于完成该第二节点的AI任务。
基于上述技术方案，第二节点接收的N个第一信息中的每个第一信息用于指示公共AI模型的模型信息，此后，该第二节点基于该N个第一信息更新本地AI模型，得到更新后的本地AI模型。其中，该公共AI模型为该第二节点所在的系统中M个节点可理解的模型。换言之，第二节点接收的第一信息所指示的公共AI模型为该第二节点所在的系统中M个节点可理解的模型，以便于在该M个节点中存在多种不同的节点类型的情况下，其它节点在接收该第一信息之后能够理解该公共AI模型并进一步执行模型处理。
应理解，公共AI模型为该第二节点所在的系统中M个节点可理解的模型，可以表述为公共AI模型为该第一节点所在的系统中M个节点通用的模型，公共AI模型的模型结构为该第一节点所在的系统中M个节点可理解的模型结构，公共AI模型的模型结构为该第一节点所在的系统中M个节点通用的模型结构。
在第四方面的一种可能的实现方式中，该第二节点基于该N个第一信息更新本地AI模型，得到更新后的本地AI模型的过程包括：该第二节点基于该N个第一信息和该第二节点的节点类型更新本地AI模型，得到更新后的本地AI模型。
基于上述技术方案，该第二节点可以基于该N个第一信息和该第二节点的节点类型确定该公共AI模型，使得该第二节点可以基于不同的节点类型可以执行不同的AI模型处理过程，解决节点功能单一的问题，以提升灵活性。
在第四方面的一种可能的实现方式中，该节点类型包括以下任一项：基于其它节点的AI模型进行融合处理的节点类型，基于该本地数据进行本地训练并基于其它节点的AI模型进行融合处理的节点类型。
基于上述技术方案，用于得到该更新后的本地AI模型的节点类型可以为上述任一项实现，以提升方案实现的灵活性。
在第四方面的一种可能的实现方式中，该方法还包括：该第二节点发送指示该第二节点的节点类型的指示信息。
基于上述技术方案，该第二节点还可以发送指示该第二节点的节点类型的指示信息，以便于其它节点基于该指示信息明确该第二节点的节点类型，后续可以基于该第二节点的节点类型与该第二节点进行交互。
在第四方面的一种可能的实现方式中，该方法还包括：该第二节点基于能力信息和/或需求信息确定该第二节点的节点类型；或，该第二节点接收指示该第二节点的节点类型的指示信息。
基于上述技术方案，该第二节点可以基于自身的能力信息和/或需求信息以明确该第二节点的节点类型，也可以基于其它节点的指示以明确该第二节点的节点类型，以提升方案实现的灵活性。
在第四方面的一种可能的实现方式中，该节点类型为基于其它节点的AI模型进行融合处理的节点类型时，该本地AI模型基于P个信息得到，该P个信息中的每个信息指示其它节点的AI模型的模型信息和该其它节点的AI模型的辅助信息，该P为正整数。
在第四方面的一种可能的实现方式中，该节点类型为基于该本地数据进行本地训练并基于其它节点的AI模型进行融合处理的节点类型时，该本地AI模型基于该本地数据得到，或，该本地AI模型基于该本地数据以及该P个信息得到，该P个信息中的每个信息指示其它节点的AI模型的模型信息和该其它节点的AI模型的辅助信息，该P为正整数。
基于上述技术方案，在节点类型不同的情况下，该本地AI模型可以为上述不同的实现，以提升方案实现的灵活性。
在第四方面的一种可能的实现方式中，该第一信息还用于指示该公共AI模型的辅助信息。
基于上述技术方案，该第二节点接收的每个第一信息还指示该公共AI模型的辅助信息。相比于不同节点之间仅交互各自的AI模型的方式，由于第一信息指示公共AI模型的模型信息之外，该第一信息还可以指示公共AI模型的辅助信息，使得该第二节点能够基于该公共AI模型的辅助信息对本地AI模型进行更新以得到更新后的本地AI模型，提升该第二节点基于该公共AI模型进行处理得到的AI模型的性能。
在第四方面的一种可能的实现方式中，该公共AI模型的辅助信息包括以下至少一项：该公共AI模型的类型信息，该第一节点的标识信息，该公共AI模型的接收节点的标识信息，该公共AI模型的版本信息，生成该公共AI模型的时间信息，生成该公共AI模型的地理位置信息，该第一节点的本地数据的分布信息。
基于上述技术方案，第一信息所指示的公共AI模型的辅助信息包括上述至少一项，以便于第二节点能够基于上述至少一项信息对该公共AI模型的模型信息进行AI模型处理，提升该第二节点基于该公共AI模型进行处理得到的AI模型的性能。
在第四方面的一种可能的实现方式中，该公共AI模型的模型信息和该公共AI模型的辅助信息分别承载于不同的传输资源，该传输资源包括时域资源和/或频域资源。
基于上述技术方案，由于该公共AI模型的模型信息对应的数据量和该公共AI模型的 辅助信息对应的数据量一般是不同的(例如前者一般多于后者)，使得两者可以分别承载于不同的传输资源上。
本申请第五方面提供了一种AI模型处理方法，该方法由第一节点执行，或者，该方法由第一节点中的部分组件(例如处理器、芯片或芯片系统等)执行，或者该方法还可以由能实现全部或部分第一节点功能的逻辑模块或软件实现。在第五方面及其可能的实现方式中，以该方法由第一节点执行为例进行描述，该第一节点可以为终端设备或网络设备。在该方法中，第一节点确定该第一节点的节点类型。
基于上述技术方案，第一节点可以确定该第一节点的节点类型，后续该第一节点可以基于节点类型执行AI模型处理。使得该第一节点可以基于不同的节点类型执行不同的AI模型处理过程，解决节点功能单一的问题，以提升灵活性。
在第五方面的一种可能的实现方式中，该第一节点发送指示该第一节点的节点类型的指示信息。
基于上述技术方案，该第一节点还可以发送指示该第一节点的节点类型的指示信息，以便于其它节点基于该指示信息明确该第一节点的节点类型，后续可以基于该第一节点的节点类型与该第一节点进行交互。
在第五方面的一种可能的实现方式中，该方法还包括：该第一节点发送第一信息，该第一信息用于指示第一AI模型的模型信息，其中，该第一AI模型为基于该第一节点的节点类型得到。
基于上述技术方案，第一节点发送的第一信息所指示的第一AI模型可以为该第一节点基于该第一节点的节点类型得到的模型。在该第一AI模型至少基于第一节点的节点类型得到的情况下，使得第一节点可以基于不同的节点类型可以执行不同的AI模型处理过程，解决节点功能单一的问题，以提升灵活性。
在第五方面的一种可能的实现方式中，该第一AI模型为基于第二AI模型和该第一节点的节点类型得到。
基于上述技术方案，在该第一AI模型至少基于第二AI模型得到的情况下，该第一节点所发送的第一信息所指示的第一AI模型可以为其它节点可理解的模型，以便于其它节点在接收该第一AI模型之后进行进一步的模型处理。
在第五方面的一种可能的实现方式中，
该第二AI模型基于本地数据得到；或，
该第二AI模型基于K个信息得到，该K个信息中的每个信息指示其它节点的AI模型的模型信息和该其它节点的AI模型的辅助信息，K为正整数；或，
该第二AI模型基于该本地数据以及该K个信息得到。
基于上述技术方案，用于得到该第一AI模型的第二AI模型可以为上述任一项实现，以提升方案实现的灵活性。
在第五方面的一种可能的实现方式中，该节点类型包括以下任一项：基于本地数据进行本地训练的节点类型，基于其它节点的AI模型进行融合处理的节点类型，基于该本地数 据进行本地训练并基于其它节点的AI模型进行融合处理的节点类型。
基于上述技术方案，用于得到该第一AI模型的节点类型可以为上述任一项实现，以提升方案实现的灵活性。
在第五方面的一种可能的实现方式中，该第一信息还用于指示该第一AI模型的辅助信息。
基于上述技术方案，该第一节点发送的第一信息还指示该第一AI模型的辅助信息。相比于不同节点之间仅交互各自的AI模型的方式，由于第一信息指示第一AI模型的模型信息之外，该第一信息还可以指示第一AI模型的辅助信息，使得该第一信息的接收方能够基于该第一AI模型的辅助信息对该第一AI模型的模型信息进行AI模型处理(例如训练、融合等)，提升该第一信息的接收方基于该第一AI模型进行处理得到的AI模型的性能。
在第五方面的一种可能的实现方式中，其中，该第一AI模型的辅助信息包括以下至少一项：该第一AI模型的类型信息，该第一节点的标识信息，该第一AI模型的接收节点的标识信息，该第一AI模型的版本信息，生成该第一AI模型的时间信息，生成该第一AI模型的地理位置信息，该第一节点的本地数据的分布信息。
基于上述技术方案，第一信息所指示的第一AI模型的辅助信息包括上述至少一项，以便于第一信息的接收方能够基于上述至少一项信息对该第一AI模型的模型信息进行AI模型处理，提升该第一信息的接收方基于该第一AI模型进行处理得到的AI模型的性能。
在第五方面的一种可能的实现方式中，该第一AI模型的模型信息和该第一AI模型的辅助信息分别承载于不同的传输资源，该传输资源包括时域资源和/或频域资源。
基于上述技术方案，由于该公共AI模型的模型信息对应的数据量和该公共AI模型的辅助信息对应的数据量一般是不同的(例如前者一般多于后者)，使得两者可以分别承载于不同的传输资源上。
在第五方面的一种可能的实现方式中，第一AI模型为该第一节点所在的系统中M个节点可理解的模型，M为大于或等于2的整数。
基于上述技术方案，第一节点发送的第一信息所指示的第一AI模型为该第一节点所在的系统中M个节点可理解的模型，以便于在该M个节点中存在多种不同的节点类型的情况下，其它节点在接收该第一信息之后能够理解该第一AI模型并进一步执行模型处理。
在第五方面的一种可能的实现方式中，该方法还包括：该第一节点基于能力信息和/或需求信息确定该第一节点的节点类型；或，该第一节点接收指示该第一节点的节点类型的指示信息。
基于上述技术方案，该第一节点可以基于自身的能力信息和/或需求信息以明确该第一节点的节点类型，也可以基于其它节点的指示以明确该第一节点的节点类型，以提升方案实现的灵活性。
本申请第六方面提供了一种AI模型处理方法，该方法由第二节点执行，或者，该方法由第二节点中的部分组件(例如处理器、芯片或芯片系统等)执行，或者该方法还可以由能实现全部或部分第二节点功能的逻辑模块或软件实现。在第六方面及其可能的实现方式 中，以该方法由第二节点执行为例进行描述，该第二节点可以为终端设备或网络设备。在该方法中，第二节点接收指示该第一节点的节点类型的指示信息；和/或，第二节点接收第一信息，该第一信息用于指示该第一AI模型的模型信息，其中，该第一AI模型为基于该第一节点的节点类型得到。
基于上述技术方案，在第二节点接收指示该第一节点的节点类型的指示信息的情况下，使得该第二节点基于该指示信息明确该第一节点的节点类型，后续该第二节点可以基于该第一节点的节点类型与该第一节点进行交互。
此外，第二节点接收用于指示该第一AI模型的模型信息的第一信息的情况下，该第一AI模型为基于该第一节点的节点类型得到。在该第一AI模型至少基于第一节点的节点类型得到时，使得第一节点可以基于不同的节点类型可以执行不同的AI模型处理过程，解决节点功能单一的问题，以提升灵活性。
在第六方面的一种可能的实现方式中，该第一AI模型为基于该第一节点的节点类型和第二AI模型得到。
基于上述技术方案，在该第一AI模型至少基于第二AI模型得到时，该第二节点所接收的第一AI模型可以为第一节点之外的其它节点可理解的模型，以便于其它节点(例如第二节点)后续在接收该目标AI模型之后进行进一步的模型处理。
在第六方面的一种可能的实现方式中，
该第二AI模型基于本地数据得到；或，
该第二AI模型基于K个信息得到，该K个信息中的每个信息指示其它节点的AI模型的模型信息和该其它节点的AI模型的辅助信息，K为正整数；或，
该第二AI模型基于该本地数据以及该K个信息得到。
基于上述技术方案，用于得到该第一AI模型的第二AI模型可以为上述任一项实现，以提升方案实现的灵活性。
在第六方面的一种可能的实现方式中，该第一节点的节点类型包括以下任一项：基于本地数据进行本地训练的节点类型，基于其它节点的AI模型进行融合处理的节点类型，基于该本地数据进行本地训练并基于其它节点的AI模型进行融合处理的节点类型。
基于上述技术方案，用于得到该第一AI模型的节点类型可以为上述任一项实现，以提升方案实现的灵活性。
可选地，该第二节点的节点类型包括以下任一项：基于其它节点的AI模型进行融合处理的节点类型，基于该本地数据进行本地训练并基于其它节点的AI模型进行融合处理的节点类型。
在第六方面的一种可能的实现方式中，该第一信息还用于指示该第一AI模型的辅助信息。
基于上述技术方案，第二节点接收的第一信息还指示该第一AI模型的辅助信息。相比于不同节点之间仅交互各自的AI模型的方式，由于第一信息指示第一AI模型的模型信息之外，该第一信息还可以指示第一AI模型的辅助信息，使得该第二节点能够基于该第一AI模型的辅助信息对该第一AI模型的模型信息进行AI模型处理(例如训练、融合等)， 提升该第二节点基于该第一AI模型进行处理得到的AI模型的性能。
在第六方面的一种可能的实现方式中，该第一AI模型的辅助信息包括以下至少一项：该第一AI模型的类型信息，该第一节点的标识信息，该第一AI模型的接收节点的标识信息，该第一AI模型的版本信息，生成该第一AI模型的时间信息，生成该第一AI模型的地理位置信息，该第一节点的本地数据的分布信息。
基于上述技术方案，第一信息所指示的第一AI模型的辅助信息包括上述至少一项，以便于第一信息的接收方能够基于上述至少一项信息对该第一AI模型的模型信息进行AI模型处理，提升该第一信息的接收方基于该第一AI模型进行处理得到的AI模型的性能。
在第六方面的一种可能的实现方式中，该第一AI模型的模型信息和该第一AI模型的辅助信息分别承载于不同的传输资源，该传输资源包括时域资源和/或频域资源。
基于上述技术方案，由于该公共AI模型的模型信息对应的数据量和该公共AI模型的辅助信息对应的数据量一般是不同的(例如前者一般多于后者)，使得两者可以分别承载于不同的传输资源上。
在第六方面的一种可能的实现方式中，第一AI模型为该第一节点所在的系统中M个节点可理解的模型，M为大于或等于2的整数。
基于上述技术方案，第二节点接收的第一信息所指示的第一AI模型为该第二节点所在的系统中M个节点可理解的模型，以便于在该M个节点中存在多种不同的节点类型的情况下，其它节点在接收该第一信息之后能够理解该第一AI模型并进一步执行模型处理。
在第六方面的一种可能的实现方式中，该方法还包括：该第二节点基于能力信息和/或需求信息确定该第二节点的节点类型；或，该第一节点接收指示该第二节点的节点类型的指示信息。
基于上述技术方案，该第二节点可以基于自身的能力信息和/或需求信息以明确该第二节点的节点类型，也可以基于其它节点的指示以明确该第二节点的节点类型，以提升方案实现的灵活性。
本申请第七方面提供了一种AI模型处理装置，该装置为第一节点，或者，该装置为第一节点中的部分组件(例如处理器、芯片或芯片系统等)，或者，该装置还可以为能够实现全部或部分第一节点功能的逻辑模块或软件。在第七方面及其可能的实现方式中，以该通信装置为第一节点执行为例进行描述，该第一节点可以为终端设备或网络设备。
该装置包括处理单元和收发单元；该处理单元用于确定第一AI模型；该收发单元用于发送第一信息，该第一信息指示该第一AI模型的模型信息和该第一AI模型的辅助信息。
在第七方面的一种可能的实现方式中，该第一AI模型的辅助信息包括以下至少一项：
该第一AI模型的类型信息，该第一节点的标识信息，该第一AI模型的接收节点的标识信息，该第一AI模型的版本信息，生成该第一AI模型的时间信息，生成该第一AI模型的地理位置信息，该第一节点的本地数据的分布信息。
在第七方面的一种可能的实现方式中，该第一AI模型的模型信息和该第一AI模型的辅助信息分别承载于不同的传输资源，该传输资源包括时域资源和/或频域资源。
在第七方面的一种可能的实现方式中，该第一AI模型为基于该第一节点的节点类型得到。
在第七方面的一种可能的实现方式中，该第一AI模型为基于第二AI模型和该第一节点的节点类型得到。
在第七方面的一种可能的实现方式中，该第二AI模型基于本地数据得到；或，该第二AI模型基于K个信息得到，该K个信息中的每个信息指示其它节点的AI模型的模型信息和该其它节点的AI模型的辅助信息，K为正整数；或，该第二AI模型基于该本地数据以及该K个信息得到。
在第七方面的一种可能的实现方式中，该节点类型包括以下任一项：基于本地数据进行本地训练的节点类型，基于其它节点的AI模型进行融合处理的节点类型，基于该本地数据进行本地训练并基于其它节点的AI模型进行融合处理的节点类型。
在第七方面的一种可能的实现方式中，该收发单元还用于发送指示该第一节点的节点类型的指示信息。
在第七方面的一种可能的实现方式中，该处理单元还用于基于能力信息和/或需求信息确定该第一节点的节点类型；或，该收发单元还用于接收指示该第一节点的节点类型的指示信息。
在第七方面的一种可能的实现方式中，第一AI模型为该第一节点所在的系统中M个节点可理解的模型，M为大于或等于2的整数。
本申请实施例第七方面中，通信装置的组成模块还可以用于执行第一方面的各个可能实现方式中所执行的步骤，并实现相应的技术效果，具体均可以参阅第一方面，此处不再赘述。
本申请第八方面提供了一种AI模型处理装置，该装置为第二节点，或者，该装置为第二节点中的部分组件(例如处理器、芯片或芯片系统等)，或者，该装置还可以为能够实现全部或部分第二节点功能的逻辑模块或软件。在第八方面及其可能的实现方式中，以该通信装置为第二节点执行为例进行描述，该第二节点可以为终端设备或网络设备。
该装置包括处理单元和收发单元；该收发单元用于接收N个第一信息，该N个第一信息中的每个第一信息指示第一AI模型的模型信息和该第一AI模型的辅助信息，N为正整数；该处理单元用于基于该N个第一信息进行模型处理，得到目标AI模型。
可选地，该目标AI模型用于完成该第二节点的AI任务，或，该目标AI模型为该第二节点的本地模型。
在第八方面的一种可能的实现方式中，该第一AI模型的辅助信息包括以下至少一项：
该第一AI模型的类型信息，第一节点的标识信息，该第一AI模型的接收节点的标识信息，该第一AI模型的版本信息，生成该第一AI模型的时间信息，生成该第一AI模型的地理位置信息，该第一节点的本地数据的分布信息。
在第八方面的一种可能的实现方式中，该第一AI模型的模型信息和该第一AI模型的辅助信息分别承载于不同的传输资源，该传输资源包括时域资源和/或频域资源。
在第八方面的一种可能的实现方式中，该处理单元具体用于基于该N个第一信息以及该第二节点的节点类型进行模型处理，得到该目标AI模型。
在第八方面的一种可能的实现方式中，在该第二节点的节点类型为基于其它节点的AI模型进行融合处理的节点类型时，该处理单元具体用于基于该N个第一信息对该N个第一AI模型进行模型融合，得到该目标AI模型。
在第八方面的一种可能的实现方式中，在该第二节点的节点类型为基于该本地数据进行本地训练并基于其它节点的AI模型进行融合处理的节点类型时，该处理单元具体用于基于该N个第一信息对该N个第一AI模型以及第二AI模型进行模型融合，得到该目标AI模型，其中，该第二AI模型为基于本地数据训练得到。
在第八方面的一种可能的实现方式中，该收发单元还用于接收用于指示该第一节点的节点类型的指示信息。
在第八方面的一种可能的实现方式中，该收发单元还用于发送用于指示该第二节点的节点类型的指示信息。
在第八方面的一种可能的实现方式中，该处理单元还用于基于能力信息和/或需求信息确定该第二节点的节点类型；或，该收发单元还用于接收指示该第二节点的节点类型的指示信息。
在第八方面的一种可能的实现方式中，第一AI模型为该第二节点所在的系统中M个节点可理解的模型，M为大于或等于2的整数。
本申请实施例第八方面中，通信装置的组成模块还可以用于执行第二方面的各个可能实现方式中所执行的步骤，并实现相应的技术效果，具体均可以参阅第二方面，此处不再赘述。
本申请第九方面提供了一种AI模型处理装置，该装置为第一节点，或者，该装置为第一节点中的部分组件(例如处理器、芯片或芯片系统等)，或者，该装置还可以为能够实现全部或部分第一节点功能的逻辑模块或软件。在第九方面及其可能的实现方式中，以该通信装置为第一节点执行为例进行描述，该第一节点可以为终端设备或网络设备。
该装置包括处理单元和收发单元；该处理单元用于获取本地AI模型，该本地AI模型用于完成该第一节点的AI任务；该处理单元还用于基于该本地AI模型确定公共AI模型，该公共AI模型为该第一节点所在的系统中M个节点可理解的模型，M为大于或等于2的整数；该收发单元用于发送第一信息，该第一信息用于指示该公共AI模型的模型信息。
在第九方面的一种可能的实现方式中，该处理单元具体用于基于该本地AI模型和该第一节点的节点类型确定公共AI模型。
在第九方面的一种可能的实现方式中，该节点类型包括以下任一项：基于本地数据进行本地训练的节点类型，基于其它节点的AI模型进行融合处理的节点类型，基于该本地数据进行本地训练并基于其它节点的AI模型进行融合处理的节点类型。
