Communication method, and apparatus

Abstract

Provided are a communication method and an apparatus, which relate to the technical field of communications, and are used for improving the performance of realizing distributed machine learning by multiple nodes in a communication system and the security of data transmission. The method comprises: receiving a first message which comprises intermediate data of machine learning, the intermediate data being from a first apparatus, and the first apparatus being a participating node of a machine learning task; verifying the intermediate data to determine that there are abnormal intermediate data; and sending a second message used for indicating abnormal data feedback, the second message further comprising a first policy used for indicating processing performed on the abnormal intermediate data or the first apparatus.

Description

Communication method and apparatus

[0001] The present application claims priority from the Chinese patent application No. 202410869532.2 filed on June 29, 2024, and entitled "A communication method and apparatus", the content of which is incorporated herein by reference in its entirety. TECHNICAL FIELD

[0002] The present application relates to the field of communication technology, and in particular to a communication method and apparatus. BACKGROUND

[0003] In the face of the needs of future communication network complexity, business demand diversification or business experience individualization, artificial intelligence (AI) can be introduced into the communication network to provide on-demand services and higher network resource utilization. For example, network data analysis function (NWDAF) can be used to analyze and process various network data, such as collecting network operation data from network function nodes, obtaining terminal and network related statistical data from operation administration and maintenance (OAM), or obtaining application data from third party application function entities (AF). Subsequently, the collected data can be used for data training or model inference.

[0004] Among them, multiple network nodes (such as NWDAF) can be used in the communication system to implement distributed machine learning (such as federated learning), and multiple nodes can implement common modeling to improve the use effect of AI model. For example, federated learning is performed between multiple network nodes, and the initiating node of federated learning can request multiple participating nodes to collect data, and the multiple participating nodes can collect data and send it to the initiating node of federated learning in succession. At present, how to improve the performance and data security of distributed machine learning between network nodes is a problem to be solved. SUMMARY

[0005] The present application provides a communication method and apparatus for improving the performance and data security of distributed machine learning implemented by multiple nodes in a communication system.

[0006] In a first aspect, a communication method is provided, which can be performed by a communication apparatus. The communication apparatus can be a network device, or a module (e.g., a chip, a chip system, or a circuit) in the network device, or a module or software that can implement all or part of the functions of the network device. For example, the communication apparatus can be an initiating node or a coordinating node of machine learning, etc. The method comprises: receiving a first message, wherein the first message comprises intermediate data of machine learning, and the intermediate data is from a first apparatus, which is a participating node of the machine learning task; verifying the intermediate data, and determining that there is abnormal intermediate data; and sending a second message, wherein the second message is used to indicate that there is an abnormal data feedback, and the second message further comprises a first policy, which is used to indicate a processing performed on the abnormal intermediate data or the first apparatus.

[0007] In the above embodiments, the intermediate data reported by the participating node of machine learning is verified, and whether the model is updated according to the intermediate data is determined according to the verification result, or the corresponding processing is performed on the abnormal intermediate data or the participating node providing the abnormal intermediate data, so that the model deviation caused by the reporting of abnormal data can be avoided, and the effect of machine learning is improved.

[0008] In an embodiment, the second message comprises information of the abnormal intermediate data and / or information of the first apparatus corresponding to the abnormal intermediate data. That is, the abnormal intermediate data is indicated by the second message, or the participating node providing the abnormal data is indicated, so that the abnormal data is fed back, the model deviation caused by the reporting of abnormal data can be avoided, and the effect of machine learning is improved.

[0009] In an embodiment, the first policy is used to indicate at least one of the following information: discarding the intermediate data from the first apparatus; discarding N times of intermediate data from the first apparatus, wherein N is a positive integer; marking the first apparatus as a first type or a second type; or updating the intermediate data.

[0010] In the above embodiments, the abnormal intermediate data is discarded or updated, or the participating node providing the abnormal data is marked, such as being marked as a black list node or a gray list node, so that the model deviation caused by the reporting of abnormal data can be avoided, the flexibility of processing the abnormal data is improved, and the effect of machine learning is improved.

[0011] In an embodiment, the method further comprises: if the intermediate data corresponding to the first apparatus is abnormal, marking the first apparatus as a first type.

[0012] In an embodiment, the method further comprises: if the number of times that the intermediate data corresponding to the first device is abnormal is greater than or equal to m, marking the first device as a second type, wherein m is a positive integer.

[0013] In the above embodiment, by marking the participating node providing abnormal data, such as marking as a blacklisted or graylisted node, the model deviation caused by the reporting of abnormal data can be avoided, the flexibility of processing abnormal data is improved, and the effect of machine learning is improved.

[0014] In an embodiment, the verifying of the intermediate data of the machine learning comprises: determining a stage of the machine learning; and determining whether the intermediate data is abnormal according to the stage of the machine learning. Considering that the data characteristics of different nodes in the machine learning process are different, when verifying the intermediate data, the node at which the machine learning is currently located can be determined, and then whether the received intermediate data is abnormal can be determined according to the characteristics of the data at this stage, thereby improving the flexibility and accuracy of data verification and improving the performance of machine learning.

[0015] In an embodiment, if the stage of the machine learning is in a convergence stage, the verifying of the intermediate data of the machine learning comprises at least one of the following: verifying a variance of a gradient in the intermediate data of the machine learning, verifying an information divergence of the gradient, or verifying whether the gradient exceeds a deviation value.

[0016] In the above embodiment, for a machine learning task in a convergence node, the change of gradient information in the intermediate data is not too large, and therefore, whether the intermediate data is abnormal can be verified by verifying the variance of the gradient, the information divergence of the gradient, or the deviation value of the gradient, thereby improving the realizability and accuracy of data verification and improving the performance of machine learning.

[0017] In an embodiment, the first message further comprises an identifier of the first device. Wherein the communication device receives the intermediate data through the first message, and the identifier of the participating node providing the intermediate data can be carried, so that the source of abnormal data can be determined, and further processing strategies for the participating node providing abnormal data can be determined, thereby improving the realizability and flexibility of data verification and improving the performance of machine learning.

[0018] In an embodiment, the method further comprises: eliminating the intermediate data that is abnormal, and updating a model of the machine learning. That is, after the participating node receives the intermediate data, verifies the intermediate data, and eliminates the intermediate data that is abnormal, the model can be updated according to other intermediate data, thereby avoiding the model deviation caused by the reporting of abnormal data, and improving the performance of machine learning.

[0019] In an embodiment, the method further comprises: sending a second message to the first device or the second device, the second message being used to indicate that the first device data feedback is abnormal, wherein the second device is a network repository function or a unified data management function.

[0020] In the above embodiment, after the communication device determines that the intermediate data is abnormal, the network repository function or the unified data management function can be notified, so that the network repository function or the unified data management function and the like can further take corresponding processing strategies on the participating node providing the abnormal data.

[0021] In an embodiment, the second message is used to indicate at least one of the following information: deregistering the first device; and stopping collecting data of external nodes of the communication network.

[0022] In the above embodiment, after the communication device determines that the intermediate data is abnormal, the network repository function or the unified data management function can be notified, so that the network repository function or the unified data management function and the like can further take corresponding processing strategies on the participating node providing the abnormal data.

[0023] In a second aspect, a communication method is provided, which can be performed by a first device. The first device can be a network device, or can be performed by a module (such as a chip, a chip system or a circuit) in the network device. Alternatively, the first device can be a terminal, or can be performed by a module (such as a chip, a chip system or a circuit) in the terminal. The method comprises: sending a first message, the first message comprising intermediate data of machine learning, the first message being used to request verification of the intermediate data; wherein the intermediate data is from a first device, and the first device is a participating node of the machine learning task; receiving a second message, the second message being used to indicate data feedback abnormality; and the second message further comprising a first strategy, the first strategy being used to indicate processing on the intermediate data with abnormality or the first device.

[0024] In an embodiment, the second message comprises information of the intermediate data with abnormality and / or information of the first device corresponding to the intermediate data with abnormality.

[0025] In a third aspect, a communication method is provided, which can be performed by a communication apparatus. The communication apparatus can be a network device, or a module (e.g., a chip, a chip system, or a circuit) in the network device, or a module or software that can implement all or part of the functions of the network device. For example, the communication apparatus can be an initiating node of machine learning. The method includes: receiving a first message from a first apparatus, the first message including intermediate data of machine learning and a first token; verifying the first token, and if the verification is passed, determining an update of a model according to the intermediate data.

[0026] In the above embodiments, the participating nodes of machine learning are verified, so that the security of data between nodes in the machine learning process can be improved, the model deviation caused by abnormal data reporting can be avoided, the interference of malicious or misused machine learning nodes on the entire machine learning process can be prevented, the security of the model is improved, and the performance of machine learning is improved.

[0027] In an embodiment, the method further includes: sending information of a coordinating node of machine learning to the first apparatus, for instructing the first apparatus to request authorization of machine learning from the coordinating node. In the above embodiment, by indicating the information of the coordinating node to the participating node, the participating node can request authorization of machine learning from the coordinating node, and only the participating node that passes the authorization can send the intermediate data to the initiating node, so that the security of data can be improved, the interference of malicious or misused machine learning nodes on the entire machine learning process can be prevented, the security of the model is improved, and the performance of machine learning is further improved.

[0028] In an embodiment, the verifying the first token includes: interacting with the coordinating node of machine learning to verify at least one of the following information: verifying whether the signature information of the valid coordinating node is included in the first token; verifying whether the certificate included in the first token is valid; or, verifying whether the first token is within a valid period. The coordinating node can provide an authorized token for the participating node, therefore, when the initiating node verifies the first token of the participating node, the initiating node can interact with the coordinating node to obtain relevant information for verifying the token, so that the reliability and accuracy of the security verification are improved, and the performance of machine learning is improved.

[0029] In a fourth aspect, a communication method is provided, which can be performed by a communication apparatus. The communication apparatus can be a network device, or a module (such as a chip, a chip system, or a circuit) in the network device, or a module or software that can implement all or part of the functions of the network device. For example, the communication apparatus can be a coordination node of machine learning. The method includes: receiving a second message from a first apparatus, the second message being used to request authorization for the first apparatus to send intermediate data of machine learning to an initiating node of machine learning; wherein the first apparatus is a participating node of the machine learning task; verifying the first apparatus, and if the verification is passed, generating a first token corresponding to the first apparatus; and sending the first token to the first apparatus.

[0030] In the above embodiments, the participating node can send an authorization request to the coordination node, the authorization request being used to request authorization for the first apparatus to send intermediate data of machine learning to the initiating node of machine learning. Thus, the coordination node generates an authorization token for the first apparatus only if the verification of the first apparatus is passed, so as to realize the identity verification of the participating node, avoid the model deviation caused by the abnormal data reporting, prevent the interference of the malicious or misused machine learning node to the whole machine learning process, improve the security of the model, and improve the performance of the machine learning.

[0031] In an embodiment, the verification of the first apparatus includes: determining a node type of the first apparatus; and verifying the first apparatus according to security information corresponding to the node type. Different types of participating nodes correspond to different identity verification manners. Thus, the coordination node can first determine the node type of the participating node, and then verify the identity of the participating node according to the security information corresponding to the node type, so as to improve the reliability and flexibility of the node verification, and improve the performance of the machine learning.

[0032] In an embodiment, the verification of the first apparatus includes: if the first apparatus is an external application function, determining security start information corresponding to the external application function, or obtaining related security information from a third-party verification and authorization server; verifying the first apparatus according to the security start information or the security information; if the first apparatus is a network function in a communication system, querying and obtaining registration information of the network function, or obtaining a checking result of configuration information of the network function from an operation and maintenance system; and determining whether the first apparatus is verified according to the registration information or the checking result.

[0033] In the above embodiments, different verification manners are flexibly configured for the external application function or the network function in the communication system, so as to improve the reliability and flexibility of the node verification, and improve the performance of the machine learning.

[0034] In an embodiment, the first token comprises a validity period, wherein the validity period of the first token corresponding to different node types is different. In the above embodiment, different token validity periods can be configured for different node types, such as different validity periods for tokens of external application functions or network functions in a communication system, thereby improving the reliability and flexibility of node verification and improving the performance of machine learning.

[0035] In an embodiment, if the verification fails, the method further comprises: sending a second message to the first device or the second device, the second message being used to indicate that the first device data feedback is abnormal; wherein the second device is a network repository function or a unified data management function.

[0036] In the above embodiment, after the communication device determines that the participating node with the abnormality exists, the network repository function or the unified data management function can be notified, so that the network repository function or the unified data management function can further take corresponding processing strategies for the participating node providing abnormal data.

[0037] In an embodiment, the second message is used to indicate at least one of the following information: deregistering the first device; and stopping collecting data of external nodes of the communication network.

[0038] In the above embodiment, after the communication device determines that the intermediate data with the abnormality exists, the network repository function or the unified data management function can be notified, and the participating node providing the abnormal data can be executed to be deregistered or to stop collecting data of external nodes of the communication network, and the like, so that subsequent abnormal data reporting can be avoided to cause model deviation, and the performance of machine learning can be improved.

