Communication method, communication node, medium, and program product

Abstract

The present application discloses a communication method, a communication node, a medium, and a program product. The communication method comprises: sending a client information reporting request to at least one second communication node; receiving client information reported by each second communication node in response to the client information reporting request; and determining a federated learning type on the basis of each piece of client information, and sending the federated learning type to each second communication node.

Description

Communication methods, communication nodes, media and software products Technical Field

[0001] This application relates to the field of communication technology, and in particular to a communication method, communication node, medium, and program product. Background Technology

[0002] In the 5th Generation Mobile Communication (5G) system, the Network Data Analytics Function (NWDAF) is a 5G Core Network Function (5GC NF) located in the control plane, which can perform statistical data and machine learning related tasks in the 5G system.

[0003] In 5G systems, multiple NWDAFs and / or Application Functions (AFs) are supported as participants for federated learning training and / or inference. However, the method for determining the type of federated learning required for each participant remains unclear. Summary of the Invention

[0004] This application provides a communication method, communication node, medium, and program product to solve the problem of unclear federated learning type selection during federated learning. It clarifies the federated learning type required by each participant during the federated learning preparation / training process, thereby improving the scenario adaptability of the model trained through federated learning.

[0005] This application provides a communication method applied to a first communication node, comprising: sending a client information reporting request to at least one second communication node; receiving client information reported by each second communication node in response to the client information reporting request; determining a federated learning type based on the client information, and sending the federated learning type to each second communication node.

[0006] This application provides a communication method applied to a second communication node, comprising: receiving a client information reporting request sent by a first communication node and reporting client information to the first communication node; and receiving a federated learning type sent by the first communication node.

[0007] This application provides a communication method applied to a first communication node, comprising: receiving first information and making a decision based on the first information; determining a target federated learning type based on the decision result, and sending the target federated learning type to at least one second communication node.

[0008] This application provides a communication node, including: a memory, a processor, a program stored in the memory and executable on the processor, and a data bus for implementing communication between the processor and the memory. When the program is executed by the processor, it implements the communication method as described in any of the embodiments of this application.

[0009] This application provides a storage medium for computer-readable storage, which stores one or more programs that can be executed by one or more processors to implement any of the communication methods described in this application.

[0010] This application provides a computer program product, including a computer program, which, when executed by a processor, implements any of the communication methods described in this application.

[0011] The communication method, communication node, medium, and program product provided in this application embodiment involve sending a client information reporting request to at least one second communication node; receiving client information reported by each second communication node in response to the client information reporting request; determining the federated learning type based on the client information; and sending the federated learning type to each second communication node. By adopting the above technical solution, during the federated learning preparation process, the type of federated learning to be performed is determined based on the received client information, clarifying the method for determining the federated learning type to be adopted by each federated learning participant. Furthermore, the process of determining the federated learning type fully considers the capability information of each participant, thereby improving the scenario adaptability of the model trained through the federated learning type. Attached Figure Description

[0012] Figure 1 is a schematic diagram of a 5G core network architecture provided in related technologies;

[0013] Figure 2 is a flowchart of a communication method provided in an embodiment of this application;

[0014] Figure 3 is a flowchart of a communication method provided in an embodiment of this application;

[0015] Figure 4 is a flowchart of a communication method provided in an embodiment of this application;

[0016] Figure 5 is a flowchart of a communication method provided in an embodiment of this application;

[0017] Figure 6 is a timing example diagram of federated learning preparation provided by an embodiment of this application;

[0018] Figure 7 is a timing example diagram of a federated learning training provided in an embodiment of this application;

[0019] Figure 8 is a timing example diagram of a federated learning capability reporting provided in an embodiment of this application;

[0020] Figure 9 is a schematic diagram of the structure of a communication device provided in an embodiment of this application;

[0021] Figure 10 is a schematic diagram of the structure of a communication device provided in an embodiment of this application;

[0022] Figure 11 is a schematic diagram of the structure of a communication device provided in an embodiment of this application;

[0023] Figure 12 is a schematic diagram of the structure of a communication device provided in an embodiment of this application;

[0024] Figure 13 is a schematic diagram of the structure of a communication node provided in an embodiment of this application. Detailed Implementation

[0025] Unless otherwise specified, the embodiments and features described in this application may be combined arbitrarily with each other.

[0026] The operations illustrated in the flowcharts in the accompanying drawings can be performed on a computer system, such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowcharts, in some cases, the operations shown or described may be performed in a different order than that presented here.

[0027] The communication method provided in this application can be applied to 5G systems to determine the federated learning type during the preparation or training process of multiple NWDAFs and / or AFs that support federated learning training and / or inference. To clearly describe federated learning in 5G systems, a brief introduction is given here of the 5G core network architecture and the NWDAFs that perform federated learning as 5GC NFs based on the 5G core network architecture.

[0028] Figure 1 is a schematic diagram of a 5G core network architecture provided in related technologies. This architecture has the following functions:

[0029] 1) User Equipment (UE).

[0030] 2) Radio Access Network (RAN). The RAN manages radio resources, transmits user data received through the N3 interface to the UE, and transmits user data from the UE through the N3 interface. The RAN maps between Quality of Service (QoS) traffic in Dedicated Radio Bearer (DRB) and Protocol Data Unit (PDU) sessions.

[0031] 3) Access and Mobility Management Function (AMF). This function includes registration management, connection management, reachability management, and mobility management. It also performs access authentication and authorization. AMF is a Network Attached Storage (NAS) security endpoint used to forward SM NAS data between the UE and the Session Management Function (SMF).

[0032] 4) SMF. This function includes the following: session establishment, modification, and release; UE IP address allocation and management (including optional authorization functions); selection and control of User Plane (UP) functions and downlink data notification, etc. The SMF controls the User Plane Function (UPF) through the N4 interface. The SMF provides the UPF with Packet Detection Rules (PDR) to indicate how to detect user data traffic; provides Forwarding and Routing Control Association Rules (FAR); and provides QoS enforcement rules and Usage Reporting Rules (URR) to indicate how the UPF performs user data traffic forwarding, QoS processing, and usage reporting on user data traffic detected using PDR.

[0033] 5) UPF. This function includes the following: serving as an anchor point for intra / inter-radio access type mobility, packet routing and forwarding, traffic usage reporting, QoS processing for the UP, downlink packet buffering, and downlink data notification triggering. The General Packet Radio Service (GPRS) Tunneling Protocol for User Plane (GTP-U) tunnel is used for the N3 interface between the RAN and UPF. The GTP-U tunnel operates on a per-PDU session basis. For downlink traffic, the UPF binds the downlink traffic to the QoS traffic within the PDU session's GTP-U tunnel using the FAR received from the SMF. For uplink traffic, the RAN transmits user plane traffic to the QoS stream identified by the UE.

[0034] 6) Policy Control Function (PCF). The PCF provides QoS policy rules to control plane functions for enforcement. The PCF translates AF requests into PCC rules applicable to PDU sessions.

[0035] 7) Unified Data Management (UDM). The UDM performs tasks such as generating 3GPPAKA authentication credentials, granting access based on subscription data, managing UE service NF registration (e.g., storing AMF for UE storage services, SMF for UE PDU session storage services), and subscription management. The UDM accesses the UDR to retrieve UE subscription data and stores the UE context in the UDR. The UDM and UDR can be deployed together.

[0036] Based on the 5G core network architecture, NWDAF is a 5GC NF located in the control plane, which performs data statistics and machine learning related tasks in the 5G system. NWDAF can interact with different entities for various purposes: collect data based on event subscriptions provided by AMF, SMF, UPF, PCF, UDM, NSCAF, AF (directly or via NEF), and Operations, Administration and Maintenance (OAM); [optionally] perform analysis and data collection using the Data Collection Coordination Function (DCCF); retrieve information from data repositories (e.g., retrieve UDRs related to users via UDM or PFD information via NEF (PFDF)); collect location information data from LCS systems; [optionally] store and retrieve information from the Analytics Data Storage Function (ADRF); [optionally] analyze and collect data from the Messaging Framework Adaptor Function (MFAF); retrieve information about NFs (e.g., retrieve NF-related information from NRF); provide analytics to consumers on demand; provide bulk data associated with analytics IDs; provide accuracy information for analytics IDs; provide ML model accuracy information or ML model accuracy degradation indicators for machine learning (ML) models.

[0037] In some examples, a single instance or multiple instances of NWDAF can be deployed in a public terrestrial mobile network. NWDAF may contain the following logical functions: Analysis Logic Function (AnLF): A logical function in NWDAF that performs inference, derives analytical information (i.e., derives statistics and / or predictions based on analytical consumer requests), and exposes analytical services (i.e., Nnwdaf_AnalyticsSubscription or Nnwdaf_AnalyticsInfo).

[0038] Model Training Logic Function (MTLF): A logic function in NWDAF used to train ML models and expose new training services (such as providing pre-trained ML models).