在第九方面的一种可能的实现方式中，该收发单元还用于发送指示该第一节点的节点类型的指示信息。
在第九方面的一种可能的实现方式中，该处理单元还用于基于能力信息和/或需求信息确定该第一节点的节点类型；或，该收发单元还用于接收指示该第一节点的节点类型的指示信息。
在第九方面的一种可能的实现方式中，
该本地AI模型基于本地数据得到；或，
该本地AI模型基于K个信息得到，该K个信息中的每个信息指示其它节点的AI模型的模型信息和该其它节点的AI模型的辅助信息，K为正整数；或，
该本地AI模型基于该本地数据以及该K个信息得到。
在第九方面的一种可能的实现方式中，该第一信息还指示该公共AI模型的辅助信息。
在第九方面的一种可能的实现方式中，该公共AI模型的辅助信息包括以下至少一项：
该公共AI模型的类型信息，该第一节点的标识信息，该公共AI模型的接收节点的标识信息，该公共AI模型的版本信息，生成该公共AI模型的时间信息，生成该公共AI模型的地理位置信息，该公共节点的本地数据的分布信息。
在第九方面的一种可能的实现方式中，该公共AI模型的模型信息和该公共AI模型的辅助信息分别承载于不同的传输资源，该传输资源包括时域资源和/或频域资源。
本申请实施例第九方面中，通信装置的组成模块还可以用于执行第三方面的各个可能实现方式中所执行的步骤，并实现相应的技术效果，具体均可以参阅第三方面，此处不再赘述。
本申请第十方面提供了一种AI模型处理装置，该装置为第二节点，或者，该装置为第二节点中的部分组件(例如处理器、芯片或芯片系统等)，或者，该装置还可以为能够实现全部或部分第二节点功能的逻辑模块或软件。在第十方面及其可能的实现方式中，以该通信装置为第二节点执行为例进行描述，该第二节点可以为终端设备或网络设备。
该装置包括处理单元和收发单元；该收发单元用于接收N个第一信息，该N个第一信息中的每个第一信息用于指示公共AI模型的模型信息，该公共AI模型为第二节点所在的系统中M个节点可理解的模型，M为大于或等于2的整数；该处理单元用于基于该N个第一信息更新本地AI模型，得到更新后的本地AI模型，该本地AI模型用于完成该第二节点的AI任务。
在第十方面的一种可能的实现方式中，该处理单元具体用于基于该N个第一信息和该第二节点的节点类型更新本地AI模型，得到更新后的本地AI模型。
在第十方面的一种可能的实现方式中，该节点类型包括以下任一项：基于其它节点的AI模型进行融合处理的节点类型，基于该本地数据进行本地训练并基于其它节点的AI模型进行融合处理的节点类型。
在第十方面的一种可能的实现方式中，该收发单元还用于发送指示该第二节点的节点类型的指示信息。
在第十方面的一种可能的实现方式中，该处理单元还用于基于能力信息和/或需求信息确定该第二节点的节点类型；或，该收发单元还用于接收指示该第二节点的节点类型的指 示信息。
在第十方面的一种可能的实现方式中，该节点类型为基于其它节点的AI模型进行融合处理的节点类型时，该本地AI模型基于P个信息得到，该P个信息中的每个信息指示其它节点的AI模型的模型信息和该其它节点的AI模型的辅助信息，该P为正整数。
在第十方面的一种可能的实现方式中，该节点类型为基于该本地数据进行本地训练并基于其它节点的AI模型进行融合处理的节点类型时，该本地AI模型基于该本地数据得到，或，该本地AI模型基于该本地数据以及该P个信息得到，该P个信息中的每个信息指示其它节点的AI模型的模型信息和该其它节点的AI模型的辅助信息，该P为正整数。
在第十方面的一种可能的实现方式中，该第一信息还用于指示该公共AI模型的辅助信息。
在第十方面的一种可能的实现方式中，该公共AI模型的辅助信息包括以下至少一项：该公共AI模型的类型信息，该第一节点的标识信息，该公共AI模型的接收节点的标识信息，该公共AI模型的版本信息，生成该公共AI模型的时间信息，生成该公共AI模型的地理位置信息，该第一节点的本地数据的分布信息。
在第十方面的一种可能的实现方式中，该公共AI模型的模型信息和该公共AI模型的辅助信息分别承载于不同的传输资源，该传输资源包括时域资源和/或频域资源。
本申请实施例第十方面中，通信装置的组成模块还可以用于执行第四方面的各个可能实现方式中所执行的步骤，并实现相应的技术效果，具体均可以参阅第四方面，此处不再赘述。
本申请第十一方面提供了一种AI模型处理装置，该装置为第一节点，或者，该装置为第一节点中的部分组件(例如处理器、芯片或芯片系统等)，或者，该装置还可以为能够实现全部或部分第一节点功能的逻辑模块或软件。在第十一方面及其可能的实现方式中，以该通信装置为第一节点执行为例进行描述，该第一节点可以为终端设备或网络设备。
该装置包括处理单元；该处理单元用于第一节点确定该第一节点的节点类型。
在第十一方面的一种可能的实现方式中，该装置还包括收发单元，该收发单元用于发送指示该第一节点的节点类型的指示信息。
在第十一方面的一种可能的实现方式中，该装置还包括收发单元，该收发单元用于发送第一信息，该第一信息用于指示该第一AI模型的模型信息，其中，该第一AI模型为基于该第一节点的节点类型得到。
可选地，该第一AI模型为基于第二AI模型和该第一节点的节点类型得到。
在第十一方面的一种可能的实现方式中，
该第二AI模型基于本地数据得到；或，
该第二AI模型基于K个信息得到，该K个信息中的每个信息指示其它节点的AI模型的模型信息和该其它节点的AI模型的辅助信息，K为正整数；或，
该第二AI模型基于该本地数据以及该K个信息得到。
在第十一方面的一种可能的实现方式中，该节点类型包括以下任一项：基于本地数据 进行本地训练的节点类型，基于其它节点的AI模型进行融合处理的节点类型，基于该本地数据进行本地训练并基于其它节点的AI模型进行融合处理的节点类型。
在第十一方面的一种可能的实现方式中，该第一信息还用于指示该第一AI模型的辅助信息，其中，该第一AI模型的辅助信息包括以下至少一项：该第一AI模型的类型信息，该第一节点的标识信息，该第一AI模型的接收节点的标识信息，该第一AI模型的版本信息，生成该第一AI模型的时间信息，生成该第一AI模型的地理位置信息，该第一节点的本地数据的分布信息。
在第十一方面的一种可能的实现方式中，该第一AI模型的模型信息和该第一AI模型的辅助信息分别承载于不同的传输资源，该传输资源包括时域资源和/或频域资源。
在第十一方面的一种可能的实现方式中，第一AI模型为该第一节点所在的系统中M个节点可理解的模型，M为大于或等于2的整数。
在第十一方面的一种可能的实现方式中，该处理单元还用于基于能力信息和/或需求信息确定该第一节点的节点类型；或，该收发单元还用于接收指示该第一节点的节点类型的指示信息。
本申请实施例第十一方面中，通信装置的组成模块还可以用于执行第五方面的各个可能实现方式中所执行的步骤，并实现相应的技术效果，具体均可以参阅第五方面，此处不再赘述。
本申请第十二方面提供了一种AI模型处理装置，该装置为第二节点，或者，该装置为第二节点中的部分组件(例如处理器、芯片或芯片系统等)，或者，该装置还可以为能够实现全部或部分第二节点功能的逻辑模块或软件。在第十二方面及其可能的实现方式中，以该通信装置为第二节点执行为例进行描述，该第二节点可以为终端设备或网络设备。
该装置包括收发单元；
该收发单元用于第二节点接收指示该第一节点的节点类型的指示信息；
和/或，
该收发单元用于接收第一信息，该第一信息用于指示该第一AI模型的模型信息，其中，该第一AI模型为基于该第一节点的节点类型得到。
可选地，该第一AI模型为基于第二AI模型和该第一节点的节点类型得到。
在第十二方面的一种可能的实现方式中，
该第二AI模型基于本地数据得到；或，
该第二AI模型基于K个信息得到，该K个信息中的每个信息指示其它节点的AI模型的模型信息和该其它节点的AI模型的辅助信息，K为正整数；或，
该第二AI模型基于该本地数据以及该K个信息得到。
在第十二方面的一种可能的实现方式中，该节点类型包括以下任一项：基于本地数据进行本地训练的节点类型，基于其它节点的AI模型进行融合处理的节点类型，基于该本地数据进行本地训练并基于其它节点的AI模型进行融合处理的节点类型。
在第十二方面的一种可能的实现方式中，该第一信息还用于指示该第一AI模型的辅助 信息，其中，该第一AI模型的辅助信息包括以下至少一项：该第一AI模型的类型信息，该第一节点的标识信息，该第一AI模型的接收节点的标识信息，该第一AI模型的版本信息，生成该第一AI模型的时间信息，生成该第一AI模型的地理位置信息，该第一节点的本地数据的分布信息。
在第十二方面的一种可能的实现方式中，该第一AI模型的模型信息和该第一AI模型的辅助信息分别承载于不同的传输资源，该传输资源包括时域资源和/或频域资源。
在第十二方面的一种可能的实现方式中，第一AI模型为该第一节点所在的系统中M个节点可理解的模型，M为大于或等于2的整数。
在第十二方面的一种可能的实现方式中，该装置还包括处理单元，该处理单元用于基于能力信息和/或需求信息确定该第二节点的节点类型；或，该收发单元还用于接收指示该第二节点的节点类型的指示信息。
本申请实施例第十二方面中，通信装置的组成模块还可以用于执行第六方面的各个可能实现方式中所执行的步骤，并实现相应的技术效果，具体均可以参阅第六方面，此处不再赘述。
本申请实施例第十三方面提供了一种通信装置，包括至少一个处理器，所述至少一个处理器与存储器耦合；
该存储器用于存储程序或指令；
该至少一个处理器用于执行该程序或指令，以使该装置实现前述第一方面或第一方面任意一种可能的实现方式所述的方法，或者，以使该装置实现前述第二方面或第二方面任意一种可能的实现方式所述的方法，或者，以使该装置实现前述第三方面或第三方面任意一种可能的实现方式所述的方法，或者，以使该装置实现前述第四方面或第四方面任意一种可能的实现方式所述的方法，或者，以使该装置实现前述第五方面或第五方面任意一种可能的实现方式所述的方法，或者，以使该装置实现前述第六方面或第六方面任意一种可能的实现方式所述的方法。
本申请实施例第十四方面提供了一种通信装置，包括至少一个逻辑电路和输入输出接口；
该逻辑电路用于执行如前述第一方面或第一方面任意一种可能的实现方式所述的方法，或，该逻辑电路用于执行如前述第二方面或第二方面任意一种可能的实现方式所述的方法，或，该逻辑电路用于执行如前述第三方面或第三方面任意一种可能的实现方式所述的方法，或，该逻辑电路用于执行如前述第四方面或第四方面任意一种可能的实现方式所述的方法，或，该逻辑电路用于执行如前述第五方面或第五方面任意一种可能的实现方式所述的方法，或，该逻辑电路用于执行如前述第六方面或第六方面任意一种可能的实现方式所述的方法。
本申请实施例第十五方面提供一种存储一个或多个计算机执行指令的计算机可读存储介质，当计算机执行指令被处理器执行时，该处理器执行如上述第一方面或第一方面任意一种可能的实现方式所述的方法，或，该处理器执行如上述第二方面或第二方面任意一种可能的实现方式所述的方法，或，该处理器执行如上述第三方面或第三方面任意一种可能的实现方式所述的方法，或，该处理器执行如上述第四方面或第四方面任意一种可能的实 现方式所述的方法，或，该处理器执行如上述第五方面或第五方面任意一种可能的实现方式所述的方法，或，该处理器执行如上述第六方面或第六方面任意一种可能的实现方式所述的方法。
本申请实施例第十六方面提供一种存储一个或多个计算机的计算机程序产品(或称计算机程序)，当计算机程序产品被该处理器执行时，该处理器执行上述第一方面或第一方面任意一种可能实现方式的方法，或，该处理器执行上述第二方面或第二方面任意一种可能实现方式的方法，或，该处理器执行上述第三方面或第三方面任意一种可能实现方式的方法，或，该处理器执行上述第四方面或第四方面任意一种可能实现方式的方法，或，该处理器执行上述第五方面或第五方面任意一种可能实现方式的方法，或，该处理器执行上述第六方面或第六方面任意一种可能实现方式的方法。
本申请实施例第十七方面提供了一种芯片系统，该芯片系统包括至少一个处理器，用于支持通信装置实现上述第一方面或第一方面任意一种可能的实现方式中所涉及的功能，或，用于支持通信装置实现上述第二方面或第二方面任意一种可能的实现方式中所涉及的功能，或，用于支持通信装置实现上述第三方面或第三方面任意一种可能的实现方式中所涉及的功能，或，用于支持通信装置实现上述第四方面或第四方面任意一种可能的实现方式中所涉及的功能，或，用于支持通信装置实现上述第五方面或第五方面任意一种可能的实现方式中所涉及的功能，或，用于支持通信装置实现上述第六方面或第六方面任意一种可能的实现方式中所涉及的功能。
在一种可能的设计中，该芯片系统还可以包括存储器，存储器，用于保存该第一通信装置必要的程序指令和数据。该芯片系统，可以由芯片构成，也可以包含芯片和其他分立器件。可选的，所述芯片系统还包括接口电路，所述接口电路为所述至少一个处理器提供程序指令和/或数据。
本申请实施例第十八方面提供了一种通信系统，该通信系统包括上述第三方面的通信装置和第四方面的通信装置，和/或，该通信系统包括上述第五方面的通信装置和第六方面的通信装置，和/或，该通信系统包括上述第七方面的通信装置和第八方面的通信装置，和/或，该通信系统包括上述第九方面的通信装置和第十方面的通信装置，和/或，该通信系统包括上述第十一方面的通信装置和第十二方面的通信装置，和/或，该通信系统包括上述第十三方面的通信装置，和/或，该通信系统包括上述第十四方面的通信装置。
其中，第七方面至第十八方面中任一种设计方式所带来的技术效果可参见上述第一方面至第六方面中不同设计方式所带来的技术效果，在此不再赘述。
从以上技术方案可以看出，本申请提供的方案具备以下有益效果：
在一些实施例中，第一节点在确定第一AI模型之后，该第一节点发送指示该第一AI模型的模型信息和该第一AI模型的辅助信息的第一信息。相比于不同节点之间仅交互各自的AI模型的方式，由于第一信息指示第一AI模型的模型信息之外，该第一信息还可以指示第一AI模型的辅助信息，使得该第一信息的接收方能够基于该第一AI模型的辅助信息对该第一AI模型的模型信息进行AI模型处理(例如训练、融合等)，提升该第一信息的接收方基于该第一AI模型进行处理得到的AI模型的性能。
在另一些实施例中，第一节点获取用于完成该第一节点的AI任务的本地AI模型，该第一节点基于该本地AI模型确定公共AI模型，该第一节点发送用于指示该公共AI模型的模型信息的第一信息。其中，该公共AI模型为该第一节点所在的系统中M个节点可理解的模型。换言之，第一节点发送的第一信息所指示的公共AI模型为该第一节点所在的系统中M个节点可理解的模型，以便于在该M个节点中存在多种不同的节点类型的情况下，其它节点在接收该第一信息之后能够理解该第一AI模型并进一步执行模型处理。
在另一些实施例中，第一节点可以确定该第一节点的节点类型，后续该第一节点可以基于节点类型执行AI模型处理。使得该第一节点可以基于不同的节点类型可以执行不同的AI模型处理过程，解决节点功能单一的问题，以提升灵活性。
附图说明
图1a为本申请提供的通信系统的一个示意图；
图1b为本申请提供的通信系统的另一个示意图；
图1c为本申请提供的通信系统的另一个示意图；
图2a为本申请涉及的AI处理过程的一个示意图；
图2b为本申请涉及的AI处理过程的另一个示意图；
图2c为本申请涉及的AI处理过程的另一个示意图；
图2d为本申请涉及的AI处理过程的另一个示意图；
图2e为本申请涉及的AI处理过程的另一个示意图；
图3a为基于联邦学习实现的AI处理过程的一个示意图；
图3b为基于分布式学习实现的AI处理过程的一个示意图；
图4为本申请提供的AI模型处理方法的一个交互示意图；
图5a为本申请提供的AI模型处理方法的一个示意图；
图5b为本申请提供的AI模型处理方法的另一个示意图；
图5c为本申请提供的AI模型处理方法的另一个示意图；
图5d为本申请提供的AI模型处理方法的另一个示意图；
图6为本申请提供的AI模型处理方法的另一个交互示意图；
图7为本申请提供的AI模型处理方法的另一个交互示意图；
图8为本申请提供的通信装置的一个示意图；
图9为本申请提供的通信装置的另一个示意图；
图10为本申请提供的通信装置的另一个示意图；
图11为本申请提供的通信装置的另一个示意图。
具体实施方式
下面将结合本申请实施例中的附图，对本申请实施例中的技术方案进行描述。基于本申请中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都属于本申请保护的范围。
首先，对本申请实施例中的部分用语进行解释说明，以便于本领域技术人员理解。
(1)终端设备：可以是能够接收网络设备调度和指示信息的无线终端设备，无线终端设备可以是指向用户提供语音和/或数据连通性的设备，或具有无线连接功能的手持式设备，或连接到无线调制解调器的其他处理设备。
终端设备可以经RAN与一个或多个核心网或者互联网进行通信，终端设备可以是移动终端设备，如移动电话(或称为“蜂窝”电话，手机(mobile phone))、计算机和数据卡，例如，可以是便携式、袖珍式、手持式、计算机内置的或者车载的移动装置，它们与无线接入网交换语音和/或数据。例如，个人通信业务(personal communication service，PCS)电话、无绳电话、会话发起协议(SIP)话机、无线本地环路(wireless local loop，WLL)站、个人数字助理(personal digital assistant，PDA)、平板电脑(Pad)、带无线收发功能的电脑等设备。无线终端设备也可以称为系统、订户单元(subscriber unit)、订户站(subscriber station)，移动站(mobile station)、移动台(mobile station，MS)、远程站(remote station)、接入点(access point，AP)、远程终端设备(remote terminal)、接入终端设备(access terminal)、用户终端设备(user terminal)、用户代理(user agent)、用户站(subscriber station，SS)、用户端设备(customer premises equipment，CPE)、终端(terminal)、用户设备(user equipment，UE)、移动终端(mobile terminal，MT)等。
作为示例而非限定，在本申请实施例中，该终端设备还可以是可穿戴设备。可穿戴设备也可以称为穿戴式智能设备或智能穿戴式设备等，是应用穿戴式技术对日常穿戴进行智能化设计、开发出可以穿戴的设备的总称，如眼镜、手套、手表、服饰及鞋等。可穿戴设备即直接穿在身上，或是整合到用户的衣服或配件的一种便携式设备。可穿戴设备不仅仅是一种硬件设备，更是通过软件支持以及数据交互、云端交互来实现强大的功能。广义穿戴式智能设备包括功能全、尺寸大、可不依赖智能手机实现完整或者部分的功能，例如：智能手表或智能眼镜等，以及只专注于某一类应用功能，需要和其它设备如智能手机配合使用，如各类进行体征监测的智能手环、智能头盔、智能首饰等。
终端还可以是无人机、机器人、设备到设备通信(device-to-device，D2D)中的终端、车到一切(vehicle to everything，V2X)中的终端、虚拟现实(virtual reality，VR)终端设备、增强现实(augmented reality，AR)终端设备、工业控制(industrial control)中的无线终端、无人驾驶(self driving)中的无线终端、远程医疗(remote medical)中的无线终端、智能电网(smart grid)中的无线终端、运输安全(transportation safety)中的无线终端、智慧城市(smart city)中的无线终端、智慧家庭(smart home)中的无线终端等。
此外，终端设备也可以是第五代(5th generation，5G)通信系统之后演进的通信系统(例如第六代(6th generation，6G)通信系统等)中的终端设备或者未来演进的公共陆地移动网络(public land mobile network，PLMN)中的终端设备等。示例性的，6G网络可以进一步扩展5G通信终端的形态和功能，6G终端包括但不限于车、蜂窝网络终端(融合卫星终端功能)、无人机、物联网(internet of things，IoT)设备。
在本申请实施例中，上述终端设备还可以获得网络设备提供的AI服务。可选地，终端设备还可以具有AI处理能力。