[0039] In a fifth aspect, a communication method is provided, which can be executed by a communication device. The communication device can be a network device, or a module (such as a chip, a chip system or a circuit) in the network device, or a module or software capable of realizing all or part of the functions of the network device. For example, the communication device can be an initiating node or a coordinating node of machine learning. The method comprises: receiving intermediate data of machine learning from a first device, the first device being a participating node of the machine learning task; performing model updating on a test set according to the intermediate data to obtain a first model; and if an accuracy deviation between the first model and a model corresponding to the test set is greater than or equal to a first threshold, sending a first message, the first message being used to indicate that the intermediate data of the first device has an abnormality.

[0040] In the above embodiment, the reliability or accuracy of the intermediate data of the participating node is verified through the model updating test of the intermediate data of the participating node according to the test set, and the intermediate data with the exception is fed back, so as to avoid the model deviation caused by the reporting of the abnormal data, prevent the interference of the malicious or misused intermediate data on the whole machine learning process, improve the security of the model, and improve the performance of the machine learning.

[0041] In an embodiment, the method further includes: if the model accuracy deviation between the first model and the test model is less than a second threshold, updating a target model according to the intermediate data. That is, the reliability of the intermediate data of the participating node is verified through the model updating test of the intermediate data of the participating node according to the test set, and the intermediate data with high reliability is used to update the model, so as to improve the performance of the machine learning.

[0042] In an embodiment, the method further includes: receiving a first policy for performing corresponding processing on the intermediate data with the exception or the first device.

[0043] In the above embodiment, through the indication of the first policy, corresponding processing can be further taken on the participating node providing the intermediate data with the exception, so as to avoid the continuous reporting of the abnormal data, avoid the model deviation caused by the reporting of the abnormal data, and improve the performance of the machine learning.

[0044] In an embodiment, the first policy is used to indicate at least one of the following information: discarding the intermediate data from the first device; discarding the intermediate data from the first device for N times, where N is a positive integer; marking the first device as a first type or a second type; or updating the intermediate data.

[0045] In the above embodiment, through the discarding processing or updating processing of the intermediate data with the exception, or the marking of the participating node providing the abnormal data, such as the marking as a black list node or a gray list node, the model deviation caused by the reporting of the abnormal data can be avoided, the flexibility of the processing of the abnormal data is improved, and the effect of the machine learning is improved.

[0046] In an embodiment, the method further includes: receiving a second message, the second message including the test set and the model corresponding to the test set. That is, the initiating node can receive the test set required for testing the intermediate data and the model corresponding to the test set, so as to perform the model updating test according to the test set after receiving the intermediate data, verify the reliability of the intermediate data, and use the intermediate data with high reliability to update the model, so as to improve the performance of the machine learning.

[0047] In an implementation, the second message is carried in a federated learning preparation message. That is, the coordination node can send the test-related test set and the corresponding model to the initiating node through the federated learning preparation message, thereby improving test flexibility and improving the performance of machine learning.

[0048] In a sixth aspect, a communication method is provided, which can be performed by a communication device. The communication device can be a network device, or a module (such as a chip, chip system, or circuit) in the network device, or a module or software that can implement all or part of the functions of the network device. For example, the communication device can be an initiating node or a coordination node of machine learning. The method includes: sending a second message, the second message including a test set and a model corresponding to the test set; receiving a first message, the first message including information of a first device, the first message indicating that intermediate data of machine learning of the first device is abnormal; and the first device being a participating node of the machine learning task.

[0049] In an implementation, the method further includes: determining a first strategy according to the first message, the first strategy being used for performing corresponding processing on the abnormal intermediate data or the first device; and sending the first strategy.

[0050] In an implementation, the first strategy is used to indicate at least one of the following information: discarding intermediate data from the first device; discarding intermediate data from the first device for N times, where N is a positive integer; marking the first device as a first type or a second type; or updating the intermediate data.

[0051] In an implementation, the method further includes: receiving a third message including requirement information of the machine learning task; and determining the test set corresponding to the machine learning task and the model corresponding to the test set according to the requirement information.

[0052] In an implementation, the second message is carried in a federated learning preparation message.

[0053] In a seventh aspect, a communication device is provided, which is used to implement the above method. The communication device can be a device that performs any one of the above methods, or a node or equipment including the above device, or a module in the above device, such as a chip, chip system, or circuit, or a logic node, logic module, or software that can implement part or all of the functions.

[0054] The apparatus includes modules, units, or means corresponding to the above-described methods, which can be implemented by hardware, software, or by executing corresponding software by hardware. The hardware or software includes one or more modules or units corresponding to the above-described functions.

[0055] In a possible implementation, the apparatus can include a processing module and a transceiver module. The processing module can be configured to implement the processing functions in any of the above aspects and any possible implementation thereof. The processing module can be, for example, a processor. The transceiver module, which can also be referred to as a transceiver unit, is configured to implement the functions of transmitting and / or receiving in any of the above aspects and any possible implementation thereof. The transceiver module can include a transceiver circuit, a transceiver, a transceiver, or a communication interface.

[0056] In a possible implementation, the transceiver module includes a transmitting module and a receiving module, which are configured to implement the functions of transmitting and receiving in any of the above aspects and any possible implementation thereof.

[0057] In an eighth aspect, a communication apparatus is provided, which includes a processor. The processor is configured to couple with a memory, and read instructions in the memory, and perform the method in any of the above aspects according to the instructions. The communication apparatus can be the apparatus in any of the first aspect to the sixth aspect, or a node or device containing the apparatus, or a module in the apparatus, such as a chip, a chip system, or a circuit, or a logic node, a logic module, or software capable of implementing part or all of the functions.

[0058] In a possible implementation, the communication apparatus further includes a memory. The memory is configured to store program instructions and / or data. Optionally, the memory and the processor are integrated together.

[0059] In a possible implementation, the communication apparatus is a chip or a chip system. Optionally, when the communication apparatus is a chip system, the communication apparatus can be composed of a chip, or can include a chip and other discrete devices.

[0060] In a ninth aspect, a communication apparatus is provided, which includes a processor and an interface circuit. The interface circuit is configured to receive a computer program or instructions and transmit the computer program or instructions to the processor. The processor is configured to execute the computer program or instructions, so that the communication apparatus performs the method in any of the above aspects. The communication apparatus can be the apparatus in any of the first aspect to the sixth aspect, or a node or device containing the apparatus, or a module in the apparatus, such as a chip, a chip system, or a circuit, or a logic node, a logic module, or software capable of implementing part or all of the functions.

[0061] In a possible implementation, the communication apparatus is a chip or a chip system. Optionally, when the communication apparatus is a chip system, the chip system can be composed of a chip or can include a chip and other discrete devices.

[0062] In a tenth aspect, a computer readable storage medium is provided, which stores instructions that, when executed on a computer, cause the computer to perform the method of any one of the preceding aspects.

[0063] In an eleventh aspect, a computer program product is provided, which includes instructions that, when executed on a computer, cause the computer to perform the method of any one of the preceding aspects.

[0064] In a twelfth aspect, a communication system is provided, which includes the communication apparatus performing the method of the first aspect and the first apparatus performing the method of the second aspect.

[0065] In a possible implementation, the communication system includes the coordination node performing the method of the first aspect, the initiation node, and the first apparatus performing the method of the second aspect.

[0066] Alternatively, the communication system includes the communication apparatus performing the method of the third aspect and the communication apparatus performing the method of the fourth aspect; or the communication system includes the communication apparatus performing the method of the fifth aspect and the communication apparatus performing the method of the sixth aspect.

[0067] The technical effects brought by any one of the possible implementations of the second aspect to the twelfth aspect can refer to the technical effects brought by the different possible implementations of the first aspect, which will not be repeated here.

[0068] It can be understood that the solutions in each of the aspects can be combined as long as the solutions are not contradictory. BRIEF DESCRIPTION OF DRAWINGS

[0069] FIG. 1 is a schematic diagram of an architecture of a communication system according to an embodiment of the present application;

[0070] FIG. 2 is a schematic diagram of an interaction flow of a data analysis service according to an embodiment of the present application;

[0071] FIG. 3 is a schematic diagram of an architecture of a communication apparatus according to an embodiment of the present application;

[0072] FIG. 4A is a flowchart of a communication method according to an embodiment of the present application;

[0073] FIG. 4B is a schematic diagram of a machine learning architecture according to an embodiment of the present application;

[0074] FIG. 5 is a flow diagram of another communication method according to an embodiment of the present application;

[0075] FIG. 6 is a flow diagram of another communication method according to an embodiment of the present application;

[0076] FIG. 7 is a flow diagram of another communication method according to an embodiment of the present application;

[0077] FIG. 8 is a flow diagram of another communication method according to an embodiment of the present application;

[0078] FIG. 9 is a flow diagram of another communication method according to an embodiment of the present application;

[0079] FIG. 10 is a schematic diagram of a communication apparatus according to an embodiment of the present application. DETAILED DESCRIPTION

[0080] Hereinafter, the terms "first" and "second" are used only for the purpose of description, and should not be understood as indicating or implying relative importance or implicitly indicating the number of the technical features indicated. Therefore, the features defined with "first" and "second" can explicitly or implicitly include one or more of the features. In the description of the embodiments, unless otherwise specified, the meaning of "a plurality of" is two or more.

[0081] It should be noted that in the present application, the words "exemplary" or "for example" are used to mean serving as an example, instance, or illustration. Any embodiment or design scheme described as "exemplary" or "for example" in the present application should not be interpreted as being more preferred or advantageous than other embodiments or design schemes. Rather, the use of "exemplary" or "for example" is intended to present relevant concepts in a specific manner.

[0082] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0083] First, the implementation environment and application scenario of the embodiments of the present application are briefly introduced.

[0084] The communication method provided by the embodiments of the present application can be applied to a fourth generation (4th generation, 4G) communication system, for example, a long term evolution (long term evolution, LTE) communication system, and can also be applied to a fifth generation (5th generation, 5G) communication system, for example, a 5G new radio (new radio, NR) communication system, or various communication systems evolved after 5G, a future communication system. Figure 1 shows a network architecture to which the present application is applicable. Among them, Figure 1 takes the network service architecture of the fifth generation (5th generation, 5G) mobile communication system as an example to show the interaction relationship between network functions (network function, NF) and entities and the corresponding interfaces.Among them, the third generation partnership project (3GPP) service-based architecture (SBA) of the 5G system contains network functions and entities mainly including: user equipment (UE), at least one access network (AN) or radio access network (RAN) node, user plane function (UPF), data network (DN), access and mobility management function (AMF), session management function (SMF), policy control function (PCF), application function (AF), unified data management (UDM), network exposure function (NEF), unified data repository (UDR), authentication server function (AUSF), network storage function (NF repository function, NRF), network slice selection function (NSSF), network slice authentication and authorization network element (NSSAAF, The Network Slice Specific Authentication and Authorization Function) and network data analysis function (Network Data Analytics Function, NWDAF) and the like.

[0085] In addition, the present application is also applicable to a communication scenario of a non-terrestrial network (NTN). The NTN communication is a deployment scenario of a non-terrestrial network including a satellite system or a high altitude platform station (HAPS), and the like, which realizes wireless communication services by means of wide-area coverage capability of a non-terrestrial network device such as a satellite. Exemplarily, the non-terrestrial network device can also be referred to as an aerial network device, for example, a network device that can be deployed in the air, including a satellite, a HAPS device, a drone device, and the like, without limitation.

[0086] Among them, the UE, the (radio) access network ((R)AN) node, the UPF and the DN are generally referred to as user plane network functions and entities (or user plane network elements), and the other parts are generally referred to as control plane network functions and entities (or control plane network elements). The control plane network element is defined by 3GPP to process functions in a network, and the control plane network element has a 3GPP-defined function behavior and a 3GPP-defined interface. The NF can be a network element running on a dedicated hardware, or a software instance running on a dedicated hardware, or a virtual function instantiated on a suitable platform, such as a cloud infrastructure.

[0087] The main functions of each network function are described below.

[0088] Among them, the user plane network functions in the communication system include:

[0089] The (R)AN node: The (R)AN can be an AN, or a RAN, or also be referred to as an access network device, a RAN entity or an access node, etc., which constitutes a part of the communication system to help the terminal device to access the communication network. For example, the (R)AN can be various forms of base stations, such as macro base stations, micro base stations, radio controllers, relay stations, access points, or vehicle-mounted devices, wearable devices, or future evolved network devices in a public land mobile network (PLMN), satellite access nodes or NTN base stations, etc. The (R)AN is mainly responsible for radio resource management, quality of service management, data compression and encryption, etc. on the air interface side. In the NTN, the (R)AN node can be deployed on a high-altitude platform or a satellite, and can be used as a layer 1 (L1) relay, or as a base station, or as a distributed unit (DU), or as an integrated access and backhaul (IAB) node. Or it can be a gateway station or a ground station.

[0090] In addition, the (R)AN node can also be an access node in an open RAN (O-RAN or ORAN), a cloud radio access network (CRAN), or a wireless fidelity (WiFi) system, and can also be an access node in a communication system that integrates two or more of the above systems.