[0039] An NWDAF can contain a Model Training Logical Function (MTLF) or an Analysis Logic Function (AnLF), or both.

[0040] The Data Collection Coordination Function (DCCF) is also an NF on the 5G core network control plane. The DCCF is responsible for coordinating the collection and distribution of data requested by NF consumers. It prevents data sources from processing multiple subscriptions to the same data and prevents multiple notifications containing the same information from being sent due to incoordination of data consumer requests.

[0041] DCCF is applicable to: NWDAFs that request data from data sources (such as those used for computational analysis); NF consumers that request analysis from NWDAF data sources; NF consumers that request data from ADRF data sources; and ADRFs that receive data from NF data sources.

[0042] To clearly describe the application of federated learning in 5G systems, the following is a brief introduction to horizontal federated learning and vertical federated learning:

[0043] Vertical federated learning: Suitable for situations where there is little overlap in the data features of participants, but a lot of overlap in sample IDs. In vertical federated learning, the training data of different client models have the same sample space but different feature spaces.

[0044] In the already defined core network vertical federated learning, NWDAF / AF can serve as a vertical federated learning server and / or a vertical federated learning client.

[0045] Horizontal federated learning is suitable for situations where participants' data features overlap significantly, while sample IDs overlap less. In horizontal federated learning, different client models train on data with the same feature space but different sample spaces.

[0046] In the already defined core network horizontal federated learning, only NWDAF can serve as a horizontal federated learning server and / or a horizontal federated learning client.

[0047] However, although 5G systems support multiple NWDAFs and / or AFs as participants in federated learning and / or inference, they do not explicitly provide a method for selecting the federated learning type during the federated learning process, making it impossible to select the most suitable federated learning type based on the capabilities of the federated learning participants. To address the above issues, embodiments of this application provide a communication method, which can be implemented at a first communication node and / or a second communication node. The first and second communication nodes are respectively nodes acting as federated learning servers and clients among the participants in federated learning in the 5G system. The first and / or second communication nodes are generally electronic devices with certain computing capabilities. In embodiments of this application, the first and / or second communication nodes can be NWDAFs and / or AFs. In some possible implementations, the communication method can be implemented by a processor calling computer-executable instructions stored in memory.

[0048] In one exemplary embodiment, Figure 2 is a flowchart of a communication method provided in an embodiment of this application. This method can be applied to the process of preparing for federated learning in a 5G system, where a federated learning type is selected for communication nodes participating in federated learning. This method can be executed by a communication device, which can be implemented by software and / or hardware and integrated on the communication node. This method can be applied to a first communication node.

[0049] As shown in Figure 2, the communication method provided in this application embodiment specifically includes S101-S103.

[0050] S101, Send a client information reporting request to at least one second communication node.

[0051] In this embodiment, the second communication node can be specifically understood as a communication node that participates in the same federated learning process as the first communication node.

[0052] In this embodiment, the client information reporting request can be specifically understood as a request for information from the communication node that needs to participate in federated learning to request its support capabilities for federated learning.

[0053] In a specific example, when preparing for federated learning, the first communication node sends a client information reporting request to at least one second communication node that it wishes to participate in the same federated learning, requesting the second communication node to provide feedback on its capabilities for federated learning.

[0054] S102. Receive the client information reported by each second communication node in response to the client information reporting request.

[0055] In this embodiment, client information can be specifically understood as information used to characterize the supported federated learning status or federated learning capability.

[0056] In a specific example, each second communication node that receives a client information reporting request feeds back its corresponding client information to the first communication node, which means that the first communication node can receive the client information reported by each second communication node in response to the client information reporting request.

[0057] S103. Determine the federated learning type based on the information from each client and send the federated learning type to each second communication node.

[0058] In this embodiment, the federated learning type can be specifically understood as a type that is determined based on the client information of each second communication node, and that each second communication node can support and carry out federated learning.

[0059] In some examples, federated learning types may include horizontal federated learning, vertical federated learning, multi-level horizontal federated learning, multi-level vertical federated learning, etc.

[0060] In a specific example, the first communication node determines the most suitable federated learning type based on the federated learning capability information of each second communication node contained in the received client information, and sends the federated learning type to each second communication node.

[0061] The communication method provided in this application involves sending a client information reporting request to at least one second communication node; receiving client information reported by each second communication node in response to the client information reporting request; determining the federated learning type based on the client information; and sending the federated learning type to each second communication node. By adopting the above technical solution, during the federated learning preparation process, the type of federated learning to be performed is determined based on the received client information, clarifying the method for determining the federated learning type to be adopted by each federated learning participant. Furthermore, the process of determining the federated learning type fully considers the capability information of each participant, thereby improving the scenario adaptability of the model trained using the federated learning type.

[0062] In some examples, after the first communication node determines the type of federated learning and, based on the determined type, determines the second communication nodes that will participate in the federated learning, it can store the list of the second communication nodes that need to participate in the federated learning and the client information corresponding to each second communication node in the list.

[0063] In one embodiment, the client information includes at least one of the following: data sample or feature information; privacy protection requirements and security capability information; data quality information; computing resource information; and a time period or duration during which horizontal or vertical federated learning can be performed.

[0064] In some examples, data sample or feature information may include sample identifiers, feature identifiers, total number of samples, feature dimensions, and time information corresponding to the sample or feature data.

[0065] In some examples, privacy requirements may include whether to support the exchange of models, intermediate training parameters, and intermediate results during federated learning.

[0066] In some examples, security capability information may include the types of encryption algorithms supported by the second communication node and homomorphic encryption capabilities.

[0067] In some examples, data quality information may include missing data information (such as the proportion of missing data values) and outlier information (such as the proportion of outliers).

[0068] In some examples, computing resource information may include computing power information (CPU / GPU specifications, computing power available for federated learning) and storage capacity available for federated learning, etc.

[0069] In one embodiment, determining the federated learning type based on each client's information includes at least one of the following: determining the federated learning type based on the number of overlapping data features and the number of overlapping sample identifiers in each client's information; determining the federated learning type based on the model exchange support capability and intermediate result transmission capability in each client's information; determining the federated learning type based on the homomorphic encryption support capability and security requirements in each client's information; determining the federated learning type based on the data quality information in each client's information; determining the federated learning type based on the computing resource information in each client's information; and determining the federated learning type based on the time period or duration in each client's information where horizontal or vertical federated learning can be performed.

[0070] In some examples, if the number of overlapping data features in the client information exceeds a preset data feature threshold, while the number of overlapping sample identifiers is less than or equal to a preset sample identifier threshold, then the federated learning type is determined to be horizontal federated learning.

[0071] In some examples, if the number of overlapping data features in each client's information is less than or equal to a preset data feature threshold, while the number of overlapping sample identifiers exceeds a preset sample identifier threshold, then the federated learning type is determined to be vertical federated learning.

[0072] In some examples, if most of the client information supports model exchange during the federated learning process but not the transfer of intermediate results, then the federated learning type is determined to be horizontal federated learning.

[0073] In some examples, if most of the client information does not support model disclosure during the federated learning process but does support intermediate result transmission, then the federated learning type is determined to be vertical federated learning.

[0074] In some examples, if most of the client information supports homomorphic encryption and the security requirements exceed a preset threshold, then the federated learning type is determined to be vertical federated learning; otherwise, the federated learning type is determined to be horizontal federated learning.

[0075] In some examples, the data quality information in each client's information is analyzed. If the number of missing samples or features in each client's information exceeds a preset missing threshold, then no federated learning can be performed.

[0076] In some examples, since horizontal federated learning requires more computing resources, the type of federated learning applicable to each second communication node can be determined based on the computing resource information in each client's information. If no second communication node has sufficient computing resources, then no federated learning can be performed.

[0077] In some examples, the appropriate type of federated learning can be determined based on the time period and duration of horizontal or vertical federated learning that can be performed in each client's information.

[0078] In one embodiment, before sending a client information reporting request to at least one second communication node, the method further includes: sending a discovery request to the Network Repository Function (NRF); wherein the discovery request includes horizontal federated learning client capabilities and / or vertical federated learning client capabilities; and receiving feedback from at least one second communication node from the Network Repository Function.

[0079] In this embodiment, the discovery request can be specifically understood as a request to the NRF to request communication nodes that can participate in federated learning.

[0080] In a specific example, during preparation for federated learning, a first communication node may send a discovery request to the NRF to request communication nodes that can participate in federated learning. The NRF then sends feedback to the first communication node indicating at least one second communication node that supports the type of federated learning requested in the discovery request.

[0081] In one embodiment, when the first communication node supports the federated learning type, sending the federated learning type to each of the second communication nodes includes sending a request to participate in federated learning to each of the second communication nodes.

[0082] In this embodiment, the request to participate in federated learning can be specifically understood as a request to the second communication node to participate in federated learning of the type of federated learning determined by the first communication node.