(2)网络设备：可以是无线网络中的设备，例如网络设备可以为将终端设备接入到无线网络的RAN节点(或设备)，又可以称为基站。目前，一些RAN设备的举例为：5G通信系统中的基站gNB(gNodeB)、传输接收点(transmission reception point，TRP)、演进型节点B(evolved Node B，eNB)、无线网络控制器(radio network controller，RNC)、节点B(Node B，NB)、家庭基站(例如，home evolved Node B，或home Node B，HNB)、基带单元(base band unit，BBU)，或无线保真(wireless fidelity，Wi-Fi)接入点AP等。另外，在一种网络结构中，网络设备可以包括集中单元(centralized unit，CU)节点、或分布单元(distributed unit，DU)节点、或包括CU节点和DU节点的RAN设备。
网络设备可以是其它为终端设备提供无线通信功能的装置。本申请的实施例对网络设备所采用的具体技术和具体设备形态不做限定。为方便描述，本申请实施例并不限定。
网络设备还可以包括核心网设备，核心网设备例如包括第四代(4th generation，4G)网络中的移动性管理实体(mobility management entity，MME)，归属用户服务器(home subscriber server，HSS)，服务网关(serving gateway，S-GW)，策略和计费规则功能(policy and charging rules function，PCRF)，公共数据网网关(public data network gateway，PDN gateway，P-GW)；5G网络中的访问和移动管理功能(access and mobility management function，AMF)、用户面功能(user plane function，UPF)或会话管理功能(session management function，SMF)等网元。此外，该核心网设备还可以包括5G网络以及5G网络的下一代网络中的其他核心网设备。
本申请实施例中，上述网络设备还可以具有AI能力的网络节点，可以为终端或其他网络设备提供AI服务，例如，可以为网络侧(接入网或核心网)的AI节点、算力节点、具有AI能力的RAN节点、具有AI能力的核心网网元等。
本申请实施例中，用于实现网络设备的功能的装置可以是网络设备，也可以是能够支持网络设备实现该功能的装置，例如芯片系统，该装置可以被安装在网络设备中。在本申请实施例提供的技术方案中，以用于实现网络设备的功能的装置是网络设备为例，描述本申请实施例提供的技术方案。
(3)配置与预配置：在本申请中，会同时用到配置与预配置。其中，配置是指网络设备/服务器通过消息或信令将一些参数的配置信息或参数的取值发送给终端，以便终端根据这些取值或信息来确定通信的参数或传输时的资源。预配置与配置类似，可以是网络设备/服务器预先与终端设备协商好的参数信息或参数值，也可以是标准协议规定的基站/网络设备或终端设备采用的参数信息或参数值，还可以是预先存储在基站/服务器或终端设备的参数信息或参数值。本申请对此不做限定。
进一步地，这些取值和参数，是可以变化或更新的。
(4)本申请实施例中的术语“系统”和“网络”可被互换使用。“多个”是指两个或两个以上。“和/或”，描述关联对象的关联关系，表示可以存在三种关系，例如，A和/或B，可以表示：单独存在A、同时存在A和B、单独存在B的情况，其中A，B可以是单数或者 复数。字符“/”一般表示前后关联对象是一种“或”的关系。“以下至少一项(个)”或其类似表达，是指的这些项中的任意组合，包括单项(个)或复数项(个)的任意组合。例如“A，B和C中的至少一项”包括A，B，C，AB，AC，BC或ABC。以及，除非有特别说明，本申请实施例提及“第一”、“第二”等序数词是用于对多个对象进行区分，不用于限定多个对象的顺序、时序、优先级或者重要程度。
本申请中，除特殊说明外，各个实施例之间相同或相似的部分可以互相参考。在本申请中各个实施例、以及各实施例中的各个方法/设计/实现方式中，如果没有特殊说明以及逻辑冲突，不同的实施例之间、以及各实施例中的各个方法/设计/实现方式之间的术语和/或描述具有一致性、且可以相互引用，不同的实施例、以及各实施例中的各个方法/设计/实现方式中的技术特征根据其内在的逻辑关系可以组合形成新的实施例、方法、或实现方式。以下所述的本申请实施方式并不构成对本申请保护范围的限定。
本申请可以应用于长期演进(long term evolution，LTE)系统、新无线(new radio，NR)系统，或者是5G之后演进的通信系统(例如6G等)。其中，该通信系统中包括至少一个网络设备和/或至少一个终端设备。
请参阅图1a，为本申请中通信系统的一种示意图。图1a中，示例性的示出了一个网络设备101和6个终端设备，6个终端设备分别为终端设备1、终端设备2、终端设备3、终端设备4、终端设备5以及终端设备6等。在图1a所示的示例中，是以终端设备1为智能茶杯，终端设备2为智能空调，终端设备3为智能加油机，终端设备4为交通工具，终端设备5为手机，终端设备6为打印机进行举例说明的。
如图1a所示，AI配置信息发送实体可以为网络设备。AI配置信息接收实体可以为终端设备1-终端设备6，此时，网络设备和终端设备1-终端设备6组成一个通信系统，在该通信系统中，终端设备1-终端设备6可以发送数据给网络设备，网络设备需要接收终端设备1-终端设备6发送的数据。同时，网络设备可以向终端设备1-终端设备6发送配置信息。
可选地，该AI配置信息可以包括后文提及的节点类型的指示信息；该数据可以包括后文提及的AI模型的模型信息和/或AI模型的辅助信息。
示例性的，在图1a中，终端设备4-终端设备6也可以组成一个通信系统。其中，终端设备5作为网络设备，即AI配置信息发送实体；终端设备4和终端设备6作为终端设备，即AI配置信息接收实体。例如车联网系统中，终端设备5分别向终端设备4和终端设备6发送AI配置信息，并且接收终端设备4和终端设备6发送的数据；相应的，终端设备4和终端设备6接收终端设备5发送的AI配置信息，并向终端设备5发送数据。
以图1a所示通信系统为例，不同的设备之间(包括网络设备与网络设备之间，网络设备与终端设备之间，和/或，终端设备和终端设备之间)除了执行通信相关业务之外，还有可能执行AI相关业务。例如，如图1b所示，以网络设备为基站为例，基站可以与一个或多个终端设备之间可以执行通信相关业务和AI相关业务，不同终端设备之间也可以执行通信相关业务和AI相关业务。又如，如图1c所示，以终端设备包括电视和手机为例，电视和手机之间也可以执行通信相关业务和AI相关业务。
随着大数据时代的到来，每台设备(包括网络设备或终端设备)都会以各种形式产生巨量的原始数据，这些数据将以“孤岛”的形式诞生并存在于世界的各个角落。传统的集中式学习要求各个分布式设备将本地数据统一传输到中心端的服务器上，其后再利用收集到的数据进行模型的训练与学习，然而这一架构随着时代的发展逐渐受到如下因素的限制：
(1)分布式设备广泛地分布于世界上各个地区和角落，这些设备将以飞快的速度源源不断地产生和积累巨大量级的原始数据。若中心端需要收集来自全部分布式设备的原始数据，势必会带来巨大的通信损耗和算力需求。
(2)随着现实生活中实际场景的复杂化，越来越多的学习任务要求分布式设备能够做出及时而有效的决策与反馈。传统的集中式学习由于涉及到大量数据的上传势必会导致较大程度的时延，致使其无法满足实际任务场景的实时需求。
(3)考虑到行业竞争、用户隐私安全、行政手续复杂等问题，将数据进行集中整合将面临越来越大的阻力制约。因而系统部署将越来越倾向于在本地存储数据，同时由分布式设备自身完成模型的本地计算。
因此，如何在满足数据隐私、安全和监管要求的前提下，设计一个机器学习框架，让AI系统能够更加高效、准确地共同使用各自的数据，成为了当前人工智能发展的一个重要议题。机器学习时人工智能领域一个重要的研究方向，而其中关于神经网络的研究则是近年来的热点。下面将本发明中可能涉及到的神经网络进行简要介绍。
1.全连接神经网络。
又叫多层感知机(multilayer perceptron，MLP)。如图2a所示，一个MLP包含一个输入层(左侧)，一个输出层(右侧)，及多个隐藏层(中间)。其中，MLP的每层包含若干个节点，称为神经元。其中,相邻两层的神经元间两两相连。
可选的，考虑相邻两层的神经元，下一层的神经元的输出h为所有与之相连的上一层神经元x的加权和并经过激活函数，可以表示为：
h＝f(wx+b)。
其中，w为权重矩阵，b为偏置向量，f为激活函数。
进一步可选的，神经网络的输出可以递归表达为：
y＝f n(w nf n-1(…)+b n)。
其中,n是神经网络层的索引，1<＝n<＝N,其中N为神经网络的总层数。
换言之，可以将神经网络理解为一个从输入数据集合到输出数据集合的映射关系。而通常神经网络都是随机初始化的，用已有数据从随机的w和b得到这个映射关系的过程被称为神经网络的训练。
可选的，训练的具体方式为采用损失函数(loss function)对神经网络的输出结果进行评价。如图2b所示，可以将误差反向传播，通过梯度下降的方法即能迭代优化神经网络参数(包括w和b)，直到损失函数达到最小值，即图2b中的“最优点”。可以理解的是，图2b中的“最优点”对应的神经网络参数可以作为训练好的AI模型信息中的神经网络参数。
进一步可选的，梯度下降的过程可以表示为：
Figure PCTCN2022110616-appb-000001
其中，θ为待优化参数(包括w和b)，L为损失函数，η为学习率，控制梯度下降的步长，
Figure PCTCN2022110616-appb-000002
表示求导运算，
Figure PCTCN2022110616-appb-000003
表示对L求θ的导数。
进一步可选的，反向传播的过程利用到求偏导的链式法则。如图2c所示，前一层参数梯度可以由后一层参数的梯度递推计算得到，可以表达为：
Figure PCTCN2022110616-appb-000004
其中，w ij为节点j连接节点i的权重，s i为节点i上的输入加权和。
2.生成对抗网络(Generative Adversarial Network，GAN)。
生成模型(Generative models,GM)是近年来计算机视觉领域中一个快速发展的研究方向。2014年，Ian Goodfellow提出了一种基于概率分布以用于生成拟真数据的的生成模型，取名为生成对抗网络(Generative adversarial network,GAN)。因其具有近似复杂概率密度函数的能力，生成对抗网络已被证实可用于包括图像生成、视频生成和自然语言处理在内的各类机器学习任务。
如图2d所示，生成对抗网络通常由至少一个生成器(Generator)与至少一个判别器(Discriminator)组成。生成器从一个潜在空间中随机取样作为输入，经过处理后输出虚假样本，并要求输出的虚假样本尽可能拟真训练集中的真实样本。判别器的输入为真实样本或生成器的虚假样本，并输出一个输入样本属于真实样本的概率值。判别器旨在将生成器所产生的虚假样本从真实样本中分辨出来，而生成器则需要尽可能去欺骗判别器，使其无法判别出输入样本的真假。通过生成器和判别器不断地相互对抗，并调整各自的模型参数，最终到达使判别器无法判断生成器的输出样本是否属于真实样本。
3.自编码器。
自编码器是一种人工神经网络结构，其示意图如图2e所示。一般用于对一组数据进行表示(也称编码)，通常用于降维，其包含编码器和译码器部分。编码器将输入数据通过一或多层神经网络进行编码，得到降维后的码字。译码器通过一层或多层神经网络，将码字重建成输出数据，该输出数据要求尽可能的与编码器输入数据相同。编码器和译码器同时训练，以实现上述要求。
近年来，AI技术在机器视觉、自然语言处理等领域取得重大进展，并且在实际生活中逐渐开始普及。可以预见，AI将在各种连接设备(如终端、边缘)中无处不在。一些可能的实现中，通信系统可以成为大规模机器学习和AI服务的平台。其中，终端设备既可以从网络享受AI推理服务或者AI模型训练服务，也可以参与网络模型训练所需要的数据收集，甚至参与分布式模型训练。
一种可能的实现方式中，联邦学习(federated learning，FL)，作为一种典型的AI数据处理模式被广泛应用。其中，FL这一概念的提出有效地解决了当前人工智能发展所面临的困境，其在充分保障用户数据隐私和安全的前提下，通过促使各个边缘设备和中心端 服务器协同合作来高效地完成模型的学习任务。如图3a所示，FL架构是当前FL领域最为广泛的训练架构，FedAvg算法是FL的基础算法，以中心节点为中心端，分布式节点为客户端为例，其算法流程大致如下：
(1)中心端初始化待训练模型
Figure PCTCN2022110616-appb-000005
并将其广播发送给所有客户端设备。
(2)在第t∈[1,T]轮中，客户端k∈[1,K]基于局部数据集
Figure PCTCN2022110616-appb-000006
对接收到的全局模型
Figure PCTCN2022110616-appb-000007
进行E个epoch的训练以得到本地训练结果
Figure PCTCN2022110616-appb-000008
将其上报给中心节点。
(3)中心节点汇总收集来自全部(或部分)客户端的本地训练结果，假设第t轮上传局部模型的客户端集合为
Figure PCTCN2022110616-appb-000009
中心端将以对应客户端的样本数为权重进行加权求均得到新的全局模型，具体更新法则为
Figure PCTCN2022110616-appb-000010
其后中心端再将最新版本的全局模型
Figure PCTCN2022110616-appb-000011
广播发送给所有客户端设备进行新一轮的训练。
(4)重复步骤(2)和(3)直至模型最终收敛或训练轮数达到上限。
可选地，收敛一般是指训练得到的模型的性能符合了预设要求，例如利用FL训练一个用于图像分类任务的模型，预设的分类准确度是95％，那么随着训练的进行，通过测试集对模型的准确度性能进行评估，发现准确度达到95％或以上时就可以确认模型已经收敛了，并停止训练。同时，由于模型结构设计、参数选择、训练方法等因素设计的不好，可能会导致永远无法预设性能要求的情况，这时就要考虑为训练轮次数设置一个上限，达到上限次数，即使模型没有达到预设性能，也要停止训练了。
除了上报本地AI模型
Figure PCTCN2022110616-appb-000012
还可以将训练的本地梯度
Figure PCTCN2022110616-appb-000013
进行上报，中心节点将本地梯度求平均，并根据这个平均梯度的方向更新全局模型。
可以看到，在FL框架中，数据集存在于分布式节点处，即分布式节点收集本地的数据集，并进行本地训练，将训练得到的本地结果(模型或梯度)上报给中心节点。中心节点本身没有数据集，只负责将分布式节点的训练结果进行融合处理，得到全局模型，并下发给分布式节点。
由上述描述可知，在联邦学习方法中，需要系统中存在一个中心节点进行模型的融合，因而是星型结构，其存在鲁棒性差的问题，一旦中心节点出现问题(如被攻击)，则会导致整个系统的瘫痪。同时，联邦学习系统中，多个本地AI模型能够用加权平均的方式进行融合的前提是各本地AI模型具有相同的结构和参数量。而实际网络中，各分布式节点的设备能力是不同的，要求他们使用相同的本地AI模型往往限制了其灵活性，也无法满足不同分布式节点的不同需求。
另一种可能的实现方式中，去中心式的AI数据处理模式作为对联邦学习的AI数据处理模式的一种改进，该模式无需设置中心节点，可以提升系统的鲁棒性。在去中心式系统中，各个节点利用本地数据和本地目标计算本地AI模型之后，与通信可达的邻居节点交互各自的AI模型，并基于交互的AI模型进一步处理(例如训练、融合等)本地AI模型。
示例性的，与联邦学习不同，去中心式的AI数据处理模式中的去中心式学习过程如图3b所示，即没有中心节点的完全分布式系统。该系统的设计目标f(x)一般是各节点目标f i(x) 的均值，即
Figure PCTCN2022110616-appb-000014
其中n是分布式节点数量，x是待优化参数，在机器学习中，x就是机器学习(如神经网络)模型的参数。各节点利用本地数据和本地目标f i(x)计算本地梯度
Figure PCTCN2022110616-appb-000015
然后将其发送给通信可达的邻居节点。任一节点收到其邻点发来的梯度信息后，可以按照下式更新本地AI模型的参数x：
Figure PCTCN2022110616-appb-000016
其中，N i是节点i的邻居节点集合，|N i|表示节点i的邻居节点集合中的元素数量，即节点i的邻居节点数量，上标k表示第k(k为正整数)轮训练，α k是第k轮训练使用的训练步长。通过节点间的信息交互，去中心式学习系统最终将学到一个统一的模型。
然而，在上述去中心式系统中，由于没有中心节点的统一调度，不同分布式节点在接收到其它节点的AI模型之后，如何实现AI模型处理，是一个亟待解决的技术问题。示例性的，不同分布式节点使用的AI模型在训练过程中会出现版本差距，例如，某些节点基于接收到的较新的(或迭代次数较多的)AI模型处理本地AI模型，而另一些节点基于接收到的较旧的(或迭代次数较少的)AI参数处理本地AI模型，将会导致分布式系统中不同节点得到的AI模型版本不一致，影响该系统得到的AI模型性能。
为了解决上述问题，本申请提供了一种AI模型处理方法及相关设备，下面将结合附图进一步说明。
请参阅图4，为本申请提供的AI模型处理方法的一个交互示意图，该方法包括如下步骤。
S401.第一节点确定第一AI模型。
S402.第一节点发送第一信息，该第一信息用于指示第一AI模型的模型信息和第一AI模型的辅助信息。
本实施例中，第一节点在步骤S402中发送第一信息。相应的，对于第二节点而言，该第二节点在步骤S402中接收来自N个第一节点发送的N个第一信息，N为正整数。其中，该第一信息用于指示第一AI模型的模型信息和第一AI模型的辅助信息。
应理解，第一AI模型的模型信息用于构建该第一AI模型。示例性的，该模型信息可以包括模型的参数信息，模型的结构信息等至少一项。
可选地，模型的结构信息可以包括模型层数、模型中每层神经元的数量、模型中层与层之间的连接关系等至少一项。
在一种可能的实现方式中，第一节点在步骤S401中确定的第一AI模型为该第一节点所在的系统中M个节点可理解的模型，M为大于或等于2的整数。相应的，第一节点在步骤S402中发送的第一信息所指示的第一AI模型为该第一节点所在的系统中M个节点可理解的模型，以便于在该M个节点中存在多种不同的节点类型的情况下，其它节点在接收该第一信息之后能够理解该第一AI模型并进一步执行模型处理。
可选地，第一节点在步骤S401中确定的第一AI模型也可以不是该第一节点所在的系统中M个节点中的部分或全部节点可理解的模型。相应的，该M个节点中的部分或全部节点在接收该第一AI模型的模型信息之后，可以将该第一AI模型的模型信息发送至其它节点(例 如服务器，控制器，网络设备，中心节点等)，使得其它节点基于该第一AI模型的模型信息进行模型处理，以得到并向该M个节点中的部分或全部节点发送可理解的模型的模型信息，以便于该M个节点中的部分或全部节点基于可理解的模型的模型信息进行后续的模型处理过程。
可选地，第一节点在步骤S401中确定的第一AI模型也可以不是该第一节点所在的系统中M个节点中的部分或全部节点可理解的模型。相应的，该M个节点中的部分或全部节点在接收该第一AI模型的模型信息之后，可以丢弃不可理解的模型。
应理解，第一AI模型为该第一节点所在的系统中M个节点可理解的模型，可以表述为第一AI模型为该第一节点所在的系统中M个节点通用的模型，第一AI模型的模型结构为该第一节点所在的系统中M个节点可理解的模型结构，第一AI模型的模型结构为该第一节点所在的系统中M个节点通用的模型结构，第一AI模型为该第一节点所在的系统的公共模型，或，第一AI模型的模型结构为该第一节点所在的系统的公共模型结构。
以第一节点所在的系统中M个节点可理解的模型为公共AI模型为例，与之相对的，第一节点所在的系统中M个节点中各个节点本地处理的AI模型可能是其它节点不可理解的模型，该模型可以称为本地AI模型。下面将对M个节点中的任一节点而言，公共AI模型和本地AI模型的获取过程进行示例性描述。
1.本地AI模型的获取过程。
在第一节点所在的系统中M个节点中，各个节点的本地AI模型用于本节点的学习任务，不同节点的本地AI模型结构、参数量可能不同。
一种可能的实现方式中，第一节点所在的系统中M个节点中各个节点可以根据能力信息(即设备自身的设备能力)和/或需求信息(即本地任务的需求)选择合适的本地AI模型。