[0091] In a possible scenario, the RAN node can be a base station, an evolved NodeB (eNodeB), an access point (AP), a transmission reception point (TRP), a base station in a future mobile communication system (next generation NodeB, gNB), or an access node in a WiFi system, etc. The RAN node can be a macro base station, a micro base station or an indoor station, a relay node or a donor node, or a wireless controller in a CRAN scenario. Optionally, the RAN node can also be a server, a wearable device, a vehicle or a vehicle-mounted device, etc. For example, the access network device in vehicle to everything (V2X) technology can be a road side unit (RSU). All or part of the functions of the RAN node in this application can also be implemented by software functions running on hardware, or by virtualized functions instantiated on a platform such as a cloud platform. The RAN node in this application can also be a logical node, a logical module or software that can implement all or part of the functions of the RAN node.

[0092] In another possible scenario, a terminal device is assisted by multiple RAN nodes to implement wireless access in cooperation, and different RAN nodes respectively implement part of functions of a base station. For example, a RAN node can be a central unit (CU), a distributed unit (DU), a CU-control plane (CP), a CU-user plane (UP), or a radio unit (RU), etc. The CU and the DU can be separately arranged, or can also be included in the same network element, for example, in a baseband unit (BBU). The RU can be included in a radio frequency device or a radio frequency unit, for example, included in a remote radio unit (RRU), an active antenna processing unit (AAU), or a remote radio head (RRH).

[0093] In different systems, the CU (or CU-CP and CU-UP), DU or RU can also have different names, but those skilled in the art can understand their meanings. For example, in an ORAN system, the CU can also be referred to as an O-CU (open CU), the DU can also be referred to as an O-DU, the CU-CP can also be referred to as an O-CU-CP, the CU-UP can also be referred to as an O-CU-UP, and the RU can also be referred to as an O-RU. For the convenience of description, the CU, CU-CP, CU-UP, DU and RU are taken as examples for description in this application. Any one of the CU (or CU-CP, CU-UP), DU and RU in this application can be implemented by a software module, a hardware module, or a combination of a software module and a hardware module.

[0094] UE: can also be referred to as a terminal, a terminal device, a mobile station (MS), a mobile terminal (MT), etc., or a device used to provide voice or data connectivity to a user, or an Internet of Things device. For example, a terminal device includes a handheld device having wireless connection functionality, a vehicle-mounted device, etc. At present, a terminal device can be: a mobile phone, a tablet computer, a notebook computer, a palm computer, a mobile Internet device (MID), a wearable device (such as a smart watch, a smart bracelet, a pedometer, smart glasses, etc.), a vehicle-mounted device (such as a car, a bicycle, an electric vehicle, an airplane, a ship, a train, a high-speed rail, etc.), a satellite terminal, a virtual reality (VR) device, an augmented reality (AR) device, a smart point of sale (POS) machine, a customer-premises equipment (CPE), a light UE, a reduced capability UE (REDCAP UE), a wireless terminal in industrial control, a smart home device (such as a refrigerator, a television, an air conditioner, an electricity meter, etc.), a smart robot, a mechanical arm, a workshop device, a wireless terminal in unmanned driving, a wireless terminal in telemedicine, a wireless terminal in a smart grid, a wireless terminal in transportation safety, a wireless terminal in a smart city, or a wireless terminal in a smart home, a flight device (such as a smart robot, a hot air balloon, a drone, an airplane), etc. A terminal device can also be a vehicle device, such as a whole vehicle device, a vehicle-mounted module, a vehicle-mounted chip, an on board unit (OBU), or a telematics box (T-BOX), etc. A terminal device can also be other devices with terminal functions, for example, a device that plays a terminal function in D2D communication.

[0095] Embodiments of the present application do not limit the device form of the terminal, and the device used to implement the function of the terminal device can be a terminal device, or can be a device capable of supporting the terminal device to implement the function, such as a chip system. The device can be installed in the terminal device or used in matching with the terminal device. In embodiments of the present application, a chip system can be composed of a chip, or can include a chip and other discrete devices. All or part of the functions of the terminal device in the present application can also be implemented by software functions running on hardware, or by virtualized functions instantiated on a platform (such as a cloud platform).

[0096] UPF: mainly responsible for forwarding and receiving user plane data. The UPF can receive downlink data from the DN and then transmit the downlink data to the UE through the (R)AN. The UPF can also receive uplink data from the UE through the (R)AN and then forward the uplink data to the DN.

[0097] DN: for example, the DN can be an operator service network, an Internet access, or a third-party service network. The DN can interact with the UE through a PDU session. Among them, the PDU session can be divided into multiple types, such as Internet Protocol version 4 (IPv4), IPv6, etc.

[0098] In addition, the control plane network functions in the communication system include:

[0099] AMF: mainly responsible for processing control plane messages, responsible for mobility management of terminal devices, including mobile state management, allocation of user temporary identity, authentication and authorization of users. For example, access control, mobility management, registration and deregistration, and network element selection, etc.

[0100] SMF: mainly used for session management, session establishment, (private) IP address allocation and management of the UE, responsible for the establishment, modification and release of the session, and quality of service (QoS) control, etc.

[0101] UDM: mainly used for authentication credit processing, responsible for managing subscription data, user identification processing, access authorization, registration / mobility management, subscription management and short message management, etc. For example, when the subscription data of the user is modified, the UDM is responsible for notifying the corresponding network element.

[0102] NEF: mainly used to provide corresponding security guarantees to ensure the security of external applications to the communication network, provide external application QoS customization capability exposure, mobility state event subscription, AF request distribution, etc.

[0103] NRF: mainly used to provide internal / external addressing functions, etc.

[0104] AUSF: mainly used for authentication processing functions, such as bidirectional authentication and authorization between the terminal and the network.

[0105] AF: mainly used to send application impact data routing information to the network side, interact with the policy framework through the network exposure function network element for policy control, etc.

[0106] NSSAAF: mainly responsible for the authentication and authorization of network slices, and can interact with an authentication, authorization, and accounting server (AAA-S) through an authentication, authorization, and accounting proxy (AAA-P).

[0107] NWDAF: mainly used for analysis of various types of network data. Exemplarily, the network data can specifically include network operation data collected from NFs, terminal and network related statistical data obtained from OAM, application data obtained from a third party AF, and information of a terminal and / or information of a user obtained from the terminal, etc. Subsequently, the NWDAF can feed back the generated analysis result to nodes such as NFs, OAM, or the third party AF, and the NFs, OAM, or AF can use the analysis result of the NWDAF to perform different optimization operations.

[0108] Exemplarily, the specific working steps of the NWDAF can include several processes such as requesting data analysis, subscribing to analysis, collecting data, and feeding back analysis results. The working process of the NWDAF will be briefly introduced below in combination with the example of the interaction diagram in FIG. 2.

[0109] As shown in FIG. 2, the process in which a consumer of the NWDAF service requests data analysis from the NWDAF can include the following steps 1-2.

[0110] Step 1: The consumer sends a data analysis request (such as a Nnwdaf_AnalyticsInfo_Request) message to the NWDAF, to request a data analysis service from the NWDAF.

[0111] Alternatively, in an embodiment, the data analysis service can also be implemented through a subscription service. For example, the data analysis request message in step 1 can be a subscription request message, such as a Nnwdaf_AnalyticsSubscription_Subscribe message.

[0112] The parameters carried in the data analysis request message can include:

[0113] One or more identification information (Analytics ID) of data analysis: used to define the type of requested analysis, such as network slice analysis, NF load analysis, etc.

[0114] Analytics Filter Information: indicates the analytics information that needs to be reported, such as Single Network Slice Selection Assistance Information (S-NSSAI), NF identifier, etc., which is used to narrow down the range of analytics data.

[0115] Target of Analytics Reporting: indicates the target of data analytics.

[0116] Notification Target Address: for a subscription service, can include a notification address of a subscription related event, such as that NF-1 subscribes to a data analytics service of the NWDAF, indicating that the analysis result corresponding to the data analytics can be notified to NF-2.

[0117] Analytics Reporting Information: indicates the analysis result corresponding to the requested data analytics.

[0118] Step 2: the NWDAF sends a data analytics response (such as a Nnwdaf_AnalyticsInfo_Request Reponse) message to the consumer, which is used to feed back the analysis result to the consumer.

[0119] For example, for a subscription service, the data analytics response message can be a Nnwdaf_AnalyticsSubscription_Notify message.

[0120] Subsequently, the NWDAF can collect data from other nodes in the communication network according to the parameters in the data analytics request, such as collecting data from NFs, OAMs, AFs, and terminals, etc.

[0121] For example, as shown in FIG. 2, the process of the NWDAF collecting data can include steps 3-4. In the following examples, the nodes requested by the NWDAF to collect data are collectively referred to as data nodes, and the type of data nodes is not limited in the present application. Any node in the communication system that can provide data required by the NWDAF for data analytics can become a data node.

[0122] Step 3: the NWDAF sends a data request message to the data node.

[0123] The data request message sent by the data node to the NF, e.g., the Nnf_EventExposure_Subscribe message, is used to subscribe data from the NF. Optionally, the data request message can include subscription parameters required by the NWDAF to subscribe data.

[0124] Optionally, for the subscription service, step 3-1 can also be included:

[0125] Step 3-1: The data node notifies the NWDAF of the success or failure of data subscription.

[0126] Optionally, for the subscription service, step 3-2 can also be included:

[0127] Step 3-2: The data node prepares data.

[0128] Step 4: The data node sends the collected data to the NWDAF.

[0129] Optionally, the NF can return the subscribed data by sending the Nnf_EventExposure_Notify message.

[0130] Optionally, the message can also be used to notify the NWDAF of the success of subscription.

[0131] Further optionally, the NF can report data to the NWDAF at a certain time interval or after collecting a certain amount of data according to the subscription parameters in the data request message.

[0132] Optionally, the NWDAF can collect data through the File Transfer Protocol (FTP).

[0133] In an application scenario, through the above interaction process, the 5G mobility management related function can request the NWDAF to predict the movement trajectory of the terminal. For example, the NWDAF can request the operation and maintenance system to collect data to obtain the historical location information of the terminal. The NWDAF can generate a mobility prediction model of the terminal by analyzing the collected data. Subsequently, the NWDAF can provide the 5G mobility management related function with mobility prediction information (such as the mobility prediction model) of the terminal according to the current location of the terminal, so that the 5G mobility management related function can formulate more accurate network strategies and complete optimized mobility management operations. For example, based on the statistical law of the terminal location, the registration area is allocated, the handover decision is assisted based on the terminal location prediction information, and the mobility anchor point is selected based on the terminal mobility trajectory.

[0134] However, it is difficult for one NWDAF to collect data distributed in different areas, and multiple NWDAFs are usually deployed in a communication network, so data collection and data analysis can be implemented through distributed processing, for example, NWDAFs distributed in an area are used to collect data in the area, and the NWDAFs can share their models or data analysis with other NWDAFs.

[0135] It should be understood that the embodiments of the present application do not limit the network nodes performing data collection and / or data analysis, and the NWDAF is taken as an example in the above examples.

[0136] In an implementation, multiple network nodes (such as NWDAFs) can implement distributed machine learning, such as federated learning (FL), multiple nodes implement common modeling, and improve the use effect of AI models. For example, federated learning is performed between multiple network nodes, and the initiating node of federated learning can request multiple participating nodes to collect data, and the multiple participating nodes can collect data and successively send the data to the initiating node of federated learning. Therefore, how to improve the efficiency of federated learning between network nodes, improve the performance of federated learning and the security of data transmission is a problem to be solved at present.

[0137] It should be noted that the functions of other network elements included in FIG. 1 can refer to the related description in the conventional technology, which will not be repeated here. The network architecture shown in FIG. 1 is only used for example, and is not used to limit the technical solutions of the present application. Those skilled in the art should understand that in the specific implementation process, other network elements or devices can also be included, and the number of access network devices, terminal devices and / or core network devices can also be determined according to specific needs.

[0138] Optionally, each network element shown in FIG. 1 can be a device, or a functional module in the device, or a logical functional unit. It can be understood that the above functions can be network elements in a hardware device, such as a communication chip in a mobile phone, or a software function running on a special hardware, or a virtualized function instantiated on a platform (such as a cloud platform).

[0139] It can be understood that the devices or network elements in the communication system of the above FIG. 1 or FIG. 2 can communicate directly or through forwarding of other devices, and the embodiments of the present application do not make specific limitation thereto.

[0140] It can be understood that the above-mentioned Fig. 1 or Fig. 2 is only a schematic diagram and does not constitute a limitation on the applicable scenarios of the technical solutions provided in the present application. It should be understood by those skilled in the art that, in the specific implementation process, the communication system can include fewer devices or network elements than those shown in Fig. 1 or Fig. 2, or the communication system can also include other devices or other network elements, and the number of devices or network elements in the communication system can also be determined according to specific needs.

[0141] It should be noted that the communication system shown in Fig. 1 or Fig. 2 is only for example and is not intended to limit the technical solutions of the present application. It should be understood by those skilled in the art that, in the specific implementation process, the communication system can also include other devices or network elements, and the number of each network element can also be determined according to specific needs.

[0142] Optionally, each network element in Fig. 1 or Fig. 2 of the embodiments of the present application can be a functional module in one device. It can be understood that the above-mentioned function can be a network element in a hardware device, such as a communication chip in a mobile phone, or a software function running on a dedicated hardware, or a virtualized function instantiated on a platform (such as a cloud platform).