[0083] In a specific example, if the first communication node supports the federated learning type it has determined, it can be considered that the first communication node can participate in the federated learning preparation corresponding to that type. In this case, the first communication node sends a request to each second communication node to participate in the federated learning of that type.

[0084] In one embodiment, when the first communication node does not support the federated learning type, sending the federated learning type to each of the second communication nodes includes: determining a target first communication node that supports the federated learning type through a network storage function; sending the federated learning type, at least one second communication node information, and / or each client information to the target first communication node; and having the target first communication node send a request to each of the second communication nodes to participate in the federated learning.

[0085] In a specific example, if the first communication node does not support the determined federated learning type, to ensure that the second communication nodes can prepare for federated learning normally, a new first communication node that supports the federated learning type will be determined through NRF, and this new first communication node will be designated as the target first communication node. The current first communication node will then send the federated learning type, information about at least one previously acquired second communication node, and / or the client information corresponding to each second communication node to the target first communication node. The target first communication node will then send participation requests to each second communication node, requesting them to participate in the federated learning of this type.

[0086] In one embodiment, participating in a federated learning request includes at least one of the following: the type of federated learning; the time period for which the federated learning will take place; and the duration for which the federated learning will take place.

[0087] In one embodiment, after sending the federated learning type to each of the second communication nodes, the method further includes: receiving a joining federated learning instruction from each of the second communication nodes.

[0088] In this embodiment, the joining federated learning instruction can be specifically understood as an instruction used to indicate whether the second communication node that receives the request to participate in federated learning should join the federated learning of this type.

[0089] In a specific example, after the first communication node sends the federated learning type to each of the second communication nodes, each of the second communication nodes will determine whether to join the federated learning of that type based on its own federated learning capabilities, and will then send a feedback instruction to the first communication node to join the federated learning based on the determined result.

[0090] In one embodiment, the first communication node is a horizontal federated learning server or a vertical federated learning server in federated learning, and the second communication node is a horizontal federated learning client or a vertical federated learning client in federated learning.

[0091] In one exemplary embodiment, Figure 3 is a flowchart of a communication method provided in an embodiment of this application. This method can be applied to the process of preparing for federated learning in a 5G system, where a federated learning type is selected for communication nodes participating in federated learning. This method can be executed by a communication device, which can be implemented by software and / or hardware and integrated on the communication node. This method can be applied to a second communication node.

[0092] As shown in Figure 3, the communication method provided in this application embodiment specifically includes S201-S202.

[0093] S201. Receive the client information reporting request sent by the first communication node, and report the client information to the first communication node.

[0094] In a specific example, after receiving a client information reporting request sent by the first communication node, the second communication node will generate client information based on the information requested in the client information reporting request and report the client information to the first communication node.

[0095] S202, Receive the federated learning type sent by the first communication node.

[0096] In a specific example, after reporting client information, the second communication node will wait for the first communication node to make a decision based on the client information reported by each second communication node, and receive the federated learning type finally determined by the first communication node.

[0097] In one embodiment, the client information includes at least one of the following: data sample or feature information; privacy protection requirements and security capability information; data quality information; computing resource information; and a time period or duration during which horizontal or vertical federated learning can be performed.

[0098] In one embodiment, receiving the federated learning type sent by the first communication node includes: receiving a request to participate in federated learning sent by the first communication node.

[0099] In one embodiment, participating in a federated learning request includes at least one of the following: the type of federated learning; the time period for which the federated learning will take place; and the duration for which the federated learning will take place.

[0100] In one embodiment, after receiving the federated learning type sent by the first communication node, the method further includes: sending a feedback instruction to the first communication node to join the federated learning.

[0101] In a specific example, after receiving the federated learning type sent by the first communication node, the second communication node will determine whether to join the federated learning of that type based on its own federated learning capabilities, and will feed back the result of whether to join to the first communication node through a join federated learning instruction.

[0102] In one embodiment, the first communication node is a horizontal federated learning server or a vertical federated learning server in federated learning, and the second communication node is a horizontal federated learning client or a vertical federated learning client in federated learning.

[0103] In one exemplary embodiment, Figure 4 is a flowchart of a communication method provided in an embodiment of this application. This method can be applied to the situation where, during the federated learning training process of a 5G system, a federated learning type is selected for the communication nodes participating in the federated learning. This method can be executed by a communication device, which can be implemented by software and / or hardware and integrated on the communication node. This method can be applied to a first communication node.

[0104] As shown in Figure 4, the communication method provided in this application embodiment specifically includes S301-S302.

[0105] S301. Receive the first information and make a decision based on the first information.

[0106] In this embodiment, the first information can be specifically understood as information used to characterize the state of federated learning, or information on changes in the state of the second communication node participating in federated learning.

[0107] In this embodiment, the second communication node can be specifically understood as a communication node that participates in the same federated learning as the first communication node and whose federated learning state can be adjusted by the first communication node.

[0108] In a specific example, the first communication node receives first information to characterize the state of federated learning or the state changes of communication nodes participating in federated learning, and determines, based on the first information, whether each communication node participating in federated learning needs to be adjusted, and obtains a decision result including whether to switch the federated learning type and the required switch to the federated learning type.

[0109] S302. Determine the target federated learning type based on the decision result, and send the target federated learning type to at least one second communication node.

[0110] In this embodiment, the target federated learning type needs to be switched to.

[0111] In a specific example, after the first communication node makes a decision on the first information based on decision logic and obtains the target federated learning type to which it needs to switch, it will send the target federated learning type to at least one second communication node so that each second communication node switches its federated learning to the target federated learning type.

[0112] The communication method provided in this application receives first information and makes a decision based on the first information; determines the target federated learning type based on the decision result, and sends the target federated learning type to at least one second communication node. By adopting the above technical solution, during the federated learning training process, the federated learning capability and state changes of at least one second communication node participating in the training are determined based on the received first information, and then a decision on whether to adjust the federated learning type is made based on the first information. This enables the first communication node to dynamically adjust the federated learning type of the object participating in the federated learning process, improving the adaptability of federated learning to state changes.

[0113] In one embodiment, the first information is provided by a network storage function and / or at least one second communication node.

[0114] In one embodiment, before receiving the first information, at least one of the following is included: subscribing to the first information from a network storage function; subscribing to the first information from at least one second communication node; and at least one second communication node periodically sending the first information.

[0115] In a specific example, the first communication node can obtain the first information from the NRF and / or at least one second communication node in different ways. For instance, during the federated learning process, the first communication node can subscribe to federated learning-related information from at least one second communication node participating in the same federated learning process, so that when the federated learning status of at least one second communication node changes, the second communication node can send the first information to the first communication node. Alternatively, during the federated learning process, the first communication node can subscribe to the NRF regarding the capabilities of at least one second communication node participating in the same federated learning process, so that the NRF can send the first information to the first communication node when the capabilities of the second communication node change. Finally, because the first communication node participates in the same federated learning process, at least one second communication node can periodically and proactively send the first information to the first communication node.

[0116] In some examples, when a first communication node subscribes to first information from the NRF, the first communication node may send a subscription message to the NRF, which may include the identifier of the second communication node to be subscribed to.

[0117] In one embodiment, the first information includes at least one of the following: changes in the data samples or feature information of the second communication node; changes in the local privacy protection requirements of the second communication node; changes in the data quality information of the second communication node; the second communication node being unable to return the model or intermediate training results within the longest response time specified by the first communication node; changes in the capabilities of the second communication node; and changes in the time period / duration during which the second communication node can perform federated learning.

[0118] In some examples, changes in data samples or feature information may include sample increases, sample decreases, feature increases, and feature decreases.

[0119] In some examples, changes in data quality information may include changes in missing data information (such as changes in the proportion of missing data values) and changes in outlier information (such as changes in the proportion of outliers).

[0120] In some examples, changes in computing resource information may include changes in computing power information (such as changes in CPU / GPU specifications, changes in the amount of computing power available for federated learning), and changes in storage capacity available for federated learning.

[0121] In one embodiment, the decision result includes at least one of the following: switching the federated learning type; switching the federated learning type and model training instructions; wherein the model training instructions include: instructions to use the original training model or instructions to resend the new model for training; applying new weights to the intermediate training results / models of each second communication node; terminating or suspending the inference and / or training of the federated learning; and finding a new second communication node to join.

[0122] In a specific example, the first communication node can make decisions based on the federated learning status changes of each second communication node contained in the first information, including whether to switch the type of federated learning, whether to use the original model for training during the switch, whether to apply new weights to each second communication node, whether to terminate or pause federated learning, or whether to find new second communication nodes to join the federated learning.

[0123] In one embodiment, making a decision based on the first information includes at least one of the following: making a decision based on changes in data characteristics and sample identifiers; making a decision based on changes in privacy protection requirements; making a decision based on changes in data quality; making a decision based on changes in computing resources; making a decision based on changes in the federated learning period / duration; and making a decision based on the overall optimization of federated learning.