具体地，能力信息(即设备自身的设备能力)主要包括计算能力和存储能力。其中，对于该计算能力而言，该计算能力一般由设备中的计算模块决定，该计算模块可以包括中央处理器(central processing units，CPU)、图形处理单元(graphics processing unit，GPU)、神经处理单元(neural processing unit，NPU)或者张量处理单元(tensor processing unit，TPU)等至少一项。例如，以每秒浮点运算次数(floating-point operations per second，flops)为单位，其描述了设备可以以怎样的速度进行模型训练和推理。对于该存储能力而言，存储能力一般由设备中的存储模块(如显存/缓存/内存等)决定，以字节(Byte)为单位(或基于字节的KB/MB/GB等单位为单位)，其描述了设备可以保存多大的模型。
可选地，需求信息(即本地任务的需求)一般指节点使用某个模型完成特定本地任务时的性能需求，如定位任务中的定位精度、分类任务中的分类准确度、信息重建任务中的均方误差、压缩任务中的压缩率等。
由上述内容可知，第一节点所在的系统中M个节点中各个节点在配置本地AI模型时，可以选择满足本地任务需求，且计算能力、存储能力要求最小的模型作为本地AI模型，例如当多个模型都可以满足本地任务需求时，选择其中模型参数量最小、模型训练/推理运算量最小的模型；也可以在计算能力、存储能力要求小于等于节点计算能力、存储能力的模型中选择性能最好的模型作为本地AI模型。
可选地，第一节点所在的系统中M个节点中各个节点也可以通过预配置的方式配置本地AI模型，例如，系统或标准可以针对给定任务提供一组可选的模型库，用于该各个节点从中选择本地AI模型。
此外，本地AI模型的配置可以由第一节点所在的系统中的管理节点进行，然后下发或安装到各个节点中，也可以由各个节点自行进行。如果是前者，则还可以是各个节点上报自身的计算能力、存储能力、任务需求至管理节点。
2.公共AI模型的获取过程。
在第一节点所在的系统中M个节点中，公共AI模型用于各个节点间的知识传递，其中，该系统的不同节点使用结构、参数量相同的公共AI模型。这种同构的公共AI模型设计有利于简化节点间模型传输的机制设计(模型传输的实现可参考前述图3b及相关描述)。
类似地，公共AI模型可以为其它节点配置或预配置的方式实现，例如，由管理节点或标准预先给定，不同的学习任务(如信道状态信息(channel state information，CSI)压缩、波束管理、定位等)可以使用不同的公共AI模型。
此外，公共AI模型的配置需要考虑第一节点所在的系统中各个节点的计算能力和/或存储能力。例如，由管理节点确定公共AI模型时，可以由管理节点收集各个节点的计算能力和存储能力信息，基于能力最弱的节点的计算能力和存储能力确定公共AI模型。再如，由预配置的方式预先给定公共AI模型时，该标准应同时给定该公共AI模型对应的计算能力要求和存储能力要求，各个节点加入系统之前，就应确保其自身相应能力不小于标准给定的能力要求。
可选地，在上述实现过程中，本地AI模型配置和公共AI模型配置一般在系统构建之初执行，当系统状态(如系统所在环境、各个节点的能力和需求)发生变化时，可以重新对本地AI模型和公共AI模型进行配置或重置。
在一种可能的实现方式中，上述提及的本地AI模型和公共AI模型可以面向相同的功能。例如，本地AI模型和公共AI模型的功能可以是相同的，例如，都用于定位、分类、重建、压缩等任务。此时两者仅存在模型结构、参数量、性能上的差别。
在另一种可能的实现方式中，上述提及的本地AI模型和公共AI模型可以面向不同的功能：即本地AI模型和公共AI模型的功能可以是不同的。例如，各个节点基于GAN结构进行数据生成，此时本地AI模型可以是GAN结构中的生成器模型，而公共AI模型可以是GAN结构中的判别器模型，即各个节点根据本地数据和邻点发来的判别器模型，训练更新本地的生成器模型和判别器模型，并把更新后的判别器模型发送给邻点。判别器模型中包含了各个节点的本地数据相关信息，交互判别器模型实现了节点间知识的交互。再如，本地AI模型和公共AI模型分别是一个完成模型的两部分，如本地AI模型是自编码器模型的编码器(encoder)部分，公共AI模型是自编码器模型的译码器(decoder)部分。各个节点可以用本地数据训练各不相同的编码器(本地AI模型)加相同的译码器(公共AI模型)，然后将译码器(公共AI模型)发送给邻点。译码器部分同样携带了各节点本地的数据信息，因此实现了节点间知识的交互。
示例性的，如图5a所示，上述提及的本地AI模型和公共AI模型之间存在“知识分享”和/或“知识吸收”的实现过程。
例如，“知识分享”是指基于本地AI模型获取(或更新)公共AI模型。当本地AI模型和公共AI模型面向相同的功能时，知识分享可以通过知识蒸馏、剪枝、扩张的现有技术实现；当本地AI模型和公共AI模型面向不同的功能时，则通过本地AI模型和公共AI模型的联合训练实现知识的分享。
又如，“知识吸收”是指基于公共AI模型更新本地AI模型。同样，当本地AI模型和公共AI模型面向相同的功能时，知识吸收可以通过知识蒸馏、剪枝、扩张的现有技术实现；当本地AI模型和公共AI模型面向不同的功能时，则通过本地AI模型和公共AI模型的联合训练实现知识的吸收。
在一种可能的实现方式中，第一节点在步骤S401中确定的该第一AI模型为基于该第一节点的节点类型得到。具体地，在前述图3b所示去中心式系统中，各节点使用相同的学习模型(参数都是x)，没有考虑各节点可能不同的能力和需求，并且，模型融合采用上述所示的求平均方式，且各节点功能相同，上述系统中未考虑节点的功能区分。而在本实施例中，第一节点发送的第一信息所指示的第一AI模型可以为该第一节点基于该第一节点的节点类型得到的模型。其中，在该第一AI模型至少基于第一节点的节点类型得到的情况下，使得第一节点可以基于不同的节点类型执行不同的AI模型处理过程，解决节点功能单一的问题，以提升灵活性。
可选地，第一节点在步骤S401中确定的该第一AI模型为基于该第一节点的节点类型和第二AI模型得到。其中，在该第一AI模型至少基于第二AI模型得到的情况下，该第一节点所发送的第一信息所指示的第一AI模型可以为其它节点可理解的模型，以便于其它节点在接收该第一AI模型之后进行进一步的模型处理。
可选地，该节点类型包括以下任一项：基于本地数据进行本地训练的节点类型(为便于后文描述，记为L节点)，基于其它节点的AI模型进行融合处理的节点类型(为便于后文描述，记为A节点)，基于该本地数据进行本地训练并基于其它节点的AI模型进行融合处理的节点类型(为便于后文描述，记为H节点)。具体地，用于得到该第一AI模型的节点类型可以为上述任一项实现，以提升方案实现的灵活性。
可选地，在图4所示方法中，该方法还包括：该第一节点发送指示该第一节点的节点类型的指示信息。具体地，该第一节点还可以发送指示该第一节点的节点类型的指示信息，以便于其它节点基于该指示信息明确该第一节点的节点类型，后续可以基于该第一节点的节点类型与该第一节点进行交互。
可选地，在图4所示方法中，在步骤S401之前，该第一节点基于能力信息和/或需求信息确定该第一节点的节点类型；或，该第一节点接收指示该第一节点的节点类型的指示信息。具体地，该第一节点可以基于自身的能力信息和/或需求信息以明确该第一节点的节点类型，也可以基于其它节点的指示以明确该第一节点的节点类型，以提升方案实现的灵活性。
示例性的，下面将结合一些实现示例对上述三种节点类型及其实现过程进一步描述。
对于L节点而言，可以称为本地训练节点，这类节点具有知识分享的能力和/或需求，无知识吸收的能力和/或需求，同时节点上有本地数据，可以用于本地训练。
对于A节点而言，可以称为模型融合节点，这类节点无知识分享的能力/需求，有知识吸收的能力和/或需求，同时节点上可能没有用于本地训练的本地数据。
对于H节点而言，可以称为混合功能节点，这类节点既有知识分享的能力/需求，又有知识吸收的能力和/或需求，同时节点上有本地数据，可以用于本地训练。
可选地，类似于前述本地AI模型和公共AI模型的配置方式，节点的类型可以由节点根据自身的能力、需求以及本地数据情况确定，也可以由管理节点指定，也可以由预配置的方式确定。
此外，节点类型确定后，需要将自身的节点类型(即角色信息)告知邻点，即角色交互。同时，节点还可以将本节点参与的学习任务类型信息和节点类型共同告知邻点。例如，各个节点接收到邻点发来的节点类型后，构建邻点信息表的实现可以如下述表1所示。
表1
Figure PCTCN2022110616-appb-000017
在上述表1所示的邻点信息表中，节点ID是该邻居节点在去中心式学习系统中的唯一标识，可以在该邻居节点加入该学习系统时统一分配；邻点角色由邻点发来的节点类型确定；任务类型表示该邻点参与的学习任务的类型，例如CSI压缩、波束管理、定位等，该任务类型也可以由邻点发来的节点类型确定；链路质量指标描述了给定邻点和本节点之间通信链路的质量，可以包括通信链路的信噪比(signal noise ratio，SNR)、吞吐、时延、丢包率等指标。
在一种可能的实现方式中，各个节点的角色并不是一成不变的，在学习过程(学习系统运行过程)中，通信网络拓扑、节点需求、环境状态等发生变化时，节点角色可能发生改变。如图5b所示，给出一些节点角色发生转变的例子。
例如，H节点无法获取新的本地数据，或基于隐私或其它考虑，关闭知识分享功能，转变为A节点。
又如，H节点判断基于本地数据足以训练获取性能较好的本地AI模型，关闭知识吸收功能，转变为L节点。
又如L节点无法获取新的本地数据，关闭本地训练功能，开启知识吸收功能，转变为A节点。
又如L节点判断基于本地数据无法训练获得性能满意的本地AI模型，开启知识吸收功能，转变为H节点。
又如A节点可以获取本地数据，开启本地训练功能，转变为H节点。
又如A节点可以获取本地数据，且判断基于本地数据足以训练获取性能较好的本地AI模型，关闭知识吸收功能，转变为L节点。
类似于表1的实现过程，在各个节点的角色发生变化后，可以通知邻点该变化，以便邻点更新各自维护的邻点信息表。
在一种可能的实现方式中，对于任一节点而言，在完成邻点信息表构建后，各个节点开始按照自身的角色，运行相关操作，进行分布式学习。学习的过程周期性的进行，每个周期分为两个阶段：模型更新和模型交互。下表2给出了各阶段不同节点的具体操作。
表2
Figure PCTCN2022110616-appb-000018
在上述操作中，模型的本地训练更新可以采用类似前述图2a描述的全连接神经网络中使用的梯度反向传递训练实现，也可以通过其它AI训练过程实现，此处不做限定。此外，基于本地AI模型获取公共AI模型和基于公共AI模型更新本地AI模型的方法可以参考前述图5a及相关描述中的介绍，在此不再赘述。
由上述实现过程可知，对于第一节点所发送的第一信息所指示的第一AI模型为基于第二AI模型得到的情况下，该第二AI模型可以是第一节点基于本地数据得到；或，该第二AI模型可以是第一节点基于K个信息得到，该K个信息中的每个信息指示其它节点的AI模型的模型信息和该其它节点的AI模型的辅助信息，K为正整数；或，该第二AI模型可以是第一节点基于该本地数据以及该K个信息得到。具体地，用于得到该第一AI模型的第二AI模型可以为上述任一项实现，以提升方案实现的灵活性。
可选地，K的取值比M的取值小1，或，K的取值小于M-1。
在一种可能的实现方式中，该第一AI模型的辅助信息包括以下至少一项：该第一AI模型的类型信息，该第一节点的标识信息，该第一AI模型的接收节点的标识信息，该第一AI模型的版本信息，生成该第一AI模型的时间信息，生成该第一AI模型的地理位置信息，该第一节点的本地数据的分布信息。具体地，第一信息所指示的第一AI模型的辅助信息包括 上述至少一项，以便于第一信息的接收方能够基于上述至少一项信息对该第一AI模型的模型信息进行AI模型处理，提升该第一信息的接收方基于该第一AI模型进行处理得到的AI模型的性能。
可选地，该第一AI模型的模型信息和该第一AI模型的辅助信息分别承载于不同的传输资源，该传输资源包括时域资源和/或频域资源。具体地，由于该第一AI模型的模型信息对应的数据量和该第一AI模型的辅助信息对应的数据量一般是不同的(例如前者一般多于后者)，使得两者可以分别承载于不同的传输资源上。
可选地，该传输资源为预配置的资源。
示例性的，如前述描述可知，第一节点所在系统中的M个节点将本节点更新后的公共AI模型发送给邻居节点(例如第一节点在步骤S402向第二节点发送的第一信息携带有第一AI模型的模型信息)。由于公共AI模型的结构、参数量是预设且统一的，因此可以设计复杂度较低的通信机制。下面将结合更多的实现示例，对第一节点在步骤S402中发送的第一信息的实现过程进行描述。
对于第一节点在步骤S402中发送的第一信息而言，该第一信息所指示的第一AI模型可以为公共AI模型。其中，第一节点可以将第一信息按照图5c所示的方式打包成数据包，其中，公共AI模型的模型信息承载于有效载荷(payload)，公共AI模型的辅助信息承载于包头部分。如图5c所示，包头部分包括payload类型、节点ID、版本号、时间戳、地理位置、数据分布等至少一项信息。
可选地，payload类型用于指示payload部分携带的公共AI模型信息的类型，该类型可能是模型参数、模型梯度、(中间)推理结果等，可以设计比特映射表隐式的指示这些信息，例如用00表示模型参数、11表示模型梯度、01或10表示(中间)推理结果。
可选地，节点ID可以包括源节点ID和目的节点ID，该节点ID和邻点信息表中使用的节点ID相同，源节点ID用于标识生成payload包含的公共AI模型信息的节点，目的节点ID用于标识payload包含的公共AI模型信息的目的地节点。
可选地，版本号用于指示payload包含的公共AI模型的版本，版本号的设计将在实施例二中具体讨论。
可选地，时间戳是payload包含的公共AI模型更新完成的时间点。
可选地，地理位置标识了payload包含的公共AI模型被更新时节点所处的地理位置信息。
可选地，数据分布为源节点本地数据的分布信息，没有本地数据的节点生成的数据包，此字段可以为空(或填0)。版本号、时间戳、地理位置、数据分布等信息将被用于模型的融合，因此统称为模型融合辅助信息。
对于第一信息的发送过程而言，第一节点可以按照上述方式完成数据包打包后，在步骤S402中将该数据包发送给相邻节点。示例性的，以第一节点和第二节点之间的通信链路为侧行链路(sidelink)为例，即侧行链路(sidelink)进行数据包发送。
可选地，数据包的包头和payload的部分可以分开发送，例如，将包头在控制信道上发送，payload在数据信道上发送，此时包头所在的控制信道还需要发送指示发送该包头对 应的payload所使用的数据信道的传输资源(时频空资源)位置。需要注意的是，由于公共AI模型的结构和参数量是预设且统一的，因此其使用的传输资源量也是可以预知且固定的，因此可以预先分配专门用于公共AI模型传输的资源，从而简化了通信机制的设计(即无需根据不同的数据包长度分确定传输资源分配和调度策略)。
S403.第二节点基于N个第一信息进行模型处理，得到目标AI模型。
本实施例中，第二节点在步骤S402中接收N个第一信息之后，该第二节点在步骤S403中基于N个第一信息进行模型处理，得到目标AI模型。
可选地，该目标AI模型用于完成该第二节点的AI任务，或，该目标AI模型为该第二节点的本地模型。
在一种可能的实现方式中，该第二节点基于该N个第一信息进行模型处理，得到目标AI模型包括：该第二节点基于该N个第一信息以及该第二节点的节点类型进行模型处理，得到该目标AI模型。具体地，在该第二AI模型至少基于第二节点的节点类型得到的情况下，使得第二节点可以基于不同的节点类型执行不同的AI模型处理过程，解决节点功能单一的问题，以提升灵活性。
可选地，K的取值比M的取值小1，或，K的取值小于M-1。
示例性的，在步骤S403中，第二节点收到N个邻点发送的包含公共AI模型信息的数据包后，可以进行解包，并获得包头和payload部分。节点根据模型融合辅助信息(版本号、时间戳、地理位置、数据分布等)进行公共AI模型的融合。可以将N个邻点发来的公共AI模型信息进行分组，采用多级融合的方式进行融合。例如，将时间戳相近的数据包中的公共AI模型信息分为一组，或将地理位置信息相近的数据包中的公共AI模型信息分为一组，或将数据分布不同的数据包中的公共AI模型信息分为一组。模型融合可以采用加权平均、蒸馏等方法实现，在此不再赘述。
下面将以第二节点接收的N个第一信息中包含了AI模型的版本号信息的过程进行示例性描述。其中，版本号可以设计为AxLyHz，A，L，H为固定不变字段，x，y，z为整数。并且，版本号更新规则如下：L节点基于本地AI模型对公共AI模型进行一次训练更新，模型版本号中的y字段加1；A节点基于收到的邻点发来的公共AI模型，每次融合后得到更新的公共AI模型，模型版本号中的x字段加1，x字段每次加1，y，z字段置0；H节点基于本地AI模型和邻点发来的公共AI模型对公共AI模型进行一次更新，模型版本号中的x，y，z字段分别加1，x字段每次加1，y，z字段置0。
如图5d所示，一个圆圈代表一个AI模型，标识某一个模型的版本号为AxLyHz。在图5d所示示例中，x的取值可以为1，2，3或4等整数；y的取值可以为0，1，y，u等整数；z的取值可以为0，x，v，w或z等整数。此外，在图5d所示基于版本号的模型融合过程中，可以将版本号相近的公共AI模型分为一组，先进行组内融合，完成后再进行分组和组内融合，直至融合过程完成。例如，如图5d所示，将x字段相同的公共AI模型分为一组，每完成一次组内融合，x字段加1，y，z字段重置为0，直至获得融合后的一个公共AI模型。
需要说明的是，第二节点所在系统中的不同的节点最后融合得到的公共模型的版本号类似，甚至有可能是相同的。整个系统的目的是为了各个节点都能获得一个模型，用于节 点本地的AI任务的完成，所以版本号的存在只是为了让更相似的模型先融合，并不一定版本越高的模型性能就会越好。
类似地，第二节点还可以基于其它AI模型的辅助信息(例如时间戳、地理位置等)进行融合(蒸馏、剪枝、扩张)，具体实现过程可以参考上述示例，此处不做赘述。
在一种可能的实现方式中，该第二节点在步骤S403中基于该N个第一信息和该第二节点的节点类型进行模型处理，得到该目标AI模型的过程包括：在该第二节点的节点类型为基于其它节点的AI模型进行融合处理的节点类型时，该第二节点基于该N个第一信息对该N个第一AI模型进行模型融合，得到该目标AI模型。具体地，在该第二节点的节点类型为基于其它节点的AI模型进行融合处理的节点类型时，该第二节点在接收N个第一信息确定N个第一AI模型之后，该第二节点对该N个第一AI模型进行模型融合，以得到该目标AI模型。
在一种可能的实现方式中，该第二节点在步骤S403中基于该N个第一信息和该第二节点的节点类型进行模型处理，得到该目标AI模型的过程包括：在该第二节点的节点类型为基于该本地数据进行本地训练并基于其它节点的AI模型进行融合处理的节点类型时，该第二节点基于该N个第一信息对该N个第一AI模型以及第二AI模型进行模型融合，得到该目标AI模型，其中，该第二AI模型为基于本地数据训练得到。具体地，在该第二节点的节点类型为基于该本地数据进行本地训练并基于其它节点的AI模型进行融合处理的节点类型时，该第二节点在接收N个第一信息确定N个第一AI模型之后，该第二节点需要基于本地数据训练得到第二AI模型，并对该N个第一AI模型以及第二AI模型进行模型融合，以得到该目标AI模型。
应理解，第二节点所使用的第二AI模型可以参考前述第一节点所使用的第二AI模型的相关描述，并实现相应的技术效果，此处不做赘述。
在一种可能的实现方式中，该方法还包括：该第二节点接收用于指示该第一节点的节点类型的指示信息。具体地，该第二节点还可以接收指示该第一节点的节点类型的指示信息，以便于该第二节点基于该指示信息明确该第一节点的节点类型，后续该第二节点可以基于该第一节点的节点类型与该第一节点进行交互。
在一种可能的实现方式中，该方法还包括：该第二节点发送用于指示该第二节点的节点类型的指示信息。具体地，该第二节点还可以发送指示该第二节点的节点类型的指示信息，以便于其它节点基于该指示信息明确该第二节点的节点类型，后续可以基于该第二节点的节点类型与该第二节点进行交互。
在一种可能的实现方式中，该方法还包括：该第二节点基于能力信息和/或需求信息确定该第二节点的节点类型；或，该第二节点接收指示该第二节点的节点类型的指示信息。具体地，该第二节点可以基于自身的能力信息和/或需求信息以明确该第二节点的节点类型，也可以基于其它节点的指示以明确该第二节点的节点类型，以提升方案实现的灵活性。
应理解，第二节点的节点类型的相关实现过程(例如节点类型的确定、节点类型的指示等)可以参考前述第一节点的节点类型的相关描述，并实现相应的技术效果，此处不做赘述。