[0143] For example, each network element in Fig. 1 or Fig. 2 can be implemented by the communication device 300 in Fig. 3. Fig. 3 shows a hardware structure schematic diagram of a communication device applicable to the embodiments of the present application. The communication device 300 includes at least one processor 301, a communication line 302, a memory 303 and at least one communication interface 304.

[0144] The processor 301 can be one or more central processing units (CPU). In the case of a CPU, the CPU can be a single-core CPU or a multi-core CPU. The processor can be a general-purpose processor, a digital signal processor, an application-specific integrated circuit, a field programmable gate array, or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component, and can implement or execute the methods, steps and logic block diagrams disclosed in the embodiments of the present application. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in combination with the embodiments of the present application can be directly embodied by a hardware processor for execution or by a combination of hardware and software modules in the processor for execution.

[0145] Optionally, the processor can include one or a combination of a central processing unit (CPU), an application-specific integrated circuit (ASIC), a digital signal processor (DSP), a microprocessor unit (MPU), a microcontroller unit (MCU), a graphics processing unit (GPU), a field programmable gate array (FPGA), an artificial intelligence processor (AI processor), or a neural network processor (NPU).

[0146] The communication line 302 can include a path for transmitting information between the above components, such as a bus.

[0147] The communication interface 304 uses any transceiver-like device for communicating with other devices or communication networks, such as an Ethernet interface, a RAN interface, a wireless local area networks (WLAN) interface, etc.

[0148] The memory 303 can include, but is not limited to, a cache, a read-only memory (ROM), a random access memory (RAM), a synchronous dynamic random access memory (SDRAM), a hard disk drive (HDD), or a solid-state drive (SSD), an erasable programmable ROM (EPROM), or a compact disc read-only memory (CD-ROM), and the like. The memory is any medium capable of carrying or storing desired program codes in the form of instructions or data structures and capable of being accessed by a computer, but is not limited thereto. The memory in the embodiments of the present application can also be a circuit or any other device capable of realizing a storage function, used for storing computer programs or instructions, and / or data. Alternatively, the memory can exist independently and be connected to the processor through the communication line 302. Alternatively, the memory can also be integrated with the processor. The memory provided in the embodiments of the present application can generally have a non-volatility. The memory 303 is used to store computer execution instructions related to the schemes of the present application, and the execution is controlled by the processor 301. The processor 301 is used to execute the computer execution instructions stored in the memory 303, so as to realize the method provided in the embodiments of the present application.

[0149] Alternatively, the computer execution instructions in the embodiments of the present application can also be referred to as application program codes, which are not specifically limited in the embodiments of the present application.

[0150] In a specific implementation, as an embodiment, the processor 301 can include one or more CPUs, for example, CPU0 and CPU1 in FIG. 3.

[0151] In a specific implementation, as an embodiment, the communication device 300 can include multiple processors, for example, the processor 301 and the processor 307 in FIG. 3. Each of the processors can be a single-CPU processor or a multi-CPU processor. The processor herein can refer to one or more devices, circuits, and / or processing cores for processing data (for example, computer program instructions).

[0152] In a particular implementation, as an example, the communication apparatus 300 can further include an output device 305 and an input device 306. The output device 305 communicates with the processor 301 and can display information in various ways. For example, the output device 305 can be a liquid crystal display (LCD), a light emitting diode (LED) display device, a cathode ray tube (CRT) display device, or a projector, etc. The input device 306 communicates with the processor 301 and can receive user input in various ways. For example, the input device 306 can be a mouse, a keyboard, a touch screen device, a sensor device, etc.

[0153] The communication apparatus 300 described above can be a general-purpose device or a special-purpose device. In a particular implementation, the communication apparatus 300 can be a portable computer, a network server, a personal digital assistant (PDA), a mobile phone, a tablet computer, a wireless terminal apparatus, an embedded device, or a device having a similar structure as shown in FIG. 3. The embodiments of the present application do not limit the type of the communication apparatus 300.

[0154] The communication method provided by the embodiments of the present application will be described in detail below.

[0155] It should be noted that the names of messages between various network elements or the names of parameters in the messages in the following embodiments of the present application are only examples, and other names can also be used in a particular implementation, which is not limited in the embodiments of the present application.

[0156] It can be understood that some or all of the steps in the embodiments of the present application are only examples, and other steps or variations of various steps can also be performed in the embodiments of the present application. In addition, various steps can be performed in different orders according to the embodiments of the present application, and it is possible that not all steps in the embodiments of the present application are performed.

[0157] The communication method provided by the embodiments of the present application verifies the intermediate data reported by the participating nodes of machine learning, and then decides whether to update the model according to the intermediate data according to the verification result, or processes the abnormal intermediate data or the participating nodes providing the abnormal intermediate data accordingly, thereby improving the effect of machine learning.

[0158] As shown in FIG. 4A, the present application provides a communication method, which can include the following steps.

[0159] 401: The communication apparatus receives a first message including intermediate data of machine learning.

[0160] 402: The communication device verifies the intermediate data of machine learning, determines that there is abnormal intermediate data, or there is an abnormal participating node.

[0161] The communication device can verify the received intermediate data of machine learning. For example, if the intermediate data is gradient information, the communication device can verify whether the gradient information is reasonable by using a preconfigured data verification method, so as to determine whether the intermediate data is abnormal.

[0162] In an embodiment, the communication device verifies the intermediate data of machine learning, which can include at least one of the following: verifying the variance of the gradient in the intermediate data of machine learning, verifying the information divergence of the gradient, or verifying whether the gradient exceeds a deviation value.

[0163] In an embodiment, the communication device verifies the intermediate data of machine learning, which can include: determining the stage of machine learning, and determining whether the intermediate data is abnormal according to the stage of machine learning. For example, the communication device verifies the gradient information or weight information included in the intermediate data. A common method is to calculate the variance of the gradient. During the convergence period of machine learning processing, most of the gradients should be approximate, and the variance should not deviate too much. If the communication device calculates a large gradient variance, it means that the gradient information is abnormal. Further, the communication device can calculate the KL divergence of the gradient to see whether there is a gradient with a large deviation value, and finally associate the ID of the related participating node of machine learning.

[0164] In an embodiment, as shown in FIG. 4B, the machine learning architecture in a communication network can include: an initiator of machine learning, and one or more participating nodes of machine learning.

[0165] The initiator can be NWDAF, application function or other network element, which is mainly used for initiating machine learning tasks, performing model aggregation, or filtering participating nodes of machine learning tasks, etc. The participating node is a node selected by the initiator into the machine learning group, which is mainly used for collecting data and generating gradient information and / or weight information, etc. intermediate data of machine learning, and sending it to the initiator for model aggregation. The participating node can be NWDAF, external application function of the communication network, AMF, SMF or terminal, etc.

[0166] The intermediate data can include, for example, locally trained models, model parameters, weight data and / or gradient data, etc.

[0167] Exemplarily, as shown in FIG. 4B, the machine learning initiating node can be a central NWDAF, which can also be referred to as a server NWDAF. The machine learning participating nodes can be client NWDAFs in different regions, such as NWDAF-1 in region 1, NWDAF-2 in region 2, and NWDAF-n in region n, and the like.

[0168] The client NWDAF in each region can collect data from nodes associated with / managed by the region, such as network functions within the communication system, application functions outside the communication system, operation and maintenance systems, or terminals, and the like.

[0169] Exemplarily, the machine learning can be federated learning.

[0170] Federated learning: also known as federated machine learning, is a possible solution to handle data privacy and security, model training, and the like. According to the type of data, federated learning can be divided into horizontal federated learning, vertical federated learning (VFL), and federated transfer learning.

[0171] The horizontal federated learning is suitable for a scenario where data features have high overlap and sample overlap is small, such as data of banks in different regions, hospitals in different regions, and the like. Their businesses are similar (i.e., the features of the data are similar), but the users are different (i.e., the data samples are different).

[0172] The vertical federated learning, also known as vertical federated learning, is suitable for a scenario where sample overlap is large, but data feature overlap is small. For example, data of hospitals, banks, or insurance agencies in the same region, and the like. The users they serve are in a large region, so the samples are the same, but the businesses of pharmacies and banks are different, and the data collected has different data features.

[0173] In addition, in the vertical federated learning process, the local data sets in different vertical federated learning participating nodes for local model training have different feature spaces for the same sample (for example, the same user). Considering the security of data, the original data cannot be transmitted between nodes, and the local model or intermediate data can be shared, thereby improving the security of data and reducing the risk of data leakage. For example, in a multi-level NWDAF architecture, the NWDAF can be co-located with a network function (NF, such as a UPF, an SMF), and due to information security issues and performance reasons, the original data cannot be shared. In this case, federated learning can be used, multiple local NWDAFs collect data in the region and train to obtain a model, and then the local NWDAF can share the obtained model to the central NWDAF, and the central NWDAF completes model aggregation to obtain a global model, an optimal model, or model parameters, and sends them to the client NWDAF to implement data inference and other data analysis tasks.

[0174] In an embodiment, the embodiment shown in FIG. 4A can be applied to the initiating node of the machine learning task.

[0175] Optionally, the initiating node of the machine learning task itself has data processing capability and can also be a participating node of federated learning.

[0176] In addition, considering the security of data, the implementation process of the machine learning task (such as a federated learning task) can optionally introduce a coordinator or intermediate node, which can be used to coordinate data exchange between nodes and security authorization functions; it can be a NWDAF, NEF, Model Training Logical Function (MTLF), Analytics Logical Function (AnLF) MTLF, or other network elements, or a separate network element.

[0177] Optionally, the initiating node and the coordinator of the machine learning task can be deployed on the same network element, or independently deployed, which is not limited in the present application.

[0178] It should be understood that the communication device in the embodiment shown in FIG. 4A can be the initiating node of the machine learning task, or can be the coordinator of the machine learning task.

[0179] 403: The communication device sends a second message for indicating that the data feedback is abnormal.

[0180] In an embodiment, the second message can include information of the intermediate data with the abnormality and / or information of the first device corresponding to the intermediate data with the abnormality.

[0181] The first device is a participating node of a machine learning task.

[0182] That is, the communication device can inform other nodes of the existence of abnormal reported data through the second message. Optionally, information of the abnormal intermediate data and / or information of the participating node providing the abnormal intermediate data can be informed, so that subsequent processing of the abnormal data or the abnormal participating node can be performed, improving data security and thus improving the effect of machine learning.

[0183] In an embodiment, the second message can include a first policy for performing corresponding processing on the abnormal intermediate data or the first device.

[0184] For example, the first policy can be used to indicate at least one of the following information:

[0185] 1. The first policy can be used to indicate discarding the intermediate data from the first device. That is, the initiating node of subsequent machine learning can no longer receive the intermediate data from the first device. For example, the first device can be marked as an abnormal node according to the indicated first policy, so that the intermediate data from the node can be discarded and no longer parsed. Further optionally, a notification message can also be sent to the first device to indicate that it is an abnormal node and can no longer send intermediate data, thereby saving data transmission overhead.

[0186] 2. The first policy can be used to indicate discarding N times of intermediate data from the first device, where N is a positive integer.

[0187] That is, the first device can be considered to be abnormal in the short term, and the initiating node of subsequent machine learning can ignore or discard the intermediate data from the first device in the short term. For example, the first device can be discarded for the 0th to Nth intermediate data from the node according to the indicated first policy.

[0188] 3. The first policy can be used to indicate marking the first device as a first type or a second type.

[0189] In an embodiment, if the intermediate data corresponding to the first device is abnormal, the first device is marked as the first type. For example, the first type can be a gray list node. That is, the communication device can mark the abnormal participating node as a gray list node or add the ID of the node to the gray list, so as to be an observation node for subsequent further processing.

[0190] In an embodiment, if the number of times that the intermediate data corresponding to the first device is abnormal is greater than or equal to m times, the first device is marked as a second type, where m is a positive integer. For example, the second type can be a blacklisted node, that is, the communication device can mark the participating node that reports the number of times of abnormal intermediate data reaching m times as a blacklisted node, or add the ID of the node to the blacklist, so as to further process it subsequently.

[0191] For example, the preset policy of the communication device can be: if the participating node in the gray list subsequently sends normal data, the mark of the gray list node is removed, or the node is deleted from the gray list; otherwise, if the participating node in the gray list subsequently sends abnormal data, or the number of times of abnormal data reaches a preset threshold, the node is marked as a blacklisted node, or the node is added to the blacklist, so that the subsequent intermediate data from the node can no longer be received.

[0192] 4. The first strategy can be used to indicate updating of the intermediate data. For example, updating of the intermediate data can specifically include: ignoring gradient information with a large deviation value, or ignoring weight information with a large error, etc. That is, the communication device can use the first strategy to indicate updating of the intermediate data, so as to avoid the bad influence of abnormal intermediate data on the model. For example, the initiating node of machine learning can ignore or exclude the abnormal intermediate data, such as ignoring part of the gradient and / or weight information, and perform model training or model updating through other normal intermediate data.