[0124] The following explains the internal logic of the first communication node making a decision based on the first information:

[0125] In one embodiment, making decisions based on changes in data features and sample identifiers includes at least one of the following: making decisions based on the amount of overlap between data features and sample identifiers; making decisions based on changes in the distribution of data features or sample identifiers.

[0126] In some examples, if the number of overlapping data features is greater than a preset data feature threshold, while the number of overlapping sample identifiers is less than or equal to a preset sample identifier threshold, the decision can be to switch to horizontal federated learning.

[0127] In some examples, if the number of overlapping data features is less than or equal to a preset data feature threshold, but the number of overlapping sample identifiers is greater than a preset sample identifier threshold, the decision can be to switch to vertical federated learning.

[0128] In some examples, if the distribution of data features or sample identifiers changes significantly (e.g., the overlap ratio of sample identifiers increases or decreases significantly due to data updates, and a sample identifier threshold can be set in advance to determine whether a significant change has occurred), the federated learning type can be re-evaluated, and the federated learning type can be switched if the evaluation result is different from the ongoing federated learning type.

[0129] In one embodiment, making decisions based on changes in privacy protection requirements includes at least one of the following: making decisions based on model exchange support capabilities and intermediate training result transfer capabilities; making decisions based on the amount of increase in privacy protection requirements.

[0130] In some examples, if a majority of the second communication nodes support model exchange during the federated learning process but do not support the transfer of intermediate results, the decision can be made to switch to horizontal federated learning.

[0131] In some examples, if a majority of the second communication nodes do not support model disclosure during the federated learning process but do support the transmission of intermediate results, then the decision can be made to switch to vertical federated learning.

[0132] In some examples, if the majority of privacy requirements in each of the second communication nodes are increased (such as requiring stronger homomorphic encryption), the decision may be to switch to vertical federated learning to reduce the risk of data breaches.

[0133] In one embodiment, making decisions based on changes in data quality includes making decisions based on the proportion of missing data or the proportion of outliers.

[0134] In some examples, if the increase in the proportion of missing data or outliers reported by the second communication node exceeds a preset threshold, it may lead to a decrease in training quality. Based on this, the following decisions can be made: adjust the weights of the second communication nodes to reduce the impact of the second communication nodes with a high proportion of outliers in federated learning; suspend or remove some second communication nodes (if the quality problem of the removed second communication nodes is serious, the severity can be determined by setting a threshold for the proportion of missing data or outliers); if the data quality of all second communication nodes participating in federated learning shows a significant decrease exceeding a preset threshold, it can be decided to terminate federated learning or switch the federated learning type to adapt to the new data quality characteristics.

[0135] In one embodiment, making decisions based on changes in computing resources includes making decisions based on the number of second communication nodes reporting insufficient computing power.

[0136] In some examples, if some of the second communication nodes report insufficient computing power (e.g., reduced CPU / GPU resources), the first communication node can decide to reduce the computational complexity of the training task (e.g., reduce the number of training rounds, reduce the model complexity at the first communication node, etc.). Alternatively, the first communication node can adjust its task allocation strategy to allow second communication nodes with sufficient computing power to undertake more computational tasks.

[0137] In some examples, if the number of reports of insufficient computing power in each of the second communication nodes exceeds a preset threshold, the first communication node may decide to switch to a federated learning type with lower computing resource consumption (such as switching from horizontal federated learning to vertical federated learning).

[0138] In some examples, if the number of second communication nodes reporting insufficient computing power exceeds a preset threshold, the first communication node may also decide to suspend federated learning training while waiting for the resources of the second communication nodes to be restored.

[0139] In one embodiment, making decisions based on changes in the federated learning time period / duration includes at least one of the following: making decisions based on the number of second communication nodes whose available time period has been shortened; and making decisions based on the state of satisfaction of training requirements after the available time period has been shortened.

[0140] In some examples, where the available time period of some second communication nodes is shortened but training needs can still be met, the first communication node can decide to adjust the training schedule to complete training within an appropriate time window.

[0141] In some examples, if the number of second communication nodes with shortened available time exceeds a preset threshold and cannot meet the training requirements of the current federated learning type, the first communication node may decide to reduce the training cycle or reduce the training tasks, or switch to a federated learning type that is more suitable for short-term training (such as switching from vertical federated learning to horizontal federated learning).

[0142] In one embodiment, making decisions based on the overall optimization of federated learning includes: making decisions on optimizing overall training efficiency or model accuracy based on the switching results.

[0143] In some examples, if the first communication node discovers that switching the federated learning type can improve the overall training efficiency or model accuracy of federated learning, it can proactively initiate the switching of the federated learning type.

[0144] In one embodiment, if the first communication node supports the target federated learning type, sending the target federated learning type to at least one second communication node includes: sending a federated learning switching notification to at least one second communication node.

[0145] In this embodiment, the federated learning switch notification can be specifically understood as a request to the second communication node participating in the current federated learning to switch the federated learning type.

[0146] In a specific example, if the first communication node supports the target federated learning type it has determined, it can be assumed that the first communication node can continue to participate in the federated learning corresponding to the target federated learning type. In this case, the first communication node sends a federated learning switch notification to each of the second communication nodes participating in the current federated learning.

[0147] In one embodiment, the federated learning switch notification includes at least one of the following: target federated learning type; indication to retain the existing training model; and new training model.

[0148] In some examples, the federated learning switch notification may include a target federated learning type to inform the second communication node which federated learning type to switch to; a retention instruction for the original training model to indicate whether the second communication node should retain the original training model if a federated learning type switch is required; and a new training model (which may include a model file or a link containing a model file) distributed to the second communication node if it does not retain the original training model.

[0149] New training model distribution is only possible when switching to the horizontal federated learning type.

[0150] In one embodiment, when the first communication node does not support the target federated learning type, sending the target federated learning type to at least one second communication node includes: determining a target first communication node that supports the target federated learning type through a network storage function; sending the target federated learning type, at least one second communication node information and / or first information to the target first communication node; and having the target first communication node send a federated learning switching notification to each of the second communication nodes.

[0151] In a specific example, if the first communication node does not support the determined target federated learning type, in order to enable the second communication nodes to switch federated learning types normally, a new first communication node that supports the target federated learning type will be determined through NRF, and this new first communication node will be designated as the target first communication node. The current first communication node will then send the target federated learning type, at least one previously acquired second communication node information, and / or the first information acquired by each second communication node to the target first communication node. The target first communication node will then send a federated learning switching notification to each second communication node.

[0152] In one embodiment, the federated learning switching notification includes at least one of the following: target federated learning type; indication to retain the original training model; new training model; first communication node switching indication and address information of the target first communication node; and applying new weights to the intermediate training results / models of each second communication node.

[0153] In one embodiment, after sending the target federated learning type to at least one second communication node, the method further includes receiving a handover notification response from each of the second communication nodes.

[0154] In this embodiment, the switching notification response can be specifically understood as response information indicating whether the second communication node has switched to the target federated learning type.

[0155] In a specific example, after the first communication node sends the target federated learning type to each of the second communication nodes, each of the second communication nodes will determine whether it can join the federated learning of the target federated learning type based on its own federated learning capabilities, and will send a switching notification response back to the first communication node based on the determined result.

[0156] In one embodiment, the switching notification response includes an indication of whether the second communication node is performing federated learning.

[0157] In one embodiment, the first communication node is a horizontal federated learning server or a vertical federated learning server in federated learning, and the second communication node is a horizontal federated learning client or a vertical federated learning client in federated learning.

[0158] In one exemplary embodiment, Figure 5 is a flowchart of a communication method provided in an embodiment of this application. This method can be applied to the situation where, during the federated learning training process of a 5G system, a federated learning type is selected for the communication nodes participating in the federated learning. This method can be executed by a communication device, which can be implemented by software and / or hardware and integrated on the communication node. This method can be applied to a second communication node.

[0159] As shown in Figure 5, the communication method provided in this application embodiment specifically includes S401-S402.

[0160] S401, Report the first information to the first communication node.

[0161] S402, Receive Federated Learning Switching Notification.

[0162] In one embodiment, the first information includes at least one of the following: changes in the data samples or feature information of the second communication node; changes in the local privacy protection requirements of the second communication node; changes in the data quality information of the second communication node; the second communication node being unable to return the model or intermediate training results within the longest response time specified by the first communication node; changes in the capabilities of the second communication node; and changes in the time period / duration during which the second communication node can perform federated learning.

[0163] In one embodiment, the federated learning switching notification includes at least one of the following: target federated learning type; indication to retain the original training model; new training model; first communication node switching indication and address information of the target first communication node; and applying new weights to the intermediate training results / models of each second communication node.