基于图4所示技术方案，第一节点在确定第一AI模型之后，该第一节点发送指示该第一AI模型的模型信息和该第一AI模型的辅助信息的第一信息。相比于不同节点之间仅交互各自的AI模型的方式，由于第一信息指示第一AI模型的模型信息之外，该第一信息还可以指示第一AI模型的辅助信息，使得该第一信息的接收方能够基于该第一AI模型的辅助信息对该第一AI模型的模型信息进行AI模型处理(例如训练、融合等)，提升该第一信息的接收方基于该第一AI模型进行处理得到的AI模型的性能。
请参阅图6，为本申请提供的AI模型处理方法的另一个交互示意图，该方法包括如下步骤。
S601.第一节点获取本地AI模型。
本实施例中，第一节点在步骤S601中获取本地AI模型，该本地AI模型用于完成该第一节点的AI任务。
在一种可能的实现方式中，第一节点在步骤S601获取的本地AI模型基于本地数据得到；或，第一节点在步骤S601获取的本地AI模型基于K个信息得到，该K个信息中的每个信息指示其它节点的AI模型的模型信息和该其它节点的AI模型的辅助信息，K为正整数；或，第一节点在步骤S601获取的本地AI模型基于该本地数据以及该K个信息得到。具体地，用于得到该公共AI模型的本地AI模型可以为上述任一项实现，以提升方案实现的灵活性。
应理解，第一节点在步骤S601获取的本地AI模型的相关实现过程(例如本地AI模型的配置、本地AI模型的模型处理过程等)可以参考前述图4及相关实施例的描述，并实现相应的技术效果，此处不做赘述。
S602.第一节点基于本地AI模型确定公共AI模型。
本实施例中，第一节点在步骤S601中获取本地AI模型之后，该第一节点在步骤S402中基于本地AI模型确定公共AI模型，其中，该公共AI模型为该第一节点所在的系统中M个节点可理解的模型，M为大于或等于2的整数。
应理解，第一节点在步骤S602确定的公共AI模型为该第一节点所在的系统中M个节点可理解的模型，可以表述为公共AI模型为该第一节点所在的系统中M个节点通用的模型，公共AI模型的模型结构为该第一节点所在的系统中M个节点可理解的模型结构，公共AI模型的模型结构为该第一节点所在的系统中M个节点通用的模型结构。
应理解，公共AI模型的相关实现过程(例如公共AI模型的配置、公共AI模型的模型处理过程等)可以参考前述图4及相关实施例的描述，并实现相应的技术效果，此处不做赘述。
在一种可能的实现方式中，在步骤S602中该第一节点基于该本地AI模型确定公共AI模型的过程包括：该第一节点基于该本地AI模型和该第一节点的节点类型确定公共AI模型。具体地，该第一节点可以基于该本地AI模型和该第一节点的节点类型确定该公共AI模型，使得该第一节点可以基于不同的节点类型可以执行不同的AI模型处理过程，解决节点功能单一的问题，以提升灵活性。
应理解，第一节点的节点类型的相关实现过程(例如节点类型的确定、节点类型的指示等)可以参考前述图4及相关实施例的描述，并实现相应的技术效果，此处不做赘述。
可选地，该节点类型包括以下任一项：基于本地数据进行本地训练的节点类型，基于其它节点的AI模型进行融合处理的节点类型，基于该本地数据进行本地训练并基于其它节点的AI模型进行融合处理的节点类型。具体地，用于得到该公共AI模型的节点类型可以为上述任一项实现，以提升方案实现的灵活性。
可选地，在图6所示方法中，该方法还包括：该第一节点发送指示该第一节点的节点类型的指示信息。具体地，该第一节点还可以发送指示该第一节点的节点类型的指示信息，以便于其它节点基于该指示信息明确该第一节点的节点类型，后续可以基于该第一节点的节点类型与该第一节点进行交互。
可选地，在图6所示方法中，该方法还包括：该第一节点基于能力信息和/或需求信息确定该第一节点的节点类型；或，该第一节点接收指示该第一节点的节点类型的指示信息。具体地，该第一节点可以基于自身的能力信息和/或需求信息以明确该第一节点的节点类型，也可以基于其它节点的指示以明确该第一节点的节点类型，以提升方案实现的灵活性。
S603.第一节点发送第一信息，该第一信息用于指示公共AI模型的模型信息。
本实施例中，第一节点在步骤S603中发送用于指示公共AI模型的模型信息的第一信息，相应的，对于第二节点而言，该第二节点在步骤S603中接收来自N个第一节点发送的N个第一信息，N为正整数。
应理解，第一信息的相关实现过程(例如第一信息的打包过程、第一信息的发送过程等)可以参考前述图4及相关实施例的描述，并实现相应的技术效果，此处不做赘述。
S604.第二节点基于N个第一信息更新本地AI模型，得到更新后的本地AI模型信息。
本实施例中，第二节点在步骤S603中接收N个第一信息之后，该第二节点在步骤S604中基于N个第一信息更新本地AI模型，得到更新后的本地AI模型信息。
在一种可能的实现方式中，第一节点在步骤S603中发送的第一信息还指示该公共AI模型的辅助信息。具体地，该第一节点发送的第一信息还指示该公共AI模型的辅助信息。相比于不同节点之间仅交互各自的AI模型的方式，由于第一信息指示公共AI模型的模型信息之外，该第一信息还可以指示公共AI模型的辅助信息，使得该第一信息的接收方能够基于该第一AI模型的辅助信息对该公共AI模型的模型信息进行AI模型处理(例如训练、融合等)，提升该第一信息的接收方基于该公共AI模型进行处理得到的AI模型的性能。
可选地，该公共AI模型的辅助信息包括以下至少一项：该公共AI模型的类型信息，该第一节点的标识信息，该公共AI模型的接收节点的标识信息，该公共AI模型的版本信息，生成该公共AI模型的时间信息，生成该公共AI模型的地理位置信息，该公共节点的本地数据的分布信息。具体地，第一信息所指示的公共AI模型的辅助信息包括上述至少一项，以便于第一信息的接收方能够基于上述至少一项信息对该公共AI模型的模型信息进行AI模型处理，提升该第一信息的接收方基于该公共AI模型进行处理得到的AI模型的性能。
可选地，该公共AI模型的模型信息和该公共AI模型的辅助信息分别承载于不同的传输资源，该传输资源包括时域资源和/或频域资源。具体地，由于该公共AI模型的模型信息对应的数据量和该公共AI模型的辅助信息对应的数据量一般是不同的(例如前者一般多于后者)，使得两者可以分别承载于不同的传输资源上。
应理解，第一信息的相关实现过程(例如第一信息的接收过程，第二节点基于第一信息进行AI模型处理过程等)可以参考前述图4及相关实施例的描述，并实现相应的技术效果，此处不做赘述。
在一种可能的实现方式中，在步骤S604中，该第二节点基于该N个第一信息更新本地AI模型的过程，得到更新后的本地AI模型的过程包括：该第二节点基于该N个第一信息和该第二节点的节点类型更新本地AI模型，得到更新后的本地AI模型。具体地，该第二节点可以基于该N个第一信息和该第二节点的节点类型确定该公共AI模型，使得该第二节点可以基于不同的节点类型可以执行不同的AI模型处理过程，解决节点功能单一的问题，以提升灵活性。
可选地该节点类型包括以下任一项：基于其它节点的AI模型进行融合处理的节点类型，基于该本地数据进行本地训练并基于其它节点的AI模型进行融合处理的节点类型。具体地，用于得到该更新后的本地AI模型的节点类型可以为上述任一项实现，以提升方案实现的灵活性。
可选地，该节点类型为基于其它节点的AI模型进行融合处理的节点类型时，该本地AI模型基于P个信息得到，该P个信息中的每个信息指示其它节点的AI模型的模型信息和该其它节点的AI模型的辅助信息，该P为正整数。
可选地，该节点类型为基于该本地数据进行本地训练并基于其它节点的AI模型进行融合处理的节点类型时，该本地AI模型基于该本地数据得到(例如在首次迭代处理过程中)，或，该本地AI模型基于该本地数据以及该P个信息得到(例如在首次迭代处理过程之外的其他迭代处理过程中)，该P个信息中的每个信息指示其它节点的AI模型的模型信息和该其它节点的AI模型的辅助信息，该P为正整数。
具体地，在节点类型不同的情况下，该本地AI模型可以为上述不同的实现，以提升方案实现的灵活性。
可选地，该方法还包括：该第二节点发送指示该第二节点的节点类型的指示信息。具体地，该第二节点还可以发送指示该第二节点的节点类型的指示信息，以便于其它节点基于该指示信息明确该第二节点的节点类型，后续可以基于该第二节点的节点类型与该第二节点进行交互。
可选地，该方法还包括：该第二节点基于能力信息和/或需求信息确定该第二节点的节点类型；或，该第二节点接收指示该第二节点的节点类型的指示信息。具体地，该第二节点可以基于自身的能力信息和/或需求信息以明确该第二节点的节点类型，也可以基于其它节点的指示以明确该第二节点的节点类型，以提升方案实现的灵活性。
应理解，第二节点的节点类型的相关实现过程(例如节点类型的确定、节点类型的指示等)可以参考前述图4及相关实施例的描述，并实现相应的技术效果，此处不做赘述。
基于图6所示技术方案，第一节点获取用于完成该第一节点的AI任务的本地AI模型，该第一节点基于该本地AI模型确定公共AI模型，该第一节点发送用于指示该公共AI模型的模型信息的第一信息。其中，该公共AI模型为该第一节点所在的系统中M个节点可理解的模型。换言之，第一节点发送的第一信息所指示的公共AI模型为该第一节点所在的系统中M 个节点可理解的模型，以便于在该M个节点中存在多种不同的节点类型的情况下，其它节点在接收该第一信息之后能够理解该第一AI模型并进一步执行模型处理。
请参阅图7，为本申请提供的AI模型处理方法的另一个交互示意图，该方法包括如下步骤。
S701.第一节点确定第一AI节点的节点类型。
本实施例中，第一节点在步骤S701中确定该第一节点的节点类型，并且，该第一节点在步骤S701之后执行步骤S702和步骤S703中的至少一个步骤。
应理解，若该第一节点在步骤S701之后执行步骤S702和步骤S703的情况下，本申请对步骤S702和步骤S703的执行顺序不做限定。例如，步骤S702先执行且步骤S703后执行，或，步骤S703先执行且步骤S702后执行。
在一种可能的实现方式中，第一节点在步骤S701中确定的节点类型包括以下任一项：基于本地数据进行本地训练的节点类型，基于其它节点的AI模型进行融合处理的节点类型，基于该本地数据进行本地训练并基于其它节点的AI模型进行融合处理的节点类型。具体地，用于得到该第一AI模型的节点类型可以为上述任一项实现，以提升方案实现的灵活性。
可选地，该方法还包括：该第一节点基于能力信息和/或需求信息确定该第一节点的节点类型；或，该第一节点接收指示该第一节点的节点类型的指示信息。具体地，该第一节点可以基于自身的能力信息和/或需求信息以明确该第一节点的节点类型，也可以基于其它节点的指示以明确该第一节点的节点类型，以提升方案实现的灵活性。
S702.第一节点发送指示第一节点的节点类型的指示信息。
本实施例中，第一节点在步骤S701中确定第一节点的节点类型之后，该第一节点在步骤S702中发送指示第一节点的节点类型的指示信息。相应的，第二节点在步骤S702中接收指示第一节点的节点类型的指示信息。
具体地，该第一节点还可以发送指示该第一节点的节点类型的指示信息，以便于其它节点基于该指示信息明确该第一节点的节点类型，后续可以基于该第一节点的节点类型与该第一节点进行交互。
应理解，第一节点的节点类型的相关实现过程(例如节点类型的确定、节点类型的指示等)可以参考前述图4及相关实施例的描述，并实现相应的技术效果，此处不做赘述。
S703.第一节点发送第一信息，第一信息用于指示第一AI模型的模型信息。
本实施例中，第一节点在步骤S701中确定第一节点的节点类型之后，该第一节点基于该第一节点的节点类型确定第一AI模型，并且，该第一节点在步骤S703中发送用于指示第一AI模型的模型信息的第一信息。相应的，第二节点在步骤S703中接收N个第一节点发送的N个第一信息。
在一种可能的实现方式中，第一节点发送的第一信息所指示的第一AI模型可以为该第一节点基于该第一节点的节点类型得到的模型。在该第一AI模型至少基于第一节点的节点类型得到的情况下，使得第一节点可以基于不同的节点类型可以执行不同的AI模型处理过程，解决节点功能单一的问题，以提升灵活性。
可选地，第一节点发送的第一信息所指示的第一AI模型可以为该第一节点基于该第一节点的节点类型和第二AI模型得到的模型。其中，在该第一AI模型至少基于第二AI模型得到的情况下，该第一节点所发送的第一信息所指示的第一AI模型可以为其它节点可理解的模型，以便于其它节点在接收该第一AI模型之后进行进一步的模型处理。
可选地，该第二AI模型基于本地数据得到；或，该第二AI模型基于K个信息得到，该K个信息中的每个信息指示其它节点的AI模型的模型信息和该其它节点的AI模型的辅助信息，K为正整数；或，该第二AI模型基于该本地数据以及该K个信息得到。具体地，用于得到该第一AI模型的第二AI模型可以为上述任一项实现，以提升方案实现的灵活性。
可选地，该第一信息还用于指示该第一AI模型的辅助信息。具体地，该第一节点发送的第一信息还指示该第一AI模型的辅助信息。相比于不同节点之间仅交互各自的AI模型的方式，由于第一信息指示第一AI模型的模型信息之外，该第一信息还可以指示第一AI模型的辅助信息，使得该第一信息的接收方能够基于该第一AI模型的辅助信息对该第一AI模型的模型信息进行AI模型处理(例如训练、融合等)，提升该第一信息的接收方基于该第一AI模型进行处理得到的AI模型的性能。
可选地，该第一AI模型的辅助信息包括以下至少一项：该第一AI模型的类型信息，该第一节点的标识信息，该第一AI模型的接收节点的标识信息，该第一AI模型的版本信息，生成该第一AI模型的时间信息，生成该第一AI模型的地理位置信息，该第一节点的本地数据的分布信息。具体地，第一信息所指示的第一AI模型的辅助信息包括上述至少一项，以便于第一信息的接收方能够基于上述至少一项信息对该第一AI模型的模型信息进行AI模型处理，提升该第一信息的接收方基于该第一AI模型进行处理得到的AI模型的性能。
可选地，该第一AI模型的模型信息和该第一AI模型的辅助信息分别承载于不同的传输资源，该传输资源包括时域资源和/或频域资源。具体地，由于该公共AI模型的模型信息对应的数据量和该公共AI模型的辅助信息对应的数据量一般是不同的(例如前者一般多于后者)，使得两者可以分别承载于不同的传输资源上。
可选地，第一AI模型为该第一节点所在的系统中M个节点可理解的模型，M为大于或等于2的整数。具体地，第一节点发送的第一信息所指示的第一AI模型为该第一节点所在的系统中M个节点可理解的模型，以便于在该M个节点中存在多种不同的节点类型的情况下，其它节点在接收该第一信息之后能够理解该第一AI模型并进一步执行模型处理。
应理解，第一信息的相关实现过程(例如第一信息的打包过程、第一信息的发送过程等、第一信息的接收过程，第二节点基于第一信息进行AI模型处理过程等)可以参考前述图4及相关实施例的描述，并实现相应的技术效果，此处不做赘述。
基于图7所示技术方案，第一节点可以确定该第一节点的节点类型，后续该第一节点可以基于节点类型执行AI模型处理。使得该第一节点可以基于不同的节点类型可以执行不同的AI模型处理过程，解决节点功能单一的问题，以提升灵活性。
请参阅图8，本申请实施例提供了一种通信装置800，该通信装置800可以实现上述方法实施例中第一节点(该第一节点为终端设备或网络设备)的功能，因此也能实现上述方 法实施例所具备的有益效果。在本申请实施例中，该通信装置800可以是第一节点，也可以是第一节点内部的集成电路或者元件等，例如芯片。下文实施例以该通信装置800为第一节点为例进行说明。
一种可能的实现方式中，当该装置800为用于执行前述任一实施例中第一节点所执行的方法时，该装置800包括处理单元801和收发单元802；该处理单元801用于确定第一AI模型；该收发单元802用于发送第一信息，该第一信息指示该第一AI模型的模型信息和该第一AI模型的辅助信息。
在一种可能的实现方式中，该第一AI模型的辅助信息包括以下至少一项：
该第一AI模型的类型信息，该第一节点的标识信息，该第一AI模型的接收节点的标识信息，该第一AI模型的版本信息，生成该第一AI模型的时间信息，生成该第一AI模型的地理位置信息，该第一节点的本地数据的分布信息。
在一种可能的实现方式中，该第一AI模型的模型信息和该第一AI模型的辅助信息分别承载于不同的传输资源，该传输资源包括时域资源和/或频域资源。
在一种可能的实现方式中，该第一AI模型为基于该第一节点的节点类型得到。
在一种可能的实现方式中，该第一AI模型为基于第二AI模型和该第一节点的节点类型得到。
在一种可能的实现方式中，该第二AI模型基于本地数据得到；或，该第二AI模型基于K个信息得到，该K个信息中的每个信息指示其它节点的AI模型的模型信息和该其它节点的AI模型的辅助信息，K为正整数；或，该第二AI模型基于该本地数据以及该K个信息得到。
在一种可能的实现方式中，该节点类型包括以下任一项：基于本地数据进行本地训练的节点类型，基于其它节点的AI模型进行融合处理的节点类型，基于该本地数据进行本地训练并基于其它节点的AI模型进行融合处理的节点类型。
在一种可能的实现方式中，该收发单元802还用于发送指示该第一节点的节点类型的指示信息。
在一种可能的实现方式中，该处理单元801还用于基于能力信息和/或需求信息确定该第一节点的节点类型；或，该收发单元802还用于接收指示该第一节点的节点类型的指示信息。
在一种可能的实现方式中，第一AI模型为该第一节点所在的系统中M个节点可理解的模型，M为大于或等于2的整数。
一种可能的实现方式中，当该装置800为用于执行前述任一实施例中第二节点所执行的方法时，该装置800包括处理单元801和收发单元802；该收发单元802用于接收N个第一信息，该N个第一信息中的每个第一信息指示第一AI模型的模型信息和该第一AI模型的辅助信息，N为正整数；该处理单元801用于基于该N个第一信息进行模型处理，得到目标AI模型。
可选地，该目标AI模型用于完成该第二节点的AI任务，或，该目标AI模型为该第二节点的本地模型。
在一种可能的实现方式中，该第一AI模型的辅助信息包括以下至少一项：该第一AI模型的类型信息，第一节点的标识信息，该第一AI模型的接收节点的标识信息，该第一AI模型的版本信息，生成该第一AI模型的时间信息，生成该第一AI模型的地理位置信息，该第一节点的本地数据的分布信息。
在一种可能的实现方式中，该第一AI模型的模型信息和该第一AI模型的辅助信息分别承载于不同的传输资源，该传输资源包括时域资源和/或频域资源。
在一种可能的实现方式中，该处理单元801具体用于基于该N个第一信息以及该第二节点的节点类型进行模型处理，得到该目标AI模型。
在一种可能的实现方式中，在该第二节点的节点类型为基于其它节点的AI模型进行融合处理的节点类型时，该处理单元801具体用于基于该N个第一信息对该N个第一AI模型进行模型融合，得到该目标AI模型。
在一种可能的实现方式中，在该第二节点的节点类型为基于该本地数据进行本地训练并基于其它节点的AI模型进行融合处理的节点类型时，该处理单元801具体用于基于该N个第一信息对该N个第一AI模型以及第二AI模型进行模型融合，得到该目标AI模型，其中，该第二AI模型为基于本地数据训练得到。
在一种可能的实现方式中，该收发单元802还用于接收用于指示该第一节点的节点类型的指示信息。
在一种可能的实现方式中，该收发单元802还用于发送用于指示该第二节点的节点类型的指示信息。
在一种可能的实现方式中，该处理单元801还用于基于能力信息和/或需求信息确定该第二节点的节点类型；或，该收发单元802还用于接收指示该第二节点的节点类型的指示信息。
在一种可能的实现方式中，第一AI模型为该第二节点所在的系统中M个节点可理解的模型，M为大于或等于2的整数。
一种可能的实现方式中，当该装置800为用于执行前述任一实施例中第一节点所执行的方法时，该装置800包括处理单元801和收发单元802；该处理单元801用于获取本地AI模型，该本地AI模型用于完成该第一节点的AI任务；该处理单元801还用于基于该本地AI模型确定公共AI模型，该公共AI模型为该第一节点所在的系统中M个节点可理解的模型，M为大于或等于2的整数；该收发单元802用于发送第一信息，该第一信息用于指示该公共AI模型的模型信息。
在一种可能的实现方式中，该处理单元801具体用于基于该本地AI模型和该第一节点的节点类型确定公共AI模型。
在一种可能的实现方式中，该节点类型包括以下任一项：基于本地数据进行本地训练的节点类型，基于其它节点的AI模型进行融合处理的节点类型，基于该本地数据进行本地训练并基于其它节点的AI模型进行融合处理的节点类型。
在一种可能的实现方式中，该收发单元802还用于发送指示该第一节点的节点类型的指示信息。