[0193] Further optionally, the communication device can send a notification message to the participating node, indicating that the participating node can adjust or update the reported intermediate data. For example, the participating node can be notified that it no longer reports part of the gradient information with a large deviation value, or weight information with a large error, etc.

[0194] In an embodiment, the communication device can also send a second message to the first device, indicating that the data feedback of the first device is abnormal. Further optionally, the second message can also be used to indicate that the first device can stop collecting data of the external application function of the communication network, or indicate that the first device can exclude a specific session management function or access and mobility management function data source, to adjust the data collection strategy and avoid abnormal data.

[0195] In another embodiment, the communication device can also send a second message to the second device, indicating that the data feedback of the first device is abnormal. For example, the second device can be a network storage function, a unified data management function, or other management functions, which are not limited in the present application.

[0196] Further optionally, the second message can be used to instruct to deregister the first device, or to deregister a data collection node managed by the first device. For example, the first device can request registration to the unified data management function when it accesses the communication network, and when there is an abnormality in the intermediate data sent by the first device to the machine learning, the second message can be used to request the unified data management function to deregister the first device, so as to avoid its participation in the machine learning task and provide abnormal data.

[0197] For another example, the source of the abnormal intermediate data sent by the first device to the machine learning is part of the session management function managed by the first device, and the second message can be used to request to deregister the abnormal session management function, so as to avoid its participation in the machine learning task and provide abnormal data.

[0198] In the above embodiments, in the scenario of distributed machine learning of multiple nodes in the communication system, the intermediate data reported by the participating nodes can be verified to screen out abnormal intermediate data, so that the abnormal intermediate data can be processed or instructed to the corresponding participating node, which can avoid the bad influence of abnormal data on the machine learning result, thereby improving the performance of machine learning.

[0199] In addition, the present application also provides a communication method for verifying the participating nodes of machine learning, thereby improving the security of the data exchanged between the nodes in the machine learning process and improving the performance of machine learning.

[0200] As shown in FIG. 5, the communication method can include the following steps.

[0201] 501: The first device sends a first message to the initiating node, including intermediate data of machine learning and a first token.

[0202] The first message is used to request the initiating node to verify the intermediate data of the first device, specifically to verify the first token.

[0203] Token is a way of identity authentication and authorization, which is usually used for verification between two communication nodes.

[0204] 502: The initiating node verifies the first token, and if the verification is passed, determines an updated model according to the intermediate data.

[0205] In an embodiment, the initiating node verifies the first token, which can specifically include verifying one or more of the following information: verifying whether the signature information of the coordinating node in the first token is valid; verifying whether the certificate included in the first token is valid; or verifying whether the first token is within a valid period.

[0206] Optionally, if the first token is verified by the initiating node, the intermediate data can be used to update the model to obtain an updated model; otherwise, if the first token is not verified or fails to be verified by the initiating node, the intermediate data can be discarded.

[0207] In an embodiment, in a scenario where the machine learning process includes a coordination node, the first token can be issued by the coordination node to the first device. That is, the participating node, i.e., the first device, can first request authorization for machine learning from the coordination node, and the coordination node can verify the identity of the first device to generate a corresponding first token for it. Then, the first device can carry the first token when sending the intermediate data to the initiating node, so that the initiating node can verify the security of the token to perform security check on the machine learning data transmission process.

[0208] For the scenario where the first token is issued by the coordination node, the initiating node can request the necessary information required for verifying the token from the coordination node during the security check of the first token, such as the validity period of the token, the signature information or certificate of the coordination node, etc., so as to perform security check on the first token of the first device according to the obtained information.

[0209] In an embodiment, considering that the participating node needs to request authorization for machine learning from the coordination node, the method can further include the following steps:

[0210] 500a: The initiating node sends information of the coordination node to the first device, for instructing the first device to request authorization for machine learning from the coordination node.

[0211] For example, the initiating node can carry the information of the coordination node, such as the IP address or ID of the coordination node, in the machine learning request message sent to the participating node, so that the first device can send an authorization request to the corresponding coordination node according to the IP address or ID, for requesting authorization for machine learning.

[0212] Optionally, the method can include the following steps.

[0213] 500b: The first device sends a second message to the coordination node, for requesting authorization for machine learning.

[0214] Specifically, the second message can be used to request the coordination node to authorize the first device to send the intermediate data of machine learning to the initiating node of machine learning. That is, the first device can send a second message to the coordination node, for requesting the coordination node to issue an authorization token for sending the intermediate data of machine learning.

[0215] 500c: The coordination node verifies the first device, and generates the first token if the verification is passed.

[0216] In an embodiment, when the coordinating node verifies the first device, the node type of the first device can be determined first, and then the first device is verified according to the security information corresponding to the node type. For example, the node type is an external application function or an internal network function of the communication network, and the corresponding verification manner can be different.

[0217] For example, when the coordinating node verifies the first device, the following manners can be used: if the first device is an external application function of the communication system, the security startup information corresponding to the application function is determined, or the related security information is obtained from a third-party verification and authorization server; the first device is verified according to the security startup information or the security information; or,

[0218] If the first device is a network function inside the communication system, the registration information of the network function is queried, or the checking result of the configuration information of the network function is obtained from an operation and maintenance system; and whether the first device passes the verification is determined according to the registration information or the checking result.

[0219] Optionally, the first token includes a validity period, and the validity period of the first token corresponding to different node types is different.

[0220] 500d: The coordinating node sends a first token to the first device.

[0221] In addition, for the foregoing embodiments, when the communication device finds that a certain participating node of machine learning is abnormal, or the reported data is abnormal, or the authorization token of the participating node fails, the corresponding adjustment measures can be triggered. For example, the communication device can send an abnormal notification message to the corresponding participating node to indicate to stop collecting data; or send a notification to the management node in the network to unregister the participating node, or unregister the data collection node managed by the participating node, and the like, to improve the performance of machine learning.

[0222] In an embodiment, the foregoing embodiments shown in FIG. 4A or FIG. 5 can further perform the following steps.

[0223] 1. The coordinating node of machine learning sends a second message to the first device, to indicate that the first device data feedback is abnormal.

[0224] For example, the second message can be an abnormal notification message, which can be used to indicate that the intermediate data reported by the first device is abnormal, or that the token verification of the first device fails. Therefore, the participating node can adjust the corresponding data collection strategy or data feedback strategy according to the abnormal notification message.

[0225] Optionally, for the scenario that the first device is an external application function of the communication system, the abnormal notification message can indicate the first device to stop reporting data to the initiating node. For example, for the scenario that the first device collects data of the external application function, the abnormal notification message can indicate the first device to stop collecting data of the external application function. For example, the first device can exclude specific access and mobility management function or session management function as data source through the abnormal notification message, so as to avoid collecting abnormal data or reporting abnormal intermediate data to the initiating node.

[0226] Optionally, in the above implementation process, the first device can send a notification message to the data source according to the abnormal notification message, for indicating to stop collecting data, such as sending a notification message to the external application function or specific access and mobility management function or session management function, for indicating that the data reporting is no longer needed.

[0227] In another embodiment, the above-mentioned embodiment shown in FIG. 4A or FIG. 5 can further perform the following steps.

[0228] 2. The coordinating node of machine learning sends a second message to the second device, for indicating that the first device has data feedback abnormality, wherein the second device can be a unified data management function or a unified data management function. That is, the coordinating node can send a notification message to a management node in the communication network, for indicating that the participating node has feedback data abnormality.

[0229] For example, the second message can be an abnormal notification message, which can be used to indicate that the intermediate data reported by the first device is abnormal, or to indicate that the token verification of the first device is failed.

[0230] Further, the second message can also be used to indicate to implement deregistration to the first device, such as performing DE-REGISTER to the first device.

[0231] For example, the coordinating node can be locally pre-configured with a response notification policy, and when the coordinating node determines that the first device meets the pre-configured notification policy, the coordinating node can send a notification message to a network storage function, for letting the network storage function to deregister the specified first device.

[0232] For example, the pre-configured notification strategy can be: the coordination node can periodically trigger the deregistration of all participating nodes in the blacklist; or if a participating node is marked as abnormal a number of times reaching or exceeding a preset threshold, the coordination node can trigger the deregistration of the participating node, or the coordination node can discard the intermediate data reported by the participating node; or the coordination node can notify the operation and maintenance system that the participating node is abnormal, so that the operation and maintenance system can further check the configuration information of the participating node. Alternatively, the operation and maintenance system can check the security start-up data of the participating node, detect whether the participating node has alarm information, suspicious log records, etc.

[0233] For example, the second message sent by the coordination node to the second device can be an abnormality notification message, which includes the identifier corresponding to the participating node.

[0234] Optionally, the abnormality notification message can further include abnormal event description and / or abnormal code information, etc.

[0235] For example, the abnormal event description can be carried by the abnormality notification message, or can be implicitly indicated in the abnormality notification message name. In one possible implementation, the abnormality notification message name itself can be the abnormal event description.

[0236] The abnormal code can be an error code, which uniquely corresponds to an error description, such as indicating that the participating node reports data abnormally, or that the participating node token verification is abnormal, etc.

[0237] Correspondingly, the second device can perform corresponding processing according to the received second message, such as deregistering the participating node, or notifying the operation and maintenance system, etc.

[0238] For example, the network storage function can obtain the abnormal event description and / or the abnormal code according to the abnormality notification message sent by the coordination node, so as to determine to perform corresponding management measures, for example: notifying the operation and maintenance system or initiating the deregistration process of the participating node, etc.

[0239] It should be understood that the above embodiments can be implemented alone or in combination with other implementation processes of the present application, such as in combination with the embodiments shown in FIG. 4A, FIG. 5, or FIG. 6 or FIG. 7 described below, etc., which are not limited by the present application.

[0240] In the following, several possible implementation modes will be introduced in combination with specific interaction processes. In this embodiment, the network elements involved include the consumption node of the data analysis function, the initiating node of the machine learning, the coordination node, at least one participating node, etc. As shown in FIG. 6, the communication method can include the following steps.

[0241] 601: The consumer node sends a request message to the machine learning initiating node, carrying the requirement of the machine learning task.

[0242] The request message is used for the consumer node to request the data processing task from the machine learning initiating node, and can trigger the machine learning task by carrying the requirement information of the machine learning task. Optionally, the initiating node can determine whether to trigger the machine learning task according to the requirement of the consumer, such as triggering federated learning.

[0243] For example, the request message can be a machine learning request, such as a federated learning request, a model request, an inference request, etc., which can be used to trigger machine learning.

[0244] 602: The initiating node of machine learning sends a request message to the coordination node to obtain the information of the participating node required by the machine learning task.

[0245] Correspondingly, the initiating node receives the request message, determines to trigger federated machine learning according to the requirement information carried, such as triggering vertical federated learning, and can send a request message to the coordination node to obtain the information of the participating node required by the machine learning task.

[0246] The request message can be the requirement of the machine learning task in the foregoing step 601. That is, the request message is used to request the coordinator to determine the participating node or candidate participating node of machine learning according to the requirement.

[0247] For example, the request message can be a federated learning preparation request message, which is used to request the coordination node to provide the information of the participating node of federated learning.

[0248] Optionally, the following steps can also be included.

[0249] 602-1: The coordination node of machine learning sends a first message to the second device to obtain the information of the candidate participating node required by the machine learning task.

[0250] For example, the second device can be a network storage function or a unified data management function, etc. For example, when the second device is a network storage function, it can be NRF, or when the second device is a unified data management function, it can be UDM.

[0251] The first message can include the requirement of the machine learning task, which is used for the second device to screen the candidate participating node capable of participating in the machine learning process according to the requirement for this machine learning task.

[0252] Exemplarily, the coordination node sends a first message to the network storage function, the first message being a network element discovery request (Nnrf_NFDiscovery_Request) message, for obtaining, from the network storage function, nodes of current registration of machine learning or nodes capable of participating in a machine learning task at present.

[0253] 602-2: The second device determines a candidate participating node.

[0254] Exemplarily, if the second device is the network storage function, the network storage function can select a registered node in the network as the candidate participating node, for example, as one of the candidate participating nodes in a candidate node list (such as a member list).

[0255] Further optionally, the second device can select the candidate participating node according to the requirement of the machine learning task of the initiating node. For example, if the current machine learning task requires a node supporting vertical federated learning, the second device can select a node supporting vertical federated learning as the candidate participating node of the current machine learning task according to the algorithm supported by the node.

[0256] 602-3: The second device sends information of the candidate participating node to the coordination node.

[0257] Exemplarily, the second device can send a network element discovery response (Nnrf_NFDiscovery_response) message to the coordination node, including information of at least one candidate participating node. For example, the message can carry a candidate participating node list such as {candidate member list}, which can carry identification information of at least one candidate node, such as an IP address of the node or an ID of a network function instance, etc.

[0258] Exemplarily, the node identity (ID) included in the response message can refer to an identifier for determining a network element in the message, which can be an identifier of an application function, an IP address, an ID of a network function instance, etc., which is not limited in the present application.

[0259] Optionally, in some possible implementation scenarios, in addition to the identification information of the node, the information of the candidate participating node can further include subscription information corresponding to the node, such as attribute information and capability information, for assisting in screening of the machine learning participating node, which is not described herein again.