[0164] In one embodiment, after receiving the federated learning switch notification, the method further includes: sending a switch notification response to the first communication node or the target first communication node; wherein the switch notification response includes an indication of whether the second communication node performs federated learning.

[0165] In one embodiment, the second communication node is a horizontal federated learning client or a vertical federated learning client in federated learning, and the first communication node is a horizontal federated learning server or a vertical federated learning server in federated learning.

[0166] The communication method of this application is illustrated below through some exemplary schemes. In the following schemes, the federated learning server refers to the first communication node, and the federated learning client refers to the second communication node, as an example for illustration.

[0167] Solution 1: A specific example is provided for selecting the federated learning type for the federated learning client and federated learning server participating in the federated learning preparation process. Figure 6 is a timing example diagram of federated learning preparation provided by an embodiment of this application. As shown in Figure 6, it specifically includes:

[0168] S0. The federated learning server (such as NWDAF, AF) receives an analysis retrieval request or a model retrieval request, which includes an analysis identifier (ID).

[0169] In some examples, the analytics fetch request or model fetch request received by the federated learning server is used to trigger the federated learning server to prepare for federated learning.

[0170] S1. The federated learning server sends a discovery request to the NRF to discover federated learning clients.

[0171] In some examples, it was found that the request covered all types of federated learning (such as horizontal federated learning, vertical federated learning, or both).

[0172] S2.NRF returns one or more NWDAFs or AFs as federated learning clients.

[0173] S3. The Federated Learning Server sends a Client Information Reporting Request to the Federated Learning Client, requesting the Federated Learning Client to report client information in the Client Information Reporting Request.

[0174] Client information includes at least one of the following: data sample or feature information (such as sample identifiers, feature identifiers, total number of samples, feature dimensions, and time information corresponding to the sample or feature data); privacy protection requirements (such as whether it supports exchanging models, intermediate training parameters, and intermediate results during federated learning) and security capability information (such as supported encryption algorithm types and homomorphic encryption capabilities); data quality information (such as missing data information (such as the proportion of missing data values) and outlier information (such as the proportion of outliers)); computing resource information (such as computing power information (CPU / GPU specifications, computing power available for federated learning) and storage capacity available for federated learning); and the time period or duration during which horizontal or vertical federated learning can be performed.

[0175] S4. The Federated Learning Client reports client information.

[0176] S5. The federated learning server determines whether to conduct federated learning and the type of federated learning (such as horizontal federated learning, vertical federated learning, multi-layer horizontal federated learning, and multi-layer vertical federated learning) based on the client information reported by the federated learning client, the network function type of the federated learning client, and the analysis ID in S0.

[0177] The federated learning server filters and stores the determined list of federated learning clients and related information (such as the client information reported by the federated learning clients in the federated learning client list in S4).

[0178] In some examples, the decision logic of the federated learning server may include: if the number of overlapping data features in each client's information exceeds a preset data feature threshold, while the number of overlapping sample identifiers is less than or equal to a preset sample identifier threshold, then the federated learning type is determined to be horizontal federated learning; if the number of overlapping data features in each client's information is less than or equal to a preset data feature threshold, while the number of overlapping sample identifiers exceeds a preset sample identifier threshold, then the federated learning type is determined to be vertical federated learning; if the majority of each client's information supports model exchange but not intermediate result transfer during the federated learning process, then the federated learning type is determined to be horizontal federated learning; if the majority of each client's information does not support model disclosure but supports intermediate result transfer during the federated learning process, then the federated learning type is determined to be vertical federated learning; If most of the client information supports homomorphic encryption and the security requirements exceed a preset threshold, then the federated learning type is determined to be vertical federated learning; otherwise, the federated learning type is determined to be horizontal federated learning. The data quality information in each client information is analyzed. If the number of missing samples or features in each client information exceeds a preset missing threshold, then no federated learning is required. Since horizontal federated learning requires more computing resources, the applicable federated learning type for each second communication node can be determined based on the computing resource information in each client information. If no second communication node has sufficient computing resources, then no federated learning is required. The appropriate federated learning type can be determined based on the time period and duration in each client information where horizontal or vertical federated learning can be performed.

[0179] S6a. [If the federated learning server supports the federated learning type determined in S5] The federated learning server sends a request to the selected federated learning clients to participate in the federated learning, specifying the type of federated learning, the time period for the federated learning, and the duration of the federated learning.

[0180] S6b. [If the federated learning server does not support the federated learning type determined in S5] The source federated learning server discovers a target federated learning server that supports the federated learning type in S5 through NRF, and sends it the determined federated learning type, a list of federated learning client network functions (including network function IDs or other network function address information), and the relevant information received in step 3, and then executes 6c.

[0181] The source federated learning server is the federated learning server that does not support the federated learning type in S5 during the current federated learning process.

[0182] S6c. The target federated learning server sends a join request to the network functions in the federated learning client list, which includes the type of federated learning, the time period for federated learning, and the duration of federated learning.

[0183] S6a and S6b+6c are parallel schemes and will not be executed simultaneously.

[0184] S7. The Federated Learning Client sends an instruction to the Federated Learning Server whether to join the Federated Learning program.

[0185] Solution 2: A specific example is provided for dynamically selecting the federated learning type for each federated learning client participating in the federated learning process during federated learning training. Figure 7 is a timing example diagram of federated learning training provided by an embodiment of this application. As shown in Figure 7, it specifically includes:

[0186] S1. The Federated Learning Server subscribes to the NRF for information on the Federated Learning Capabilities of the Federated Learning Clients.

[0187] The federated learning server can subscribe to federated learning capability information of federated learning clients by sending subscription messages to the NRF. The subscription message may contain the NWDAF ID or AF ID.

[0188] In some examples, reports from the federated learning client can be triggered periodically by the federated learning client or requested by the federated learning server.

[0189] S2.NRF notifies the federated learning server that the federated learning client's federated learning capabilities have changed, or that the time period / duration for horizontal / vertical federated learning has changed.

[0190] S3. The Federated Learning Server subscribes to Federated Learning-related information from Federated Learning Clients.

[0191] S4. The Federated Learning client sends the following information to the Federated Learning server:

[0192] 1) Instructions and reasons for withdrawing from federated learning (insufficient computing power, federated learning is not supported at this time).

[0193] 2) Changes in data samples or feature information (increase in samples, decrease in samples, increase in features, decrease in features).

[0194] 3) Changes in local privacy protection requirements.

[0195] 4) Changes in data quality information (changes in missing data information (e.g., changes in the proportion of missing data values), changes in outlier information (e.g., changes in the proportion of outliers)).

[0196] 5) Changes in computing resource information (changes in computing power information (changes in CPU / GPU specifications, changes in computing power available for federated learning), changes in storage capacity available for federated learning).

[0197] 6) Unable to return intermediate training results within the maximum response time specified by the federated learning server.

[0198] The S5 Federated Learning Server may make the following decisions:

[0199] 1) Switch the federated learning type.

[0200] 2) Switch the federated learning type and the model training instructions; the model training instructions include: instructions to use the original training model or instructions to resend the new model for training.

[0201] 3) Apply new weights to the intermediate training results / models of each second communication node; where the intermediate training result weights correspond to vertical federated learning, and the model weights correspond to horizontal federated learning.

[0202] 4) Terminate or suspend the reasoning and / or training of federated learning.

[0203] 5) Find a new second communication node to join.

[0204] The internal decision-making logic of the federated learning server includes at least one of the following: making decisions based on changes in data characteristics and sample identifiers; making decisions based on changes in privacy protection requirements; making decisions based on changes in data quality; making decisions based on changes in computing resources; making decisions based on changes in the federated learning period / duration; and making decisions based on the overall optimization of federated learning.

[0205] In some examples, decisions based on changes in data features and sample identifiers may include: if the number of overlapping data features is greater than a preset data feature threshold, while the number of overlapping sample identifiers is less than or equal to a preset sample identifier threshold, the decision may be to switch to horizontal federated learning; if the number of overlapping data features is less than or equal to a preset data feature threshold, while the number of overlapping sample identifiers is greater than a preset sample identifier threshold, the decision may be to switch to vertical federated learning; if the distribution of data features or sample identifiers changes significantly (e.g., due to data updates causing a significant increase or decrease in the proportion of overlapping sample identifiers, a significant change can be determined by setting a preset sample identifier threshold), the federated learning type may be re-evaluated, and if the evaluation result differs from the ongoing federated learning type, the federated learning type may be switched; the federated learning may be terminated; or a new federated learning client may be found to join.

[0206] In some examples, decisions based on changes in privacy requirements may include: if a majority of federated learning clients support model exchange but not intermediate result delivery during the federated learning process, the decision may be to switch to horizontal federated learning; if a majority of federated learning clients do not support model disclosure but support intermediate result delivery during the federated learning process, the decision may be to switch to vertical federated learning; if a majority of federated learning clients have increased privacy requirements (such as requiring stronger homomorphic encryption), the decision may be to switch to vertical federated learning to reduce the risk of data leakage; terminate federated learning; or find new federated learning clients to join.