在一种可能的实现方式中，该处理单元802还用于基于能力信息和/或需求信息确定该第一节点的节点类型；或，该收发单元802还用于接收指示该第一节点的节点类型的指示信息。
在一种可能的实现方式中，
该本地AI模型基于本地数据得到；或，
该本地AI模型基于K个信息得到，该K个信息中的每个信息指示其它节点的AI模型的模型信息和该其它节点的AI模型的辅助信息，K为正整数；或，
该本地AI模型基于该本地数据以及该K个信息得到。
在一种可能的实现方式中，该第一信息还指示该公共AI模型的辅助信息。
在一种可能的实现方式中，该公共AI模型的辅助信息包括以下至少一项：
该公共AI模型的类型信息，该第一节点的标识信息，该公共AI模型的接收节点的标识信息，该公共AI模型的版本信息，生成该公共AI模型的时间信息，生成该公共AI模型的地理位置信息，该公共节点的本地数据的分布信息。
在一种可能的实现方式中，该公共AI模型的模型信息和该公共AI模型的辅助信息分别承载于不同的传输资源，该传输资源包括时域资源和/或频域资源。
一种可能的实现方式中，当该装置800为用于执行前述任一实施例中第二节点所执行的方法时，该装置800包括处理单元801和收发单元802；该收发单元802用于接收N个第一信息，该N个第一信息中的每个第一信息用于指示公共AI模型的模型信息，该公共AI模型为第一节点所在的系统中M个节点可理解的模型，M为大于或等于2的整数；该处理单元801用于基于该N个第一信息更新本地AI模型，得到更新后的本地AI模型，该本地AI模型用于完成该第二节点的AI任务。
在一种可能的实现方式中，该处理单元801具体用于基于该N个第一信息和该第二节点的节点类型更新本地AI模型，得到更新后的本地AI模型。
在一种可能的实现方式中，该节点类型包括以下任一项：基于其它节点的AI模型进行融合处理的节点类型，基于该本地数据进行本地训练并基于其它节点的AI模型进行融合处理的节点类型。
在一种可能的实现方式中，该收发单元802还用于发送指示该第二节点的节点类型的指示信息。
在一种可能的实现方式中，该处理单元801还用于基于能力信息和/或需求信息确定该第二节点的节点类型；或，该收发单元802还用于接收指示该第二节点的节点类型的指示信息。
在一种可能的实现方式中，该节点类型为基于其它节点的AI模型进行融合处理的节点类型时，该本地AI模型基于P个信息得到，该P个信息中的每个信息指示其它节点的AI模型的模型信息和该其它节点的AI模型的辅助信息，P为正整数。
在一种可能的实现方式中，该节点类型为基于该本地数据进行本地训练并基于其它节点的AI模型进行融合处理的节点类型时，该本地AI模型基于该本地数据得到，或，该本地AI模型基于该本地数据以及该P个信息得到，该P个信息中的每个信息指示其它节点的AI模型的模型信息和该其它节点的AI模型的辅助信息，P为正整数。
在一种可能的实现方式中，该第一信息还用于指示该公共AI模型的辅助信息。
在一种可能的实现方式中，该公共AI模型的辅助信息包括以下至少一项：该公共AI模型的类型信息，该第一节点的标识信息，该公共AI模型的接收节点的标识信息，该公共AI模型的版本信息，生成该公共AI模型的时间信息，生成该公共AI模型的地理位置信息，该第一节点的本地数据的分布信息。
在一种可能的实现方式中，该公共AI模型的模型信息和该公共AI模型的辅助信息分别承载于不同的传输资源，该传输资源包括时域资源和/或频域资源。
一种可能的实现方式中，当该装置800为用于执行前述任一实施例中第一节点所执行的方法时，该装置800包括处理单元801和收发单元802；该处理单元801用于第一节点确定该第一节点的节点类型。
在一种可能的实现方式中，该装置还包括收发单元，该收发单元802用于发送指示该第一节点的节点类型的指示信息。
在一种可能的实现方式中，该装置还包括收发单元802，该收发单元802用于发送第一信息，该第一信息用于指示该第一AI模型的模型信息，其中，该第一AI模型为基于该第一节点的节点类型得到。
可选地，该第一AI模型为基于第二AI模型和该第一节点的节点类型得到。
在一种可能的实现方式中，
该第二AI模型基于本地数据得到；或，
该第二AI模型基于K个信息得到，该K个信息中的每个信息指示其它节点的AI模型的模型信息和该其它节点的AI模型的辅助信息，K为正整数；或，
该第二AI模型基于该本地数据以及该K个信息得到。
在一种可能的实现方式中，该节点类型包括以下任一项：基于本地数据进行本地训练的节点类型，基于其它节点的AI模型进行融合处理的节点类型，基于该本地数据进行本地训练并基于其它节点的AI模型进行融合处理的节点类型。
在一种可能的实现方式中，该第一信息还用于指示该第一AI模型的辅助信息，其中，该第一AI模型的辅助信息包括以下至少一项：该第一AI模型的类型信息，该第一节点的标识信息，该第一AI模型的接收节点的标识信息，该第一AI模型的版本信息，生成该第一AI模型的时间信息，生成该第一AI模型的地理位置信息，该第一节点的本地数据的分布信息。
在一种可能的实现方式中，该第一AI模型的模型信息和该第一AI模型的辅助信息分别承载于不同的传输资源，该传输资源包括时域资源和/或频域资源。
在一种可能的实现方式中，第一AI模型为该第一节点所在的系统中M个节点可理解的模型，M为大于或等于2的整数。
在一种可能的实现方式中，该处理单元801还用于基于能力信息和/或需求信息确定该第一节点的节点类型；或，该收发单元802还用于接收指示该第一节点的节点类型的指示信息。
一种可能的实现方式中，当该装置800为用于执行前述任一实施例中第二节点所执行的方法时，该装置800包括收发单元802；该收发单元802用于第二节点接收指示该第一节点的节点类型的指示信息；和/或，该收发单元802用于接收第一信息，该第一信息用于指示该第一AI模型的模型信息，其中，该第一AI模型为基于该第一节点的节点类型得到。
可选地，该第一AI模型为基于第二AI模型和该第一节点的节点类型得到。
在一种可能的实现方式中，
该第二AI模型基于本地数据得到；或，
该第二AI模型基于K个信息得到，该K个信息中的每个信息指示其它节点的AI模型的模型信息和该其它节点的AI模型的辅助信息，K为正整数；或，
该第二AI模型基于该本地数据以及该K个信息得到。
在一种可能的实现方式中，该节点类型包括以下任一项：基于本地数据进行本地训练的节点类型，基于其它节点的AI模型进行融合处理的节点类型，基于该本地数据进行本地训练并基于其它节点的AI模型进行融合处理的节点类型。
在一种可能的实现方式中，该第一信息还用于指示该第一AI模型的辅助信息，其中，该第一AI模型的辅助信息包括以下至少一项：该第一AI模型的类型信息，该第一节点的标识信息，该第一AI模型的接收节点的标识信息，该第一AI模型的版本信息，生成该第一AI模型的时间信息，生成该第一AI模型的地理位置信息，该第一节点的本地数据的分布信息。
在一种可能的实现方式中，该第一AI模型的模型信息和该第一AI模型的辅助信息分别承载于不同的传输资源，该传输资源包括时域资源和/或频域资源。
在一种可能的实现方式中，第一AI模型为该第一节点所在的系统中M个节点可理解的模型，M为大于或等于2的整数。
在一种可能的实现方式中，该装置还包括处理单元801，该处理单元801用于基于能力信息和/或需求信息确定该第二节点的节点类型；或，该收发单元还用于接收指示该第二节点的节点类型的指示信息。
需要说明的是，上述通信装置800的单元的信息执行过程等内容，具体可参见本申请前述所示的方法实施例中的叙述，此处不再赘述。
请参阅图9，为本申请提供的通信装置900的另一种示意性结构图，通信装置900至少包括输入输出接口902。其中，通信装置900可以为芯片或集成电路。
可选的，该通信装置还包括逻辑电路901。
其中，图8所示收发单元802可以为通信接口，该通信接口可以是图9中的输入输出接口902，该输入输出接口902可以包括输入接口和输出接口。或者，该通信接口也可以是收发电路，该收发电路可以包括输入接口电路和输出接口电路。
可选的，逻辑电路901用于确定第一AI模型；输入输出接口902用于发送第一信息，该第一信息指示该第一AI模型的模型信息和该第一AI模型的辅助信息。
可选的，输入输出接口902用于接收N个第一信息，该N个第一信息中的每个第一信息指示第一AI模型的模型信息和该第一AI模型的辅助信息；逻辑电路901用于基于该N个第一信息进行模型处理，得到目标AI模型。
可选的，逻辑电路901用于获取本地AI模型，该本地AI模型用于完成该第一节点的AI任务；逻辑电路901用于基于该本地AI模型确定公共AI模型，该公共AI模型为该第一节点所在的系统中M个节点可理解的模型，M为大于或等于2的整数；输入输出接口902用于发送第一信息，该第一信息用于指示该公共AI模型的模型信息。
可选的，输入输出接口902用于接收N个第一信息，该N个第一信息中的每个第一信息用于指示公共AI模型的模型信息，该公共AI模型为第一节点所在的系统中M个节点可理解的模型，M为大于或等于2的整数；逻辑电路901用于基于该公共AI模型的模型信息更新本地AI模型，得到更新后的本地AI模型，该本地AI模型用于完成该第二节点的AI任务。
可选的，逻辑电路901用于用于第一节点确定该第一节点的节点类型。
可选地，输入输出接口902用于第二节点接收指示该第一节点的节点类型的指示信息；
可选地，输入输出接口902用于接收第一信息，该第一信息用于指示该第一AI模型的模型信息，其中，该第一AI模型为基于该第一节点的节点类型得到。
其中，逻辑电路901和输入输出接口902还可以执行任一实施例中网络设备执行的其他步骤并实现对应的有益效果，此处不再赘述。
在一种可能的实现方式中，图8所示处理单元801可以为图9中的逻辑电路901。
可选的，逻辑电路901可以是一个处理装置，处理装置的功能可以部分或全部通过软件实现。其中，处理装置的功能可以部分或全部通过软件实现。
可选的，处理装置可以包括存储器和处理器，其中，存储器用于存储计算机程序，处理器读取并执行存储器中存储的计算机程序，以执行任意一个方法实施例中的相应处理和/或步骤。
可选地，处理装置可以仅包括处理器。用于存储计算机程序的存储器位于处理装置之外，处理器通过电路/电线与存储器连接，以读取并执行存储器中存储的计算机程序。其中，存储器和处理器可以集成在一起，或者也可以是物理上互相独立的。
可选地，该处理装置可以是一个或多个芯片，或一个或多个集成电路。例如，处理装置可以是一个或多个现场可编程门阵列(field-programmable gate array，FPGA)、专用集成芯片(application specific integrated circuit，ASIC)、系统芯片(system on chip，SoC)、中央处理器(central processor unit，CPU)、网络处理器(network processor，NP)、数字信号处理电路(digital signal processor，DSP)、微控制器(micro controller unit，MCU)，可编程控制器(programmable logic device，PLD)或其它集成芯片，或者上述芯片或者处理器的任意组合等。
请参阅图10，为本申请的实施例提供的上述实施例中所涉及的通信装置1000，该通信装置1000具体可以为上述实施例中的作为终端设备的通信装置，图10所示示例为终端设备通过终端设备(或者终端设备中的部件)实现。
其中，该通信装置1000的一种可能的逻辑结构示意图，该通信装置1000可以包括但不限于至少一个处理器1001以及通信端口1002。
进一步可选的，该装置还可以包括存储器1003、总线1004中的至少一个，在本申请的实施例中，该至少一个处理器1001用于对通信装置1000的动作进行控制处理。
此外，处理器1001可以是中央处理器单元，通用处理器，数字信号处理器，专用集成电路，现场可编程门阵列或者其他可编程逻辑器件、晶体管逻辑器件、硬件部件或者其任意组合。其可以实现或执行结合本申请公开内容所描述的各种示例性的逻辑方框，模块和电路。该处理器也可以是实现计算功能的组合，例如包含一个或多个微处理器组合，数字信号处理器和微处理器的组合等等。所属领域的技术人员可以清楚地了解到，为描述的方便和简洁，上述描述的系统，装置和单元的具体工作过程，可以参考前述方法实施例中的对应过程，在此不再赘述。
需要说明的是，图10所示通信装置1000具体可以用于实现前述方法实施例中终端设备所实现的步骤，并实现终端设备对应的技术效果，图10所示通信装置的具体实现方式，均可以参考前述方法实施例中的叙述，此处不再一一赘述。
请参阅图11，为本申请的实施例提供的上述实施例中所涉及的通信装置1100的结构示意图，该通信装置1100具体可以为上述实施例中的作为网络设备的通信装置，图11所示示例为网络设备通过网络设备(或者网络设备中的部件)实现，其中，该通信装置的结构可以参考图11所示的结构。
通信装置1100包括至少一个处理器1111以及至少一个网络接口1114。进一步可选的，该通信装置还包括至少一个存储器1112、至少一个收发器1113和一个或多个天线1115。处理器1111、存储器1112、收发器1113和网络接口1114相连，例如通过总线相连，在本申请实施例中，该连接可包括各类接口、传输线或总线等，本实施例对此不做限定。天线1115与收发器1113相连。网络接口1114用于使得通信装置通过通信链路，与其它通信设备通信。例如网络接口1114可以包括通信装置与核心网设备之间的网络接口，例如S1接口，网络接口可以包括通信装置和其他通信装置(例如其他网络设备或者核心网设备)之间的网络接口，例如X2或者Xn接口。
处理器1111主要用于对通信协议以及通信数据进行处理，以及对整个通信装置进行控制，执行软件程序，处理软件程序的数据，例如用于支持通信装置执行实施例中所描述的动作。通信装置可以包括基带处理器和中央处理器，基带处理器主要用于对通信协议以及通信数据进行处理，中央处理器主要用于对整个终端设备进行控制，执行软件程序，处理软件程序的数据。图11中的处理器1111可以集成基带处理器和中央处理器的功能，本领域技术人员可以理解，基带处理器和中央处理器也可以是各自独立的处理器，通过总线等技术互联。本领域技术人员可以理解，终端设备可以包括多个基带处理器以适应不同的网络 制式，终端设备可以包括多个中央处理器以增强其处理能力，终端设备的各个部件可以通过各种总线连接。该基带处理器也可以表述为基带处理电路或者基带处理芯片。该中央处理器也可以表述为中央处理电路或者中央处理芯片。对通信协议以及通信数据进行处理的功能可以内置在处理器中，也可以以软件程序的形式存储在存储器中，由处理器执行软件程序以实现基带处理功能。
存储器主要用于存储软件程序和数据。存储器1112可以是独立存在，与处理器1111相连。可选的，存储器1112可以和处理器1111集成在一起，例如集成在一个芯片之内。其中，存储器1112能够存储执行本申请实施例的技术方案的程序代码，并由处理器1111来控制执行，被执行的各类计算机程序代码也可被视为是处理器1111的驱动程序。
图11仅示出了一个存储器和一个处理器。在实际的终端设备中，可以存在多个处理器和多个存储器。存储器也可以称为存储介质或者存储设备等。存储器可以为与处理器处于同一芯片上的存储元件，即片内存储元件，或者为独立的存储元件，本申请实施例对此不做限定。
收发器1113可以用于支持通信装置与终端之间射频信号的接收或者发送，收发器1113可以与天线1115相连。收发器1113包括发射机Tx和接收机Rx。具体地，一个或多个天线1115可以接收射频信号，该收发器1113的接收机Rx用于从天线接收该射频信号，并将射频信号转换为数字基带信号或数字中频信号，并将该数字基带信号或数字中频信号提供给该处理器1111，以便处理器1111对该数字基带信号或数字中频信号做进一步的处理，例如解调处理和译码处理。此外，收发器1113中的发射机Tx还用于从处理器1111接收经过调制的数字基带信号或数字中频信号，并将该经过调制的数字基带信号或数字中频信号转换为射频信号，并通过一个或多个天线1115发送该射频信号。具体地，接收机Rx可以选择性地对射频信号进行一级或多级下混频处理和模数转换处理以得到数字基带信号或数字中频信号，该下混频处理和模数转换处理的先后顺序是可调整的。发射机Tx可以选择性地对经过调制的数字基带信号或数字中频信号时进行一级或多级上混频处理和数模转换处理以得到射频信号，该上混频处理和数模转换处理的先后顺序是可调整的。数字基带信号和数字中频信号可以统称为数字信号。
收发器1113也可以称为收发单元、收发机、收发装置等。可选的，可以将收发单元中用于实现接收功能的器件视为接收单元，将收发单元中用于实现发送功能的器件视为发送单元，即收发单元包括接收单元和发送单元，接收单元也可以称为接收机、输入口、接收电路等，发送单元可以称为发射机、发射器或者发射电路等。
需要说明的是，图11所示通信装置1100具体可以用于实现前述方法实施例中网络设备所实现的步骤，并实现网络设备对应的技术效果，图11所示通信装置1100的具体实现方式，均可以参考前述方法实施例中的叙述，此处不再一一赘述。
本申请实施例还提供一种存储一个或多个计算机执行指令的计算机可读存储介质，当计算机执行指令被处理器执行时，该处理器执行如前述实施例中终端设备可能的实现方式所述的方法。
本申请实施例还提供一种存储一个或多个计算机执行指令的计算机可读存储介质，当计算机执行指令被处理器执行时，该处理器执行如前述实施例中网络设备可能的实现方式所述的方法。
本申请实施例还提供一种存储一个或多个计算机的计算机程序产品(或称计算机程序)，当计算机程序产品被该处理器执行时，该处理器执行上述终端设备可能实现方式的方法。
本申请实施例还提供一种存储一个或多个计算机的计算机程序产品，当计算机程序产品被该处理器执行时，该处理器执行上述网络设备可能实现方式的方法。
本申请实施例还提供了一种芯片系统，该芯片系统包括至少一个处理器，用于支持通信装置实现上述通信装置可能的实现方式中所涉及的功能。可选的，所述芯片系统还包括接口电路，所述接口电路为所述至少一个处理器提供程序指令和/或数据。在一种可能的设计中，该芯片系统还可以包括存储器，存储器，用于保存该通信装置必要的程序指令和数据。该芯片系统，可以由芯片构成，也可以包含芯片和其他分立器件，其中，该通信装置具体可以为前述方法实施例中终端设备。
本申请实施例还提供了一种芯片系统，该芯片系统包括至少一个处理器，用于支持通信装置实现上述通信装置可能的实现方式中所涉及的功能。可选的，所述芯片系统还包括接口电路，所述接口电路为所述至少一个处理器提供程序指令和/或数据。在一种可能的设计中，芯片系统还可以包括存储器，存储器，用于保存该通信装置必要的程序指令和数据。该芯片系统，可以由芯片构成，也可以包含芯片和其他分立器件，其中，该通信装置具体可以为前述方法实施例中网络设备。
本申请实施例还提供了一种通信系统，该网络系统架构包括上述任一实施例中的第一节点和第二节点，该第一节点可以为终端设备或网络设备，该第二节点也可以为终端设备或网络设备。
在本申请所提供的几个实施例中，应该理解到，所揭露的系统，装置和方法，可以通过其它的方式实现。例如，以上所描述的装置实施例仅仅是示意性的，例如，所述单元的划分，仅仅为一种逻辑功能划分，实际实现时可以有另外的划分方式，例如多个单元或组件可以结合或者可以集成到另一个系统，或一些特征可以忽略，或不执行。另一点，所显示或讨论的相互之间的耦合或直接耦合或通信连接可以是通过一些接口，装置或单元的间接耦合或通信连接，可以是电性，机械或其它的形式。
所述作为分离部件说明的单元可以是或者也可以不是物理上分开的，作为单元显示的部件可以是或者也可以不是物理单元，即可以位于一个地方，或者也可以分布到多个网络单元上。可以根据实际的需要选择其中的部分或者全部单元来实现本实施例方案的目的。
另外，在本申请各个实施例中的各功能单元可以集成在一个处理单元中，也可以是各个单元单独物理存在，也可以两个或两个以上单元集成在一个单元中。上述集成的单元既可以采用硬件的形式实现，也可以采用软件功能单元的形式实现。所述集成的单元如果以软件功能单元的形式实现并作为独立的产品销售或使用时，可以存储在一个计算机可读取存储介质中。基于这样的理解，本申请的技术方案本质上或者说对现有技术做出贡献的部 分或者该技术方案的全部或部分可以以软件产品的形式体现出来，该计算机软件产品存储在一个存储介质中，包括若干指令用以使得一台计算机设备(可以是个人计算机，服务器，或者网络设备等)执行本申请各个实施例所述方法的全部或部分步骤。而前述的存储介质包括：U盘、移动硬盘、只读存储器(ROM，Read-Only Memory)、随机存取存储器(RAM，Random Access Memory)、磁碟或者光盘等各种可以存储程序代码的介质。