[0260] Correspondingly, the coordination node can obtain the information of the candidate participating node from the second device.

[0261] 603: The coordination node of machine learning determines a participating node (or a candidate participating node) and sends it to the initiating node.

[0262] The coordination node determines the candidate participating nodes available at present according to the requirement of the current machine learning task, and sends a response message to the initiating node to indicate the information of the determined candidate participating nodes.

[0263] Optionally, the coordination node can determine whether there is a participating node marked as black list in the candidate participating nodes, and exclude the participating node if there is.

[0264] Optionally, when the coordination node sends the information of the candidate participating nodes to the initiating node, the capability information of the candidate participating nodes can be carried, so that the initiating node of machine learning can confirm whether the participating nodes meet the requirement of the current machine learning according to the capability information. For example, the coordination node sends the information of the candidate participating nodes to the initiating node, including the identification ID of at least one candidate participating node. Optionally, the response message can also include the capability information corresponding to the identification of each candidate participating node.

[0265] In an embodiment, the capability information includes at least one of the following information: machine learning algorithm supported by the first device, data processing capability of the first device, or performance index of the first device. For example, the machine learning algorithm can include different types of federated learning, such as vertical federated learning, horizontal federated learning, or federated transfer learning. The data processing capability of the first device can include index information or parameters representing strong or weak data processing capability, such as the performance of central processing unit (CPU), such as the clock speed, core number and cache size of CPU. The performance index of the first device can include index information or parameters representing the transmission performance of the first device, such as bandwidth, packet loss rate, throughput, frame rate, etc.

[0266] For example, in step 603, the coordination node can send a machine learning preparation response message to the initiating node to provide the determined candidate participating nodes of machine learning to the initiating node.

[0267] Optionally, in step 603, the coordination node can send a machine learning information determination request (such as federated learning information confirm request) message to the initiating node, including the (candidate) participating nodes filtered by the coordination node according to the black list, capability information of the nodes and other factors. Optionally, the request message can also carry the capability information corresponding to the participating nodes, for the initiating node to further filter the participating nodes.

[0268] Correspondingly, the initiating node receives the information of the candidate participating nodes from the coordination node.

[0269] Optionally, the initiating node can further screen the candidate participating nodes to determine the final participating nodes. Further optionally, the initiating node can send a notification message to the coordination node, for indicating information of the final participating nodes determined by the initiating node.

[0270] Optionally, the method can comprise the following step 604.

[0271] 604: The initiating node sends a notification message to the coordination node.

[0272] The notification message can be used to indicate information of the determined participating nodes. Exemplarily, the notification message can be an acknowledge (ACK) response. Optionally, the ACK response can carry information of the participating nodes such as node ID, or only the ACK response is sent without carrying the node ID, which can be used to indicate that all the candidate participating nodes received in the determining step 603 are determined as the final participating nodes by the initiating node.

[0273] 605: The initiating node sends a machine learning request message to the participating nodes.

[0274] The step 605 is optional, and the machine learning request message is used to request the participating nodes to participate in the current round of federated learning.

[0275] Correspondingly, the participating nodes receive the machine learning request message, and start local machine learning according to the locally collected data to obtain intermediate data such as model, gradient information and / or weight information.

[0276] 606: The participating nodes send the intermediate data to the initiating node.

[0277] The participating nodes send the locally obtained intermediate data of machine learning to the initiating node.

[0278] 607: The initiating node sends a verification request to the coordination node.

[0279] Correspondingly, the initiating node receives the intermediate data and triggers verification of the intermediate data.

[0280] Optionally, to save signaling overhead, the initiating node can wait to receive multiple intermediate data according to a preconfigured strategy, and then trigger verification of the multiple intermediate data.

[0281] Exemplarily, the initiating node sends a gradient information verification request message to the coordination node, which can carry an identifier of the gradient information. Optionally, the gradient information verification request message can also carry an identifier of a data source of the gradient information, i.e., an identifier of a participating node corresponding to the gradient information.

[0282] 608: The coordination node verifies the intermediate data, and determines that there is abnormal intermediate data or abnormal participating node.

[0283] The verification process can refer to the description in the aforementioned step 402.

[0284] For example, the coordination node can determine the stage of the current federated learning, for example, in the convergence period, generally the gradient change of the federated learning is relatively small, if the intermediate data includes a gradient with a large change, it can be judged as abnormal data.

[0285] For another example, the verification can be performed by calculating the variance of the gradient, most of the gradients in the convergence period should be approximate, and the variance should not deviate too much, if the calculated variance is too large, it means that the gradient information is abnormal, and optionally, the KL divergence of the gradient can be further calculated to see if there is a gradient with a large deviation value, if there is, it can be determined as abnormal data, and the participating node corresponding to the abnormal data is determined.

[0286] Further optionally, the coordination node can perform corresponding marking processing on the abnormal data according to the preconfigured strategy.

[0287] For example, the participating node corresponding to the abnormal data, such as the first device, can be marked as the first type, for example, the first type can be a gray list node. If the first device has abnormal intermediate data more than or equal to m times, the first device can be marked as the second type, for example, the second type can be a black list node, wherein m is a positive integer.

[0288] 609: The coordination node sends a verification response to the initiating node.

[0289] Specifically, the coordination node can send a second message to the initiating node, indicating the information of the intermediate data with abnormalities and / or the information of the first device corresponding to the intermediate data with abnormalities.

[0290] Optionally, the verification response sent by the coordination node to the initiating node can also include the processing of the abnormal data or the abnormal participating node, such as the first strategy described above.

[0291] Correspondingly, the initiating node receives the verification response, and can perform corresponding processing on the intermediate data with abnormalities or the participating node with abnormalities indicated in the verification response according to the preconfigured strategy, for example, the initiating node can propose the intermediate data with abnormalities according to the preconfigured strategy. Alternatively, the initiating node can perform corresponding processing on the intermediate data with abnormalities or the participating node with abnormalities according to the first strategy indicated. For example, the initiating node can perform corresponding marking processing on the intermediate data with abnormalities according to the first strategy, such as marking as a black list node or a gray list node.

[0292] 610: The initiating node sends a notification message to the participating node.

[0293] The notification message can be used to indicate the processing strategy of the initiator node to the participating node, for example, the initiator node will discard the intermediate information from the participating node subsequently, or the initiator node corrects the gradient information of the participating node, such as eliminating gradient information with large deviation. Correspondingly, the participating node receiving the notification message can perform corresponding processing, for example, no longer sending intermediate data to the initiator node, or no longer sending abnormal gradient information to the initiator node, or updating or screening the gradient information that may be abnormal, and performing local adjustment.

[0294] 611: The initiator node performs model updating and sends the updated determined model to the participating node.

[0295] For example, the initiator node can eliminate the intermediate data with abnormality according to the verification response of step 609, then update the model according to the intermediate data, and send the updated determined model to the participating node for local training.

[0296] It should be noted that the execution order of the above steps 610 and 611 is not limited by the present application, and the model updating can be performed first, and then the notification message is sent.

[0297] 612: The initiator node performs model aggregation and sends a response message to the consumer node.

[0298] For example, after the initiator node completes the model aggregation of federated learning, the response message can be sent to the consumer node, which can carry the model obtained by aggregation processing.

[0299] Further, the above steps 606-612 can be repeated in the process of machine learning to obtain the target model.

[0300] In addition, the present application also provides a communication method for verifying the participating node of machine learning, thereby improving the security of data interaction between nodes in the process of machine learning and improving the performance of machine learning.

[0301] As shown in FIG. 7, the method can include the following steps.

[0302] 701: The initiator node sends a machine learning request message to the participating node.

[0303] The step 701 is optional, and the machine learning request message is used to request the first device to participate in federated learning.

[0304] Before this step, the steps 601-604 in the embodiment of the preceding FIG. 6 can be performed, which will not be described here.

[0305] The initiating node sends a machine learning request message to the participating node, indicating that the participating node can report intermediate data required for machine learning. The request message can carry an identifier of the current machine learning, or an identifier of a data analysis task, etc.

[0306] In addition, the request message can also carry information of the coordination node, such as an IP address or an ID of the coordination node, for the participating node to request authorization verification for machine learning from the coordination node.

[0307] 702: The participating node sends an authorization request to the coordination node.

[0308] Correspondingly, the participating node sends an authorization request to the coordination node according to the information of the coordination node obtained in the foregoing steps, to obtain authorization for machine learning. Optionally, the authorization request can carry an identifier of the current machine learning, or an identifier of a data analysis task, etc., to distinguish different machine learning processes.

[0309] 703: The coordination node verifies the participating node, and generates a first token if the verification is passed.

[0310] Specifically, the coordination node receives an authorization request of the participating node, such as a token authorization request, and verifies that the participating node is a secure node, and then generates a token for the participating node, for verification of the participating node reporting intermediate data to the initiating node.

[0311] Optionally, the coordination node can first determine the type of the participating node, and verify it according to the type.

[0312] For example, for an application function outside the communication system, the coordination node can determine corresponding security start information, or find a third-party verification and authorization server to obtain security information related to the application function. For a network function inside the communication system, the coordination node can query registration information of the network element, or notify an operation and maintenance system to check configuration information of the network element, and confirm whether the verification is passed according to the checking result.

[0313] 704: The coordination node sends an authorization response to the participating node, carrying the first token.

[0314] Optionally, the first token generated by the coordination node generally has a validity period, which can be different according to different network element types and different checking manners. For example, if an external authorization server helps to check, the validity period of the token can be set to 1 hour, if an operation and maintenance system checks a configuration file, the token can be set to be valid for a long time, and if the registration information is checked, the token can be set to be valid for 1 day.

[0315] 705: The participating node sends the intermediate data to the initiating node, carrying the first token.

[0316] 706: The initiating node verifies the first token.

[0317] The initiating node can verify the token validity of the participating node, and if the verification is passed, the intermediate data sent by the participating node is adopted and used to update the model (such as optimizing the model using gradient information). If the verification is not passed, the intermediate data is discarded.

[0318] For example, the initiating node verifies the token in one or more of the following ways: verifying whether the token contains valid coordinator signature information, verifying whether the certificate in the token is valid, or verifying whether the validity period of the token is expired, etc. The initiating node can interact with the coordinating node to obtain necessary information for token verification, such as the validity period, signature information, etc. For details, refer to the relevant token verification process, which is not limited in the present application.

[0319] For example, when the participating node finds that the token is expired, it can trigger a new token application process, i.e. repeat the steps 702-704. That is, the token application process does not necessarily be triggered immediately after step 701, but can be executed multiple times during the machine learning process.

[0320] For example, the steps 702-704 of triggering the token application can be triggered after step 701, or the participating node can check its own token expiration to trigger, or the initiating node can send a message to the participating node to trigger the token application after checking the invalid token in step 706. The present application is not limited in this regard.

[0321] 707: The initiating node performs model aggregation and sends a response message to the consuming node.

[0322] In addition, the present application also provides a communication method, which verifies the reliability or accuracy of the intermediate data of the participating node by testing the model update of the intermediate data of the participating node according to the test set, thereby improving the performance of machine learning.

[0323] As shown in FIG. 8, the method can include the following steps.

[0324] 801: The first device sends the intermediate data of machine learning to the initiating node.

[0325] The first device is a participating node of the machine learning task.

[0326] 802: The initiating node updates the model according to the intermediate data and the test set, obtains a first model, and compares the model accuracy of the first model and a test model.

[0327] If the accuracy deviation between the first model and the model corresponding to the test set is greater than or equal to a first threshold, a first message is sent to indicate that the intermediate data of the first device is abnormal.

[0328] For example, if the model accuracy deviation between the first model and the test model is less than a second threshold, the target model is updated according to the intermediate data. The first threshold and the second threshold can be the same or different, and the application does not limit the specific threshold value or configuration.

[0329] That is, the initiating node can first update the model according to the received intermediate data, test according to the pre-configured test set, and determine whether the updated model accuracy is improved. If the model accuracy is improved, the target model can be updated according to the intermediate data. If the accuracy is not improved or is worse, the intermediate data can be discarded.

[0330] In addition, the initiating node can notify the coordinating node that the intermediate data of the first device is abnormal.

[0331] Optionally, the method further includes the following steps.

[0332] 803: The initiating node sends a first message to the coordinating node to indicate that the intermediate data of the first device is abnormal.

[0333] Optionally, the first message can include information of the first device, such as an IP address or a network element ID; or the first message can include an identifier of the intermediate data.

[0334] Further optionally, the coordinating node can send a first policy to the initiating node, which, similar to the first policy in the foregoing embodiments, can be used to perform corresponding processing on the intermediate data or the first device that is abnormal. For example, the first message can carry the first policy, or the first policy can be sent separately.

[0335] In an embodiment, the first policy can be used to indicate at least one of the following information: discard the intermediate data from the first device; discard N times of intermediate data from the first device, where N is a positive integer; mark the first device as a first type or a second type; or update the intermediate data.

[0336] In an embodiment, the initiating node can obtain relevant data for testing before updating the intermediate data reported by the participating nodes. Optionally, the method further includes the following steps.

[0337] 800: The coordination node sends a second message to the initiating node, including a test set and a test model corresponding to the test set.