[0207] In some examples, decisions based on changes in data quality may include: if the increase in the proportion of missing data or outliers reported by the federated learning clients exceeds a preset threshold, it may lead to a decline in training quality. Based on this, possible decisions include: adjusting the weights of federated learning clients to reduce the impact of clients with high outlier rates in federated learning; pausing or removing some federated learning clients (if the quality problems of the removed clients are severe, the severity can be determined by setting a threshold for the proportion of missing data or outliers); if the data quality of all participating federated learning clients shows a significant decline exceeding a preset threshold, a decision can be made to terminate federated learning or switch the federated learning type to adapt to the new data quality characteristics; or finding new federated learning clients to join.

[0208] In some examples, decisions based on changes in computing resources may include: if some federated learning clients report insufficient computing power (e.g., reduced CPU / GPU resources), the federated learning server may decide to reduce the computational complexity of the training tasks (e.g., reduce the number of training epochs, reduce the model complexity at the federated learning server). The federated learning server may also adjust its task allocation strategy to allow federated learning clients with sufficient computing power to undertake more computational tasks; if the number of federated learning clients reporting insufficient computing power exceeds a preset threshold, the federated learning server may decide to switch to a federated learning type with lower computational resource consumption (e.g., switch from horizontal federated learning to vertical federated learning); if the number of second communication nodes reporting insufficient computing power exceeds a preset threshold, the federated learning server may also decide to pause federated learning training to wait for the resources of the federated learning clients to recover; and find new federated learning clients to join to supplement computing resources.

[0209] In some examples, decisions based on changes in the federated learning time period / duration may include: if the available time period of some federated learning clients is shortened but the training requirements can still be met, the federated learning server may decide to adjust the training schedule to complete the training within a suitable time window; if the number of federated learning clients with shortened available time exceeds a preset threshold and cannot meet the training requirements of the current federated learning type, the federated learning server may decide to reduce the training cycle or reduce the number of training tasks, or switch to a federated learning type that is more suitable for short-term training (such as switching from vertical federated learning to horizontal federated learning).

[0210] In some examples, the overall optimization of federated learning can be decided by proactively initiating a switch in the federated learning type if the federated learning server finds that switching the federated learning type can improve the overall training efficiency or model accuracy of federated learning.

[0211] S6. [If the federated learning server does not support the federated learning type determined in S5] The source federated learning server discovers a target federated learning server through NRF and submits the federated learning type to be switched, the list of federated learning clients, and the relevant information updated by the federated learning clients in step 3 or the relevant information reported in the federated learning preparation phase to the target federated learning server.

[0212] S7. [If the federated learning server does not support the federated learning type determined in S5] The source federated learning server or the target federated learning server may send the following information to the federated learning client (including newly discovered federated learning clients):

[0213] 1) Switch to which type of federated learning? If switching, instruct the client whether to retain the original trained model. If not, distribute a new trained model (model file or a link containing the model file). Note that a new model can only be distributed if switching to lateral federated learning.

[0214] 2) Federated learning server switch notification, including the address information of the new federated learning server.

[0215] 3) Apply new weights to the intermediate training results / models of the federated learning client.

[0216] S8. The federated learning client sends a switch notification response to the source federated learning server or the target federated learning server, which includes an instruction on whether to continue federated learning.

[0217] Solution 3: A specific example is provided whereby, before federated learning, an NWDAF / AF, acting as a federated learning client and / or server, reports its own federated learning capabilities. Figure 8 is a timing example diagram of federated learning capability reporting provided by an embodiment of this application. As shown in Figure 8, it may specifically include:

[0218] S1.NWDAF and / or AF register or update their federated learning capabilities with NRF, as well as the time periods during which they support various types of federated learning.

[0219] Federated learning capabilities include: whether it supports being used as a horizontal federated learning server, whether it supports being used as a vertical federated learning server, whether it supports being used as a horizontal federated learning client, and whether it supports being used as a vertical federated learning client.

[0220] S2.NRF stores the aforementioned capabilities.

[0221] S3.NRF returns a registration or update instruction.

[0222] In one exemplary embodiment, FIG9 is a schematic diagram of a communication device provided in an embodiment of this application. The communication device is applied to a first communication node. As shown in FIG9, the device includes: a reporting request sending module 510, configured to send a client information reporting request to at least one second communication node; and a client information receiving module 520, configured to receive client information reported by each second communication node in response to the client information reporting request.

[0223] The first type of sending module 530 is configured to determine the federated learning type based on the information of each client and send the federated learning type to each second communication node.

[0224] In one embodiment, the client information includes at least one of the following: data sample or feature information; privacy protection requirements and security capability information; data quality information; computing resource information; and a time period or duration during which horizontal or vertical federated learning can be performed.

[0225] In one embodiment, determining the federated learning type based on each client's information includes at least one of the following: determining the federated learning type based on the number of overlapping data features and the number of overlapping sample identifiers in each client's information; determining the federated learning type based on the model exchange support capability and intermediate result transmission capability in each client's information; determining the federated learning type based on the homomorphic encryption support capability and security requirements in each client's information; determining the federated learning type based on the data quality information in each client's information; determining the federated learning type based on the computing resource information in each client's information; and determining the federated learning type based on the time period or duration in each client's information where horizontal or vertical federated learning can be performed.

[0226] In one embodiment, before sending a client information reporting request to at least one second communication node, the method further includes: sending a discovery request to the network storage function; wherein the discovery request includes horizontal federated learning client capabilities and / or vertical federated learning client capabilities; and receiving feedback from at least one second communication node from the network storage function.

[0227] In one embodiment, when the first communication node supports the federated learning type, sending the federated learning type to each of the second communication nodes includes sending a request to participate in federated learning to each of the second communication nodes.

[0228] In one embodiment, when the first communication node does not support the federated learning type, sending the federated learning type to each of the second communication nodes includes: determining a target first communication node that supports the federated learning type through a network storage function; sending the federated learning type, at least one second communication node information, and / or each client information to the target first communication node; and having the target first communication node send a request to each of the second communication nodes to participate in the federated learning.

[0229] In one embodiment, participating in a federated learning request includes at least one of the following: the type of federated learning; the time period for which the federated learning will take place; and the duration for which the federated learning will take place.

[0230] In one embodiment, after sending the federated learning type to each of the second communication nodes, the method further includes: receiving a joining federated learning instruction from each of the second communication nodes.

[0231] In one embodiment, the first communication node is a horizontal federated learning server or a vertical federated learning server in federated learning, and the second communication node is a horizontal federated learning client or a vertical federated learning client in federated learning.

[0232] In one exemplary embodiment, FIG10 is a schematic diagram of a communication device provided in an embodiment of this application. The communication device is applied to a second communication node. As shown in FIG10, the device includes: a first information reporting module 610, configured to receive a client information reporting request sent by the first communication node and report client information to the first communication node; and a first type receiving module 620, configured to receive federated learning types sent by the first communication node.

[0233] In one embodiment, the client information includes at least one of the following: data sample or feature information; privacy protection requirements and security capability information; data quality information; computing resource information; and a time period or duration during which horizontal or vertical federated learning can be performed.

[0234] In one embodiment, the first type receiving module 620 is specifically configured to receive a request to participate in federated learning sent by a first communication node.

[0235] In one embodiment, participating in a federated learning request includes at least one of the following: the type of federated learning; the time period for which the federated learning will take place; and the duration for which the federated learning will take place.

[0236] In one embodiment, after receiving the federated learning type sent by the first communication node, the method further includes: sending a feedback instruction to the first communication node to join the federated learning.

[0237] In one embodiment, the first communication node is a horizontal federated learning server or a vertical federated learning server in federated learning, and the second communication node is a horizontal federated learning client or a vertical federated learning client in federated learning.

[0238] In one exemplary embodiment, FIG11 is a schematic diagram of the structure of a communication device provided in an embodiment of the present application. The communication device is applied to a first communication node. As shown in FIG11, the device includes: a decision module 710, configured to receive first information and make a decision based on the first information.

[0239] The second type of sending module 720 is configured to determine the target federated learning type based on the decision result and send the target federated learning type to at least one second communication node.

[0240] In one embodiment, the first information is provided by a network storage function and / or at least one second communication node.

[0241] In one embodiment, before receiving the first information, at least one of the following is included: subscribing to the first information from a network storage function; subscribing to the first information from at least one second communication node; and at least one second communication node periodically sending the first information.

[0242] In one embodiment, the first information includes at least one of the following: changes in the data samples or feature information of the second communication node; changes in the local privacy protection requirements of the second communication node; changes in the data quality information of the second communication node; the second communication node being unable to return the model or intermediate training results within the longest response time specified by the first communication node; changes in the capabilities of the second communication node; and changes in the time period / duration during which the second communication node can perform federated learning.