Claims (43)

  1. 一种人工智能AI模型处理方法，其特征在于，包括：
    第一节点确定第一AI模型；
    所述第一节点发送第一信息，所述第一信息指示所述第一AI模型的模型信息和所述第一AI模型的辅助信息。
  2. 根据权利要求1所述的方法，其特征在于，所述第一AI模型的辅助信息包括以下至少一项：
    所述第一AI模型的类型信息，所述第一节点的标识信息，所述第一AI模型的接收节点的标识信息，所述第一AI模型的版本信息，生成所述第一AI模型的时间信息，生成所述第一AI模型的地理位置信息，所述第一节点的本地数据的分布信息。
  3. 根据权利要求1或2所述的方法，其特征在于，所述第一AI模型的模型信息和所述第一AI模型的辅助信息分别承载于不同的传输资源，所述传输资源包括时域资源和/或频域资源。
  4. 根据权利要求1至3任一项所述的方法，其特征在于，所述第一AI模型为基于所述第一节点的节点类型得到。
  5. 根据权利要求1至4任一项所述的方法，其特征在于，所述第一AI模型为基于第二AI模型和所述第一节点的节点类型得到；
    所述第二AI模型基于本地数据得到；或，
    所述第二AI模型基于K个信息得到，所述K个信息中的每个信息指示其它节点的AI模型的模型信息和所述其它节点的AI模型的辅助信息，所述K为正整数；或，
    所述第二AI模型基于所述本地数据以及所述K个信息得到。
  6. 根据权利要求4或5所述的方法，其特征在于，所述节点类型包括以下任一项：
    基于本地数据进行本地训练的节点类型，基于其它节点的AI模型进行融合处理的节点类型，基于所述本地数据进行本地训练并基于其它节点的AI模型进行融合处理的节点类型。
  7. 根据权利要求4至6任一项所述的方法，其特征在于，所述方法还包括：
    所述第一节点发送指示所述第一节点的节点类型的指示信息。
  8. 根据权利要求4至7任一项所述的方法，其特征在于，所述方法还包括：
    所述第一节点基于能力信息和/或需求信息确定所述第一节点的节点类型；或，
    所述第一节点接收指示所述第一节点的节点类型的指示信息。
  9. 根据权利要求1至8任一项所述的方法，其特征在于，
    所述第一AI模型为所述第一节点所在的系统中M个节点可理解的模型，M为大于或等于2的整数。
  10. 一种人工智能AI模型处理方法，其特征在于，包括：
    第二节点接收来自N个第一节点的N个第一信息，所述N个第一信息中的每个第一信息指示第一AI模型的模型信息和所述第一AI模型的辅助信息，N为正整数；
    所述第二节点基于所述N个第一信息进行模型处理，得到目标AI模型。
  11. 根据权利要求10所述的方法，其特征在于，所述第一AI模型的辅助信息包括以下至少一项：
    所述第一AI模型的类型信息，第一节点的标识信息，所述第一AI模型的接收节点的标识信息，所述第一AI模型的版本信息，生成所述第一AI模型的时间信息，生成所述第一AI模型的地理位置信息，所述第一节点的本地数据的分布信息。
  12. 根据权利要求10或11所述的方法，其特征在于，所述第一AI模型的模型信息和所述第一AI模型的辅助信息分别承载于不同的传输资源，所述传输资源包括时域资源和/或频域资源。
  13. 根据权利要求10至12任一项所述的方法，其特征在于，所述第二节点基于所述N个第一信息进行模型处理，得到目标AI模型包括：
    所述第二节点基于所述N个第一信息和所述第二节点的节点类型进行模型处理，得到所述目标AI模型。
  14. 根据权利要求13所述的方法，其特征在于，所述第二节点基于所述N个第一信息和所述第二节点的节点类型进行模型处理，得到所述目标AI模型包括：
    在所述第二节点的节点类型为基于其它节点的AI模型进行融合处理的节点类型时，所述第二节点基于所述N个第一信息对所述N个第一AI模型进行模型融合，得到所述目标AI模型。
  15. 根据权利要求13或14所述的方法，其特征在于，所述第二节点基于所述N个第一信息和所述第二节点的节点类型进行模型处理，得到所述目标AI模型包括：
    在所述第二节点的节点类型为基于所述本地数据进行本地训练并基于其它节点的AI模型进行融合处理的节点类型时，所述第二节点基于所述N个第一信息对所述N个第一AI模型以及第二AI模型进行模型融合，得到所述目标AI模型，其中，所述第二AI模型为基于本地数据训练得到。
  16. 根据权利要求13至15任一项所述的方法，其特征在于，所述方法还包括：
    所述第二节点接收用于指示所述第一节点的节点类型的指示信息。
  17. 根据权利要求13至16任一项所述的方法，其特征在于，所述方法还包括：
    所述第二节点发送用于指示所述第二节点的节点类型的指示信息。
  18. 根据权利要求13至17任一项所述的方法，其特征在于，所述方法还包括：
    所述第二节点基于能力信息和/或需求信息确定所述第二节点的节点类型；或，
    所述第二节点接收指示所述第二节点的节点类型的指示信息。
  19. 根据权利要求10至18任一项所述的方法，其特征在于，
    第一AI模型为所述第二节点所在的系统中M个节点可理解的模型，M为大于或等于2的整数。
  20. 一种人工智能AI模型处理装置，其特征在于，包括处理单元和收发单元；
    所述处理单元用于确定第一AI模型；
    所述收发单元用于发送第一信息，所述第一信息指示所述第一AI模型的模型信息和所述第一AI模型的辅助信息。
  21. 根据权利要求20所述的装置，其特征在于，所述第一AI模型的辅助信息包括以下至少一项：
    所述第一AI模型的类型信息，所述装置对应的节点的标识信息，所述第一AI模型的接收节点的标识信息，所述第一AI模型的版本信息，生成所述第一AI模型的时间信息，生成所述第一AI模型的地理位置信息，所述装置对应的节点的本地数据的分布信息。
  22. 根据权利要求20或21所述的装置，其特征在于，所述第一AI模型的模型信息和所述第一AI模型的辅助信息分别承载于不同的传输资源，所述传输资源包括时域资源和/或频域资源。
  23. 根据权利要求20至22任一项所述的装置，其特征在于，所述第一AI模型为基于所述装置对应的的节点类型得到。
  24. 根据权利要求23所述的装置，其特征在于，所述第一AI模型为基于第二AI模型和所述节点类型得到；
    所述第二AI模型基于本地数据得到；或，
    所述第二AI模型基于K个信息得到，所述K个信息中的每个信息指示其它节点的AI模型的模型信息和所述其它节点的AI模型的辅助信息，所述K为正整数；或，
    所述第二AI模型基于所述本地数据以及所述K个信息得到。
  25. 根据权利要求23或24所述的装置，其特征在于，所述节点类型包括以下任一项：
    基于本地数据进行本地训练的节点类型，基于其它节点的AI模型进行融合处理的节点类型，基于所述本地数据进行本地训练并基于其它节点的AI模型进行融合处理的节点类型。
  26. 根据权利要求23至25任一项所述的装置，其特征在于，
    所述收发单元还用于发送指示所述装置对应的节点类型的指示信息。
  27. 根据权利要求23至26任一项所述的装置，其特征在于，
    所述处理单元还用于基于能力信息和/或需求信息确定所述装置对应的节点的节点类型；或，
    所述收发单元还用于接收指示所述装置对应的节点的节点类型的指示信息。
  28. 根据权利要求20至27任一项所述的装置，其特征在于，
    第一AI模型为所述装置对应的节点所在的系统中M个节点可理解的模型，M为大于或等于2的整数。
  29. 一种人工智能AI模型处理装置，其特征在于，包括处理单元和收发单元；
    所述收发单元用于接收来自N个第一节点的N个第一信息，所述N个第一信息中的每个第一信息指示第一AI模型的模型信息和所述第一AI模型的辅助信息，N为正整数；
    所述处理单元用于基于所述N个第一信息进行模型处理，得到目标AI模型。
  30. 根据权利要求29所述的装置，其特征在于，所述第一AI模型的辅助信息包括以下至少一项：
    所述第一AI模型的类型信息，第一节点的标识信息，所述第一AI模型的接收节点的标识信息，所述第一AI模型的版本信息，生成所述第一AI模型的时间信息，生成所述第一AI模型的地理位置信息，所述第一节点的本地数据的分布信息。
  31. 根据权利要求29或30所述的装置，其特征在于，所述第一AI模型的模型信息和所述第一AI模型的辅助信息分别承载于不同的传输资源，所述传输资源包括时域资源和/或频域资源。
  32. 根据权利要求29至31任一项所述的装置，其特征在于，所述处理单元具体用于基于所述N个第一信息和所述处理装置对应的节点类型进行模型处理，得到所述目标AI模型。
  33. 根据权利要求32所述的装置，其特征在于，在所述节点类型为基于其它节点的AI 模型进行融合处理的节点类型时，所述处理单元具体用于基于所述N个第一信息对所述N个第一AI模型进行模型融合，得到所述目标AI模型。
  34. 根据权利要求32或33所述的装置，其特征在于，在所述节点类型为基于所述本地数据进行本地训练并基于其它节点的AI模型进行融合处理的节点类型时，所述处理单元具体用于基于所述N个第一信息对所述N个第一AI模型以及第二AI模型进行模型融合，得到所述目标AI模型，其中，所述第二AI模型为基于本地数据训练得到。
  35. 根据权利要求32至34任一项所述的装置，其特征在于，
    所述收发单元还用于接收用于指示所述第一节点的节点类型的指示信息。
  36. 根据权利要求32至35任一项所述的装置，其特征在于，
    所述收发单元还用于发送用于指示所述装置对应的节点类型的指示信息。
  37. 根据权利要求32至36任一项所述的装置，其特征在于，
    所述处理单元还用于基于能力信息和/或需求信息确定所述装置对应的节点类型；或，
    所述收发单元还用于接收指示所述装置对应的节点类型的指示信息。
  38. 根据权利要求29至37任一项所述的装置，其特征在于，
    第一AI模型为所述装置所在的系统中M个节点可理解的模型，M为大于或等于2的整数。
  39. 一种通信装置，其特征在于，包括至少一个逻辑电路和输入输出接口；
    所述至少一个逻辑电路用于确定第一AI模型；
    所述输入输出接口用于输出第一信息，所述第一信息指示所述第一AI模型的模型信息和所述第一AI模型的辅助信息；
    所述逻辑电路还用于执行如权利要求1至9中任一项所述的方法。
  40. 一种通信装置，其特征在于，包括至少一个逻辑电路和输入输出接口；
    所述输入输出接口用于输入N个第一信息，所述N个第一信息中的每个第一信息指示第一AI模型的模型信息和所述第一AI模型的辅助信息，N为正整数；
    所述至少一个逻辑电路用于基于所述N个第一信息进行模型处理，得到目标AI模型；
    所述逻辑电路还用于执行如权利要求10至19中任一项所述的方法。
  41. 一种通信系统，其特征在于，
    所述通信系统包括如权利要求20至28中任一项所述的装置，以及如权利要求29至38中任一项所述的装置；
    或者，
    所述通信系统包括权利要求39所述的装置和权利要求40所述的装置。
  42. 一种计算机可读存储介质，其特征在于，所述介质存储有指令，当所述指令被计算机执行时，实现权利要求1至19中任一项所述的方法。
  43. 一种计算机程序产品，其特征在于，包括指令，当所述指令在计算机上运行时，使得计算机执行如权利要求1至19中任一项所述的方法。
PCT/CN2022/110616 2022-08-05 2022-08-05 一种人工智能模型处理方法及相关设备 WO2024026846A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
PCT/CN2022/110616 WO2024026846A1 (zh) 2022-08-05 2022-08-05 一种人工智能模型处理方法及相关设备