[0338] For example, the second message can include a test set and a test model corresponding to the test set, for model update testing on the intermediate data, so as to verify the reliability of the intermediate data and whether the performance of the machine learning can be improved.

[0339] Optionally, the second message can be carried in a federated learning preparation message, or can be sent separately.

[0340] In an embodiment, the coordination node obtains test-related data, and specifically can determine a test set and a corresponding test model according to the requirements of the machine learning task. Alternatively, the coordination node can obtain a test set from a consumption node, or pre-configure a test set, etc.

[0341] The above possible embodiments will be introduced below in combination with a specific interaction process, by testing and updating the model to verify the effect of the received intermediate data on improving the model accuracy, so as to avoid invalid model updating and improve the performance of the machine learning. In this embodiment, the consumption node, the initiating node of machine learning, the coordination node, at least one participating node, etc. are involved. As shown in FIG. 9, the communication method can include the following steps.

[0342] 901: The consumption node sends a request message to the initiating node of machine learning, carrying the requirements of the machine learning task.

[0343] Reference can be made to the foregoing step 601.

[0344] 902: The initiating node sends a machine learning preparation request to the coordination node, for obtaining the information of the participating nodes required by the machine learning task.

[0345] Reference can be made to the foregoing step 602. In addition, in this embodiment, the request message can also be used to indicate the test data required for the machine learning.

[0346] 903: The coordination node determines a test set and a test model.

[0347] Optionally, the test set can be obtained from the consumption node, or the coordination node can generate a test set according to the requirements of the machine learning task, or the test set can be pre-configured in the coordination node.

[0348] 904: The coordination node determines a candidate participating node.

[0349] The specific implementation process of the coordination node determining the candidate participating node can refer to the foregoing embodiment steps 602-1 to 602-3.

[0350] 905: The coordination node sends a machine learning preparation response to the initiating node.

[0351] For example, the response message can be a federated learning preparation response message, which carries information such as a test set and a test model. The test result can be included in the test set.

[0352] 906: The initiating node sends a notification message to the coordination node.

[0353] The notification message can be used to indicate the information of the determined participating node, which can refer to the aforementioned step 604.

[0354] 907: The initiating node sends a machine learning request message to the participating node.

[0355] Refer to the aforementioned step 605.

[0356] 908: The participating node sends intermediate data to the initiating node.

[0357] Refer to the aforementioned step 606.

[0358] Correspondingly, the initiating node receives the intermediate data and verifies the intermediate data according to the test set obtained in the aforementioned step.

[0359] 909: The initiating node updates and verifies the intermediate data according to the test set.

[0360] For example, if the intermediate data includes gradient and / or weight information, the model of the test set can be updated using the intermediate data, for example, to obtain a first model. Then, the test set is used for testing, and the test set result is compared. If the updated first model has improved accuracy compared to the test model of the current test set, the intermediate data is used to update the target model. Otherwise, the intermediate data is discarded.

[0361] Optionally, if the updated test model has a deviation greater than a preconfigured value compared to the test set result, the following steps can be performed.

[0362] 910: The initiating node sends a notification message to the coordination node, which is used to indicate abnormal data or an abnormal node.

[0363] Optionally, the initiating node sends a notification message to the coordination node. Specifically, the notification message can be triggered according to a preset strategy based on the specific deviation of the test result. For example, if the deviation value reaches or exceeds a first threshold, the notification message is triggered to be sent to the coordination node. Alternatively, if the number of deviations of the test result of the participating node is greater than or equal to a preset number of times, the notification message is triggered to be sent to the coordination node. The present application does not make specific limitations on this.

[0364] Optionally, the notification message can include the identification of the participating node, such as the first device, or the identification of the abnormal intermediate data.

[0365] 912: The coordination node processes according to a preset strategy.

[0366] Specifically, reference can be made to steps 609-610 in the foregoing embodiments.

[0367] For example, according to at least one of the first strategies described above, the first device is added to a gray list or a black list, and the intermediate data of the first device is discarded.

[0368] 913: The initiating node performs model aggregation and sends a response message to the consumption node.

[0369] For example, after the initiating node completes the model aggregation of federated learning, the initiating node can send a response message to the consumption node, which can carry the model obtained by the aggregation processing.

[0370] The model aggregation method refers to generating a single global model by combining several models when the initiating node receives models sent by several participating nodes (such as NWDAF in each region). The present application does not make limitations on the specific implementation of model aggregation. For example, any one of the following methods can be used to aggregate the model:

[0371] Simple average: the models of all participating nodes are averaged. For example, for models of the same structure, the model parameters can be averaged to obtain a global model. For example, the initiating node receives model 1, model 2, and model 3 of three participating nodes, and obtains a global model by using the simple average model aggregation method. The model parameters of the global model can be the average of the model parameters corresponding to the above three models.

[0372] Weighted average: before averaging the model, the weighted average can be performed according to the quality of the model or the number of training data. For example, according to a preconfigured rule, the model can be labeled with a weight corresponding to the model, and the weight coefficient is multiplied when the average is taken to obtain a global model.

[0373] Federal Average: an algorithm for calculating weighted average, which can get a weighted average value by weighting the data points according to a certain weight. For example, at the beginning of each round of model aggregation, a small proportion of participating nodes are selected randomly, and then the selected nodes can send the current global algorithm state (current model parameters) to each selected participating node. After repeating the selection process for multiple rounds, the aggregated global model is obtained by averaging according to the preconfigured algorithm.

[0374] Hybrid method: combination of the above model aggregation techniques.

[0375] The various embodiments mentioned in the foregoing of the present application can be combined without contradiction in the scheme, without limitation.

[0376] The above mainly introduces the scheme provided by the present application from the perspective of interaction between various network nodes. Correspondingly, the present application also provides a communication device, which can be each communication device or node in the above method embodiment, or a component such as a chip that can be used for the above communication device or node.

[0377] It can be understood that the communication device described above contains the corresponding hardware structure and / or software module for implementing each function. Those skilled in the art should easily realize that, in combination with the unit and algorithm operation of each example described in the embodiments disclosed in the present application, the present application can be realized in the form of hardware or a combination of hardware and computer software. Whether a certain function is executed in hardware or computer software driven hardware depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0378] It should be understood that the above only describes the interaction between each network element node by example. In fact, the processing performed by the above communication device or node is also not limited to being performed only by a single network element.

[0379] The present application can divide the functional modules of the communication device according to the above method examples, for example, each functional module can be divided according to each function, or two or more functions can be integrated into one processing module. The above integrated module can be realized in the form of hardware or software functional module. It can be understood that the division of modules in the present application is illustrative, and is only a logical functional division. Actual implementation can have another division method.

[0380] The various embodiments mentioned in the foregoing of the present application can be combined without contradiction in the scheme, without limitation.

[0381] The above describes the scheme provided by the present application mainly from the perspective of interaction between network nodes. Correspondingly, the present application also provides a communication device, which can be each communication device or node in the above method embodiments, or a component such as a chip that can be used for the above communication device or node.

[0382] It can be understood that the communication device described above contains hardware structures and / or software modules corresponding to the implementation of each function in order to achieve the above functions. Those skilled in the art should easily realize that, in combination with the unit and algorithm operation of each example described in the embodiments disclosed herein, the present application can be realized in the form of hardware or a combination of hardware and computer software. Whether a certain function is executed in the form of hardware or computer software driving hardware depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0383] It should be understood that the above only describes the interaction between each network element node by way of example. In fact, the processing performed by the above communication device or node is also not limited to being performed by a single network element.

[0384] The present application can divide the functional modules of the communication device according to the above method examples, for example, each functional module can be divided according to each function, or two or more functions can be integrated into one processing module. The above integrated module can be realized in the form of hardware or software functional module. It can be understood that the division of modules in the present application is illustrative, and is only a logical functional division. Actual implementation can have another division manner.

[0385] For example, in the case of dividing each functional module in an integrated manner, FIG. 10 shows a structural schematic diagram of a communication device 1000. The communication device 1000 includes an interface module 1001 and a processing module 1002.

[0386] In some embodiments, the communication device 1000 can also include a storage module (not shown in FIG. 10) for storing program instructions and data.

[0387] Exemplarily, the communication device 1000 can be used to implement the functions of the communication device in the above embodiment of FIG. 4A. The communication device 1000 is, for example, the coordination node or the initiation node described in the above embodiment of FIG. 6.

[0388] The interface module 1001 can be used to receive a first message, the first message including intermediate data of machine learning, the intermediate data coming from a first device, the first device being a participating node of the machine learning task.

[0389] The processing module 1002 can be configured to verify the intermediate data, and determine that the intermediate data is abnormal.

[0390] The interface module 1001 can also be configured to send a second message, where the second message is used to indicate that the data feedback is abnormal, and the second message further includes a first policy used to indicate processing on the intermediate data or the first device.

[0391] In an embodiment, the second message includes information of the intermediate data, and / or information of the first device corresponding to the intermediate data.

[0392] In an embodiment, the first policy is used to indicate at least one of the following: discarding the intermediate data from the first device; discarding N times of intermediate data from the first device, where N is a positive integer; marking the first device as a first type or a second type; or updating the intermediate data.

[0393] In an embodiment, if the intermediate data corresponding to the first device is abnormal, the processing module 1002 can be configured to mark the first device as the first type.

[0394] In an embodiment, if the number of times that the intermediate data corresponding to the first device is abnormal is greater than or equal to m times, the processing module 1002 can be configured to mark the first device as the second type, where m is a positive integer.

[0395] In an embodiment, the processing module 1002 can be configured to determine a stage of the machine learning, and determine whether the intermediate data is abnormal according to the stage of the machine learning.

[0396] In an embodiment, if the stage of the machine learning is in a convergence stage, the processing module 1002 can be configured to perform at least one of the following: verifying variance of a gradient in the intermediate data of the machine learning, verifying information divergence of the gradient, or verifying whether the gradient exceeds a deviation value.

[0397] In an embodiment, the first message further includes an identifier of the first device.

[0398] In an embodiment, the processing module 1002 can be configured to discard the intermediate data, and perform model updating on the model of the machine learning.

[0399] In an embodiment, the interface module 1001 can also be configured to send a second message to the first device or a second device, where the second message is used to indicate that the data feedback of the first device is abnormal, and the second device is a network repository function or a unified data management function.

[0400] In an embodiment, the second message is used to indicate at least one of the following information: de-registering the first device; de-registering a data collection node managed by the first device; and stopping collecting data of an external node of the communication network.

[0401] In addition, the present application also provides a communication device, the communication device 1000 can be used to implement the function of the machine learning participating node in the above-mentioned embodiments of FIG. 4A or FIG. 6.

[0402] The interface module 1001 can be used to send a first message, the first message including intermediate data of machine learning, the first message being used to request verification of the intermediate data; wherein the intermediate data is from a first device, the first device being a participating node of the machine learning task.

[0403] The interface module 1001 can also be used to receive a second message, the second message being used to indicate data feedback exception; the second message further including a first policy, the first policy being used to indicate processing of the intermediate data with the exception or the first device performing the processing.

[0404] In an embodiment, the second message includes information of the intermediate data with the exception and / or information of the first device corresponding to the intermediate data with the exception.

[0405] In addition, the present application also provides a communication device, the communication device 1000 can be used to implement the function of the machine learning initiating node in the above-mentioned embodiments of FIG. 5 or FIG. 7.

[0406] The interface module 1001 can be used to receive a first message from a first device, the first message including intermediate data of machine learning and the first token.

[0407] The processing module 1002 can be used to verify the first token, and if the verification is passed, determine an update of a model according to the intermediate data.

[0408] In an embodiment, the interface module 1001 can be used to send information of a machine learning coordination node to the first device, the information being used to indicate that the first device requests authorization of machine learning from the coordination node.

[0409] In an embodiment, the processing module 1002 can be used to interact with the machine learning coordination node, and verify at least one of the following information: verifying whether valid signature information of the coordination node is included in the first token; verifying whether the certificate included in the first token is valid; or, verifying whether the first token is within a valid period.

[0410] In addition, the application further provides a communication device, and the communication device 1000 can be used to implement the function of the machine learning coordination node in the above-mentioned embodiments of FIG. 5 or FIG. 7.

[0411] The interface module 1001 can be used to receive a second message from a first device, and the second message is used to request authorization of the first device to send machine learning intermediate data to a machine learning initiation node; wherein the first device is a participating node of the machine learning task.

[0412] The processing module 1002 can be used to verify the first device, and if the verification is passed, generate a first token corresponding to the first device; and send the first token to the first device.

[0413] In an embodiment, the processing module 1002 can be used to determine the node type of the first device; and then verify the first device according to the security information corresponding to the node type.

[0414] In an embodiment, if the first device is an external application function, the processing module 1002 can be used to determine the security start information corresponding to the external application function, or obtain the related security information from a third-party verification and authorization server; and verify the first device according to the security start information or the security information; if the first device is a network function within the communication system, the processing module 1002 can be used to query the registration information of the network function, or obtain the checking result of the configuration information of the network function from an operation and maintenance management system; and determine whether the first device is verified according to the registration information or the checking result.

[0415] In an embodiment, the first token includes a validity period, wherein the validity period of the first token corresponding to different node types is different.