[0243] In one embodiment, the decision result includes at least one of the following: switching the federated learning type; switching the federated learning type and model training instructions; wherein the model training instructions include: instructions to use the original training model or instructions to resend the new model for training; applying new weights to the intermediate training results / models of each second communication node; terminating or suspending the inference and / or training of the federated learning; and finding a new second communication node to join.

[0244] In one embodiment, making a decision based on the first information includes at least one of the following: making a decision based on changes in data characteristics and sample identifiers; making a decision based on changes in privacy protection requirements; making a decision based on changes in data quality; making a decision based on changes in computing resources; making a decision based on changes in the federated learning period / duration; and making a decision based on the overall optimization of federated learning.

[0245] In one embodiment, making decisions based on changes in data features and sample identifiers includes at least one of the following: making decisions based on the amount of overlap between data features and sample identifiers; making decisions based on changes in the distribution of data features or sample identifiers.

[0246] In one embodiment, making decisions based on changes in privacy protection requirements includes at least one of the following: making decisions based on model exchange support capabilities and intermediate training result transfer capabilities; making decisions based on the amount of increase in privacy protection requirements.

[0247] In one embodiment, making decisions based on changes in data quality includes making decisions based on the proportion of missing data or the proportion of outliers.

[0248] In one embodiment, making decisions based on changes in computing resources includes making decisions based on the number of second communication nodes reporting insufficient computing power.

[0249] In one embodiment, making decisions based on changes in the federated learning time period / duration includes at least one of the following: making decisions based on the number of second communication nodes whose available time period has been shortened; and making decisions based on the state of satisfaction of training requirements after the available time period has been shortened.

[0250] In one embodiment, making decisions based on the overall optimization of federated learning includes: making decisions on optimizing overall training efficiency or model accuracy based on the switching results.

[0251] In one embodiment, if the first communication node supports the target federated learning type, sending the target federated learning type to at least one second communication node includes: sending a federated learning switching notification to at least one second communication node.

[0252] In one embodiment, the federated learning switch notification includes at least one of the following: target federated learning type; indication to retain the existing training model; and new training model.

[0253] In one embodiment, when the first communication node does not support the target federated learning type, sending the target federated learning type to at least one second communication node includes: determining a target first communication node that supports the target federated learning type through a network storage function; sending the target federated learning type, at least one second communication node information and / or first information to the target first communication node; and having the target first communication node send a federated learning switching notification to each of the second communication nodes.

[0254] In one embodiment, the federated learning switching notification includes at least one of the following: target federated learning type; indication to retain the original training model; new training model; first communication node switching indication and address information of the target first communication node; and applying new weights to the intermediate training results / models of each second communication node.

[0255] In one embodiment, after sending the target federated learning type to at least one second communication node, the method further includes receiving a handover notification response from each of the second communication nodes.

[0256] In one embodiment, the first communication node is a horizontal federated learning server or a vertical federated learning server in federated learning, and the second communication node is a horizontal federated learning client or a vertical federated learning client in federated learning.

[0257] In one exemplary embodiment, FIG12 is a schematic diagram of a communication device provided in an embodiment of this application. The communication device is applied to a second communication node. As shown in FIG12, the device includes: a second information reporting module 810, configured to report first information to a first communication node; and a handover notification receiving module 820, configured to receive federated learning handover notifications.

[0258] In one embodiment, the first information includes at least one of the following: changes in the data samples or feature information of the second communication node; changes in the local privacy protection requirements of the second communication node; changes in the data quality information of the second communication node; the second communication node being unable to return the model or intermediate training results within the longest response time specified by the first communication node; changes in the capabilities of the second communication node; and changes in the time period / duration during which the second communication node can perform federated learning.

[0259] In one embodiment, the federated learning switching notification includes at least one of the following: target federated learning type; indication to retain the original training model; new training model; first communication node switching indication and address information of the target first communication node; and applying new weights to the intermediate training results / models of each second communication node.

[0260] In one embodiment, after receiving the federated learning switch notification, the method further includes: sending a switch notification response to the first communication node or the target first communication node; wherein the switch notification response includes an indication of whether the second communication node performs federated learning.

[0261] In one embodiment, the second communication node is a horizontal federated learning client or a vertical federated learning client in federated learning, and the first communication node is a horizontal federated learning server or a vertical federated learning server in federated learning.

[0262] This application embodiment also provides a communication node. Figure 13 is a structural schematic diagram of a communication node provided in this application embodiment. As shown in Figure 13, the communication node provided in this application embodiment includes a memory 920, a processor 910, and a computer program stored in the memory and executable on the processor. When the processor 910 executes the program, it implements the above-mentioned communication method.

[0263] The communication node may also include a memory 920; the processor 910 in the communication node may be one or more, with one processor 910 as an example in Figure 13; the memory 920 is used to store one or more programs; the one or more programs are executed by the one or more processors 910, so that the one or more processors 910 implement the communication method as described in the embodiments of this application.

[0264] The communication node also includes: a communication device 930, an input device 940, and an output device 950.

[0265] The processor 910, memory 920, communication device 930, input device 940 and output device 950 in the communication node can be connected by a bus or other means. Figure 13 shows an example of connection by bus.

[0266] Input device 940 can be used to receive input digital or character information, and to generate key signal inputs related to user settings and function control of the communication node. Output device 950 may include display devices such as a display screen.

[0267] The communication device 930 may include a receiver and a transmitter. The communication device 930 is configured to perform information transmission and reception communication under the control of the processor 910.

[0268] The memory 920, as a computer-readable storage medium, can be configured to store software programs, computer-executable programs, and modules, such as program instructions / modules corresponding to the communication method described in the embodiments of this application (e.g., a reporting request sending module 510, a client information receiving module 520, a first type sending module 530; or a first information reporting module 610, a first type receiving module 620; or a decision module 710, a second type sending module 720; or a second information reporting module 810, a switching notification receiving module 820). The memory 920 may include a program storage area and a data storage area, wherein the program storage area may store the operating system, an application program required for at least one function, and the data storage area may store data created according to the use of the communication node, etc. In addition, the memory 920 may include high-speed random access memory and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some instances, the memory 920 may further include memory remotely located relative to the processor 910, and these remote memories can be connected to the communication node via a network. Examples of the aforementioned networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0269] This application also provides a storage medium storing a computer program, which, when executed by a processor, implements any of the communication methods described in this application.

[0270] Optionally, the communication method, applied to a first communication node, includes: sending a client information reporting request to at least one second communication node; receiving client information reported by each second communication node in response to the client information reporting request; determining the federated learning type based on the client information; and sending the federated learning type to each second communication node.

[0271] Optionally, the communication method, applied to the second communication node, includes: receiving a client information reporting request sent by the first communication node and reporting client information to the first communication node; and receiving a federated learning type sent by the first communication node.

[0272] Optionally, the communication method, applied to a first communication node, includes: receiving first information and making a decision based on the first information; determining a target federated learning type based on the decision result, and sending the target federated learning type to at least one second communication node.

[0273] Optionally, this communication method, applied to the second communication node, includes: reporting first information to the first communication node; and receiving a federated learning switch notification.

[0274] The computer storage medium in this application embodiment can be any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. For example, a computer-readable storage medium can be, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical fiber, portable CD-ROM, optical storage device, magnetic storage device, or any suitable combination thereof. The computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0275] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, which can send, propagate, or transmit programs for use by or in connection with an instruction execution system, apparatus, or device.

[0276] Program code contained on a computer-readable medium may be transmitted using any suitable medium, including but not limited to: wireless, wire, optical fiber, radio frequency (RF), etc., or any suitable combination thereof.

[0277] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as "C" or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0278] Optionally, embodiments of this application also provide a computer program product, including a computer program that, when executed by a processor, implements the communication method provided in any embodiment of this application.

[0279] The above description is merely an exemplary embodiment of this application and is not intended to limit the scope of protection of this application.

[0280] Those skilled in the art will understand that the term user terminal encompasses any suitable type of wireless user equipment, such as mobile phones, portable data processing devices, portable web browsers, or vehicle-mounted mobile stations.

[0281] Generally, the various embodiments of this application can be implemented in hardware or dedicated circuitry, software, logic, or any combination thereof. For example, some aspects can be implemented in hardware, while others can be implemented in firmware or software that can be executed by a controller, microprocessor, or other computing device, although this application is not limited thereto.

[0282] Embodiments of this application can be implemented by executing computer program instructions through the data processor of a mobile device, for example, in a processor entity, or through hardware, or through a combination of software and hardware. The computer program instructions can be assembly instructions, Instruction Set Architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, status setting data, or source code or object code written in any combination of one or more programming languages.