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2022/110616 WO2024026846A1 (zh) 2022-08-05 2022-08-05 一种人工智能模型处理方法及相关设备

Publications (1)

Publication Number Publication Date
WO2024026846A1 true WO2024026846A1 (zh) 2024-02-08

Family

ID=89848283

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/110616 WO2024026846A1 (zh) 2022-08-05 2022-08-05 一种人工智能模型处理方法及相关设备

Country Status (1)

Country Link
WO (1) WO2024026846A1 (zh)

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111612153A (zh) * 2019-02-22 2020-09-01 华为技术有限公司 训练模型的方法和装置
CN112070240A (zh) * 2020-09-07 2020-12-11 清华大学 一种高效通信的分层联邦学习框架及其优化方法和系统
CN112765677A (zh) * 2020-12-30 2021-05-07 杭州溪塔科技有限公司 一种基于区块链的联邦学习方法、装置和系统
CN113919512A (zh) * 2021-09-26 2022-01-11 重庆邮电大学 基于计算资源逻辑分层的联邦学习通信优化方法及系统
CN113935469A (zh) * 2021-10-26 2022-01-14 城云科技（中国）有限公司 基于去中心化联邦学习的模型训练方法
CN114491623A (zh) * 2021-12-30 2022-05-13 北京邮电大学 一种基于区块链的异步联邦学习方法及系统

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111612153A (zh) * 2019-02-22 2020-09-01 华为技术有限公司 训练模型的方法和装置
CN112070240A (zh) * 2020-09-07 2020-12-11 清华大学 一种高效通信的分层联邦学习框架及其优化方法和系统
CN112765677A (zh) * 2020-12-30 2021-05-07 杭州溪塔科技有限公司 一种基于区块链的联邦学习方法、装置和系统
CN113919512A (zh) * 2021-09-26 2022-01-11 重庆邮电大学 基于计算资源逻辑分层的联邦学习通信优化方法及系统
CN113935469A (zh) * 2021-10-26 2022-01-14 城云科技（中国）有限公司 基于去中心化联邦学习的模型训练方法
CN114491623A (zh) * 2021-12-30 2022-05-13 北京邮电大学 一种基于区块链的异步联邦学习方法及系统

Similar Documents

Publication Publication Date Title
WO2021243619A1 (zh) 信息传输方法、装置、通信设备及存储介质
US20210326701A1 (en) Architecture for machine learning (ml) assisted communications networks
WO2021244334A1 (zh) 一种信息处理方法及相关设备
WO2022121804A1 (zh) 半异步联邦学习的方法和通信装置
WO2023020502A1 (zh) 数据处理方法及装置
CN116848828A (zh) 机器学习模型分布
Jere et al. Distributed learning meets 6G: A communication and computing perspective
WO2024026846A1 (zh) 一种人工智能模型处理方法及相关设备
WO2023104169A1 (zh) 一种无线网络中的人工智能ai模型训练方法及装置
WO2023098860A1 (zh) 通信方法和通信装置
WO2023040971A1 (zh) 一种人工智能模型下载方法、装置及系统
US11997693B2 (en) Lower analog media access control (MAC-A) layer and physical layer (PHY-A) functions for analog transmission protocol stack
WO2022001822A1 (zh) 获取神经网络的方法和装置
WO2024061125A1 (zh) 一种通信方法及装置
WO2023125599A1 (zh) 一种通信方法及通信装置
WO2023125598A1 (zh) 一种通信方法及通信装置
WO2024060139A1 (zh) 通信方法、装置、存储介质及程序产品
WO2023226650A1 (zh) 一种模型训练的方法和装置
WO2022268027A1 (zh) 一种gan的训练方法、机器学习系统及通信装置
WO2023231706A1 (zh) 数据信道模型的发送方法及装置、信息发送方法及装置
WO2023115254A1 (zh) 处理数据的方法及装置
WO2023179577A1 (zh) 一种通信方法及相关装置
WO2022183362A1 (zh) 通信方法、设备及存储介质
WO2023169389A1 (zh) 通信方法和通信装置
WO2024044881A1 (zh) 一种数据处理方法、训练方法及相关装置

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22953651

Country of ref document: EP

Kind code of ref document: A1