[0416] In an embodiment, if the verification fails, the interface module 1001 can be used to send a second message to the first device or a second device, and the second message is used to indicate that the first device data feedback is abnormal; wherein the second device is a network repository function or a unified data management function.

[0417] In an embodiment, the second message is used to indicate at least one of the following information: deregister the first device; deregister the data collection node managed by the first device; and stop collecting data of external nodes of the communication network.

[0418] In addition, the application further provides a communication device, and the communication device 1000 can be used to implement the function of the machine learning initiation node in the above-mentioned embodiments of FIG. 8 or FIG. 9.

[0419] The interface module 1001 can be configured to receive intermediate data of machine learning from a first device, the first device being a participating node of the machine learning task; and perform model updating on a test set according to the intermediate data, to obtain a first model.

[0420] If an accuracy deviation between the first model and a model corresponding to the test set is greater than or equal to a first threshold, the interface module 1001 can be configured to send a first message, to indicate that the intermediate data of the first device is abnormal.

[0421] In an embodiment, if the model accuracy deviation between the first model and the test model is less than a second threshold, the processing module 1002 can be configured to update a target model according to the intermediate data.

[0422] In an embodiment, the interface module 1001 can be configured to receive a first policy, to perform corresponding processing on the abnormal intermediate data or the first device.

[0423] In an embodiment, the first policy is used to indicate at least one of the following information: discarding the intermediate data from the first device; discarding intermediate data from the first device for N times, where N is a positive integer; marking the first device as a first type or a second type; or updating the intermediate data.

[0424] In an embodiment, the interface module 1001 can be configured to receive a second message, the second message including the test set and a model corresponding to the test set.

[0425] In an embodiment, the second message is carried in a federated learning preparation message.

[0426] In addition, the present application also provides a communication device, and the communication device 1000 can be configured to implement the function of the coordination node of machine learning in the embodiments of FIG. 8 or FIG. 9.

[0427] The interface module 1001 can be configured to send a second message, the second message including the test set and a model corresponding to the test set.

[0428] The interface module 1001 can also be configured to receive a first message, the first message including information of a first device, the first message being used to indicate that intermediate data of machine learning of the first device is abnormal; and the first device being a participating node of the machine learning task.

[0429] In an embodiment, the processing module 1002 can be configured to determine a first policy according to the first message, the first policy being used for performing corresponding processing on the abnormal intermediate data or the first device; and the interface module 1001 can be further configured to send the first policy.

[0430] In an embodiment, the first policy is used to indicate at least one of the following information: discarding intermediate data from the first device; discarding N times of intermediate data from the first device, where the N is a positive integer; marking the first device as a first type or a second type; or updating the intermediate data.

[0431] In an embodiment, the interface module 1001 can be further configured to receive a third message including requirement information of the machine learning task; determine the test set corresponding to the machine learning task and the model corresponding to the test set according to the requirement information.

[0432] In an embodiment, the second message is carried in a federated learning preparation message.

[0433] In summary, when the communication device 1000 is configured to implement the functions performed by the communication device or the node in the above embodiments, other functions that can be implemented by the communication device 1000 can be referred to the related descriptions of any of the above embodiments.

[0434] In a simple embodiment, those skilled in the art can conceive that the communication device 1000 can adopt the form shown in FIG. 3. For example, the processor 301 in FIG. 3 can execute the above-mentioned method by invoking the computer execution instructions stored in the memory 303, so that the communication device 300 performs the method described in the above method embodiments.

[0435] For example, the functions / implementation processes of the processing module 1002 in FIG. 10 can be implemented by the processor 301 in FIG. 3.

[0436] For example, the functions / implementation processes of the interface module 1001 in FIG. 10 can be implemented by the communication interface 304 in FIG. 3.

[0437] It can be understood that one or more of the above modules or units can be implemented in software, hardware or a combination of both. When any of the above modules or units is implemented in software, the software exists in the form of computer program instructions and is stored in the memory, and the processor can be used to execute the program instructions and implement the above method flow. The processor can be built in the SoC (System on Chip) or ASIC, or be a separate semiconductor chip. The processor further includes the core for executing software instructions to perform operations or processing, and can further include necessary hardware accelerators, such as field programmable gate array (FPGA), programmable logic device (PLD), or logic circuit for implementing special logic operations.

[0438] When any of the above modules or units is implemented in hardware, the hardware can be any one or any combination of CPU, microprocessor, digital signal processing (DSP) chip, microcontroller unit (MCU), artificial intelligence processor, ASIC, SoC, FPGA, PLD, special purpose digital circuit, hardware accelerator or non-integrated discrete device, which can run necessary software or be independent of software to execute the above method flow.

[0439] In a possible implementation, the present application also provides a chip system, including at least one processor and an interface, the at least one processor is coupled with the memory through the interface, when the at least one processor executes the computer program or instructions in the memory, the method in any of the above method embodiments is executed. In a possible implementation, the chip system further includes the memory. Optionally, the chip system can be composed of a chip, or can include the chip and other discrete devices, and the present application does not make specific limitation hereon.

[0440] Optionally, the present application also provides a computer readable storage medium. All or part of the processes in the above method embodiments can be instructed by a computer program to relevant hardware to complete, and the program can be stored in the above computer readable storage medium. When the program is executed, the program can include the processes of the above method embodiments. The computer readable storage medium can be an internal storage unit of the communication device in any of the above embodiments, such as a hard disk or a memory of the communication device. The computer readable storage medium can also be an external storage device of the communication device, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, and the like. Further, the computer readable storage medium can include both the internal storage unit and the external storage device of the communication device. The computer readable storage medium is used to store the above computer program and other programs and data required by the communication device. The computer readable storage medium can also be used to temporarily store data that has been output or will be output.

[0441] In a possible implementation, the present application also provides a computer program product. All or part of the processes in the above method embodiments can be instructed by a computer program to relevant hardware to complete, and the program can be stored in the above computer program product. When the program is executed, the program can include the processes of the above method embodiments.

[0442] In a possible implementation, the present application also provides a computer instruction. All or part of the processes in the above method embodiments can be instructed by a computer instruction to relevant hardware (such as a computer, a processor, a network device, or a terminal device) to complete. The program can be stored in the above computer readable storage medium or the above computer program product.

[0443] In a possible implementation, the present application also provides a communication system, including the first device in the above embodiments and a communication device. Exemplarily, the first device can be a participating node of machine learning, and the communication device can be an initiating node or a coordinating node of machine learning.

[0444] In a possible implementation, the communication system can include a participating node (such as the first device), an initiating node, and a coordinating node of machine learning in the above embodiments.

[0445] From the above description of the embodiments, those skilled in the art can clearly understand that, for the convenience and brevity of description, only the division of the above functional modules is taken as an example for illustration. In actual applications, the above functions can be completed by different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the above described functions.

[0446] In several embodiments provided in the present application, it should be understood that the disclosed apparatus and method can be implemented in other manners. For example, the embodiments of the apparatus described above are merely schematic; the division of the modules or units is merely logical function division; an actual implementation can be another division manner, for example, multiple units or components can be combined or integrated into another apparatus, or some features can be ignored or not executed. In addition, the displayed or discussed mutual couplings or direct couplings or communication connections can be indirect couplings or communication connections through some interfaces, apparatuses or units, and can be in electrical, mechanical or other forms.

[0447] The units described as separated components can or can not be physically separated, and the components displayed as units can be located in one place or distributed to multiple places. Some or all of the units can be selected according to actual needs to achieve the purposes of the embodiments of the present application.

[0448] In addition, each functional unit in the embodiments of the present application can be integrated in a processing unit, or each unit can exist physically as a separate unit, or two or more units can be integrated in one unit. The integrated unit can be implemented in the form of hardware, or in the form of a software functional unit.

[0449] The above describes only specific embodiments of the present application, but the protection scope of the present application is not limited thereto, and any changes or replacements within the technical scope disclosed in the present application should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A communication method, characterized in that, The method includes: Receive a first message, the first message including intermediate data for machine learning, the intermediate data coming from a first device, the first device being a participating node in the machine learning task; The intermediate data is verified to determine if there is any abnormal intermediate data. A second message is sent to indicate that the data feedback is abnormal; the second message also includes a first strategy to indicate the processing to be performed on the abnormal intermediate data or the first device.

2. The method according to claim 1, characterized in that, The second message includes information about the abnormal intermediate data and / or information about the first device corresponding to the abnormal intermediate data.

3. The method according to claim 1 or 2, characterized in that, The first strategy is used to indicate at least one of the following information: Discard intermediate data from the first device; Discard N times of intermediate data from the first device, where N is a positive integer; The first device is labeled as either type one or type two; or, The intermediate data is then updated.

4. The method according to any one of claims 1-3, characterized in that, The method further includes: If the intermediate data corresponding to the first device is abnormal, the first device will be marked as the first type.

5. The method according to any one of claims 1-4, characterized in that, The method further includes: If the intermediate data corresponding to the first device has an anomaly more than or equal to m times, then the first device is marked as the second type, where m is a positive integer.

6. The method according to any one of claims 1-5, characterized in that, The validation of intermediate data for machine learning includes: Determine the stage of the machine learning; The presence of anomalies in the intermediate data is determined based on the stage of the machine learning process.

7. The method according to any one of claims 1-6, characterized in that, If the machine learning process is in the convergence phase, then the verification of the intermediate machine learning data includes at least one of the following: Verify the variance of the gradient in the intermediate data of the machine learning, verify the information divergence of the gradient, or verify whether the gradient exceeds the deviation value.

8. The method according to any one of claims 1-7, characterized in that, The first message also includes the identifier of the first device.

9. The method according to any one of claims 1-8, characterized in that, The method further includes: The abnormal intermediate data is removed, and the machine learning model is updated.

10. The method according to any one of claims 1-9, characterized in that, The method further includes: A second message is sent to the first device or the second device to indicate that the first device has a data feedback anomaly, wherein the second device is a network repository function or a unified data management function.

11. The method according to claim 10, characterized in that, The second message is used to indicate at least one of the following: Go to register the first device; Register the data collection node managed by the first device; Stop collecting data from external nodes of the communication network.

12. A communication method, characterized in that, The method includes: Send a first message, the first message including intermediate data of machine learning, the first message being used to request verification of the intermediate data; wherein, the intermediate data comes from a first device, the first device being a participating node of the machine learning task; A second message is received, which indicates that the data feedback is abnormal; the second message also includes a first strategy, which indicates the processing to be performed on the abnormal intermediate data or the first device.

13. The method according to claim 12, characterized in that, The second message includes information about the abnormal intermediate data and / or information about the first device corresponding to the abnormal intermediate data.

14. A communication method, characterized in that, The method includes: Receive a first message from a first device, the first message including intermediate data from machine learning and the first token; The first token is verified. If the verification passes, the model update is determined based on the intermediate data.

15. The method according to claim 14, characterized in that, The method further includes: The system sends information about a machine learning coordination node to the first device, instructing the first device to request authorization for machine learning from the coordination node.

16. The method according to claim 14 or 15, characterized in that, The verification of the first token includes: Interact with the coordinating node of the machine learning to verify at least one of the following: Verify whether the first token contains the signature information of a valid coordinating node; Verify whether the certificate included in the first token is valid; or, Verify whether the first token is within its validity period.

17. A communication method, characterized in that, The method includes: A second message is received from the first device, which requests authorization for the first device to send intermediate data of machine learning to the initiating node of the machine learning task; wherein the first device is a participating node of the machine learning task. The first device is verified. If the verification is successful, a first token corresponding to the first device is generated. Send the first token to the first device.

18. The method according to claim 17, characterized in that, Verification of the first device includes: Determine the node type of the first device; The first device is verified based on the security information corresponding to the node type.

19. The method according to claim 17 or 18, characterized in that, Verification of the first device includes: If the first device is an external application function, then determine the security startup information corresponding to the external application function, or obtain relevant security information from a third-party verification and authorization server; verify the first device based on the security startup information or the security information; If the first device is a network function within a communication system, then the registration information of the network function is queried and obtained, or the verification result of the configuration information of the network function is obtained from the operation and maintenance management system; based on the registration information or verification result, it is determined whether the first device has passed verification.

20. The method according to any one of claims 17-19, characterized in that, The first token includes an expiration period, wherein the expiration period of the first token varies for different node types.

21. The method according to any one of claims 17-20, characterized in that, If verification fails, the method further includes: A second message is sent to the first device or the second device to indicate that the first device has a data feedback anomaly; wherein the second device is a network repository function or a unified data management function.

22. The method according to claim 21, characterized in that, The second message is used to indicate at least one of the following: Go to register the first device; Register the data collection node managed by the first device; Stop collecting data from external nodes of the communication network.

23. A communication device, characterized in that, It includes at least one module or unit for implementing the method as described in any one of claims 1-22.

24. A communication device, characterized in that, include: A processor coupled to a memory for storing a program or instructions which, when executed by the processor, cause the method as described in any one of claims 1-22 to be performed.

25. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, When the computer program or instructions are executed, the method as described in any one of claims 1-22 is performed.

26. A computer program product, the computer program product comprising computer program code, characterized in that, When the computer program code is run on a computer, the method as described in any one of claims 1-22 is performed.

Images