[0283] Any block diagram of logical flow in the accompanying drawings of this application may represent program operations, or may represent interconnected logic circuits, modules, and functions, or may represent a combination of program operations and logic circuits, modules, and functions. The computer program may be stored in memory. The memory may be of any type suitable to the local technical environment and may be implemented using any suitable data storage technology, such as, but not limited to, read-only memory (ROM), random access memory (RAM), optical storage devices and systems (Digital Video Disc (DVD) or CompactDisk (CD), etc.). Computer-readable media may include non-transitory storage media. The data processor may be of any type suitable to the local technical environment, such as, but not limited to, general-purpose computers, special-purpose computers, microprocessors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), and processors based on multi-core processor architectures.

Claims

1. A communication method, applied to a first communication node, comprising: Send a client information reporting request to at least one second communication node; Receive client information reported by the at least one second communication node in response to the client information reporting request; The federated learning type is determined based on the client information, and the federated learning type is sent to the at least one second communication node.

2. The communication method according to claim 1, wherein, The client information includes at least one of the following: Data samples or feature information; Privacy protection requirements and security capabilities information; Data quality information; Computational resource information; The time period or duration during which horizontal or vertical federated learning can be conducted.

3. The communication method according to claim 1, wherein, The determination of the federated learning type based on the client information includes at least one of the following: The federated learning type is determined based on the number of overlapping data features and the number of overlapping sample identifiers in the client information. The federated learning type is determined based on the model exchange support capability and intermediate result transmission capability in the client information. The federated learning type is determined based on the homomorphic encryption support capabilities and security requirements in the client information; The federated learning type is determined based on the data quality information in the client information; The federated learning type is determined based on the computing resource information in the client information; The type of federated learning is determined based on the time period or duration in the client information that allows for horizontal or vertical federated learning.

4. The communication method according to claim 1, further comprising, before sending the client information reporting request to at least one second communication node: Send a discovery request to the network storage function; wherein the discovery request includes at least one of the horizontal federated learning client capabilities and the vertical federated learning client capabilities; The at least one second communication node receives feedback from the network storage function.

5. The communication method according to claim 1, wherein, If the first communication node supports the federated learning type, sending the federated learning type to each of the second communication nodes includes: Send a request to participate in federated learning to each of the second communication nodes.

6. The communication method according to claim 1, wherein, If the first communication node does not support the federated learning type, sending the federated learning type to the at least one second communication node includes: A target first communication node that supports the federated learning type is identified through network storage functionality; Send at least one of the federated learning type, at least one second communication node information, and the client information to the target first communication node; The target first communication node sends a request to participate in federated learning to the at least one second communication node.

7. The communication method according to claim 5 or 6, wherein, The request to participate in federated learning includes at least one of the following: The type of federated learning; The time period for federal learning; The duration of federal learning.

8. The communication method according to claim 1, further comprising, after sending the federated learning type to the at least one second communication node: Receive the joining federated learning instruction from at least one of the second communication nodes.

9. The communication method according to any one of claims 1-6 and 8, wherein, The first communication node is a horizontal federated learning server or a vertical federated learning server in federated learning, and the second communication node is a horizontal federated learning client or a vertical federated learning client in federated learning.

10. A communication method applied to a second communication node, comprising: Receive a client information reporting request sent by the first communication node, and report the client information to the first communication node; Receive the federated learning type sent by the first communication node.

11. The communication method according to claim 10, wherein, The client information includes at least one of the following: Data samples or feature information; Privacy protection requirements and security capabilities information; Data quality information; Computational resource information; The time period or duration during which horizontal or vertical federated learning can be conducted.

12. The communication method according to claim 10, wherein, The type of federated learning received from the first communication node includes: Receive the request to participate in federated learning sent by the first communication node.

13. The communication method according to claim 12, wherein, The request to participate in federated learning includes at least one of the following: The type of federated learning; The time period for federal learning; The duration of federal learning.

14. The communication method according to claim 10, further comprising, after receiving the federated learning type sent by the first communication node: The instruction to join the federated learning is sent back to the first communication node.

15. The communication method according to any one of claims 10-14, wherein, The first communication node is a horizontal federated learning server or a vertical federated learning server in federated learning, and the second communication node is a horizontal federated learning client or a vertical federated learning client in federated learning.

16. A communication method applied to a first communication node, comprising: Receive the first information and make a decision based on the first information; The target federated learning type is determined based on the decision result, and the target federated learning type is sent to at least one second communication node.

17. The communication method according to claim 16, wherein, The first information is provided by at least one of the network storage function and the at least one second communication node.

18. The communication method according to claim 17, further comprising at least one of the following before receiving the first information: Subscribe to the first information by the network storage function; Subscribe to the first information from the at least one second communication node; The at least one second communication node sends the first information at regular intervals.

19. The communication method according to claim 16, wherein, The first information includes at least one of the following: Changes in the data samples or feature information of the second communication node; The local privacy protection requirements of the second communication node have changed; Changes in the data quality information of the second communication node; The second communication node is unable to return the model or intermediate training results within the maximum response time specified by the first communication node; The capabilities of the second communication node have changed; The second communication node can perform federated learning time period / duration changes.

20. The communication method according to claim 16, wherein, The decision outcome includes at least one of the following: Switch the federated learning type; Switch the federated learning type and the model training instructions; wherein, the model training instructions include: instructions to use the original training model or instructions to resend the new model for training; Apply new weights to the intermediate training results / model of the at least one second communication node; Terminate or suspend the reasoning and / or training of federated learning; Search for a new second communication node to join.

21. The communication method according to claim 16, wherein, The decision-making based on the first information includes at least one of the following: Make decisions based on changes in data characteristics and sample identifiers; Make decisions based on changes in privacy protection requirements; Make decisions based on changes in data quality; Make decisions based on changes in computing resources; Decisions are made based on changes in federal learning time periods / durations; Decisions are made based on the overall optimization of federated learning.

22. The communication method according to claim 21, wherein, The decision-making based on changes in data characteristics and sample identifiers includes at least one of the following: Decisions are made based on the amount of overlap between the data features and the sample identifiers; Decisions are made based on the distribution changes of the data characteristics or the sample identifiers.

23. The communication method according to claim 21, wherein, The decision-making process based on changes in privacy protection requirements includes at least one of the following: Decisions are made based on the model's ability to exchange support and the ability to transfer intermediate training results. Decisions are made based on the increased level of privacy protection requirements.

24. The communication method according to claim 21, wherein, The decision-making based on changes in data quality includes: Make decisions based on the proportion of missing data or outliers.

25. The communication method according to claim 21, wherein, The decision-making based on changes in computing resources includes: Decisions are made based on the number of second communication nodes with insufficient computing power reported.

26. The communication method according to claim 21, wherein, The decision-making based on changes in the federal learning time period / duration includes at least one of the following: Decisions are made based on the number of second communication nodes with shortened available time periods; Decisions are made based on the degree to which training needs are met after the available time period is shortened.

27. The communication method according to claim 21, wherein, The decision-making based on the overall optimization of federated learning includes: Decisions are made on optimizing overall training efficiency or model accuracy based on the switching results.

28. The communication method according to claim 16, wherein, If the first communication node supports the target federated learning type, sending the target federated learning type to at least one second communication node includes: Send a federated learning switch notification to the at least one second communication node.

29. The communication method according to claim 28, wherein, The federated learning switch notification includes at least one of the following: The target federated learning type; Instructions to retain existing trained models; A new training model.

30. The communication method according to claim 16, wherein, If the first communication node does not support the target federated learning type, sending the target federated learning type to at least one second communication node includes: A target first communication node that supports the target federated learning type is identified through network storage functionality; Send at least one of the target federated learning type, at least one second communication node information, and the first information to the target first communication node; The target first communication node sends a federated learning switch notification to the at least one second communication node.

31. The communication method according to claim 30, wherein, The federated learning switch notification includes at least one of the following: The target federated learning type; Instructions to retain existing trained models; New training model; The first communication node switching indication, and the address information of the target first communication node; Apply new weights to the intermediate training results / model of the at least one second communication node.

32. The communication method according to any one of claims 16-31, further comprising, after sending the target federated learning type to at least one second communication node: Receive the handover notification response from the at least one second communication node.

33. The communication method according to claim 32, wherein, The switching notification response includes an indication of whether the second communication node should perform federated learning.

34. The communication method according to any one of claims 16-31, wherein, The first communication node is a horizontal federated learning server or a vertical federated learning server in federated learning, and the second communication node is a horizontal federated learning client or a vertical federated learning client in federated learning.

35. A communication node, comprising: The program includes a memory, a processor, a program stored in the memory and executable on the processor, and a data bus for implementing communication between the processor and the memory, wherein the program, when executed by the processor, implements the communication method as described in any one of claims 1-34.

36. A storage medium for computer-readable storage, the storage medium storing at least one program that can be executed by at least one processor to implement the communication method as described in any one of claims 1-34.

37. A computer program product comprising a computer program that, when executed by a processor, implements the communication method as described in any one of claims 1-34.

Images