Terminal configuration method, terminal, network equipment, system and storage medium
By providing distributed node configuration methods for terminals in wireless communication networks, the lack of distributed model training configuration on the RAN side is solved, efficient control and stability of distributed model training is achieved, data transmission pressure is reduced, and the Data security.
Patent Information
- Application Number
- CN202310656930.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-06-05
- Publication Date
- 2024-12-06
AI Technical Summary
In wireless communication networks, the current RAN side does not support the configuration of distributed model training, which leads to difficulty in monitoring and maintaining performance of distributed model training, affecting the efficiency and reliability of distributed model training.
提出一种终端的配置方法,通过在终端的RRC状态为非激活或连接态时接收配置信息,进行分布式节点的激活、去激活、添加或删除操作,提高分布式模型训练的控制能力,并通过中心节点对分布式节点进行配置和模型参数更新,实现分布式模型训练的高效进行。
It improves the control capability and reliability of distributed model training, reduces data transmission pressure, enhances data security, and improves the training efficiency and stability of distributed model.
Smart Images

Figure CN119095090A_ABST
Abstract
Description
Technical Field
[0001] The present disclosure relates to the field of wireless communication technology, and in particular to a terminal configuration method, a terminal, a network device, a system and a storage medium. Background Art
[0002] With the development of wireless communication technology, more and more difficulties are faced due to the increasing complexity of communication scenarios. Introducing machine learning into wireless communication technology provides the possibility of solving problems more efficiently.
[0003] Implementing wireless machine learning model training through a distributed learning framework can reduce transmission overhead, reduce the computing pressure on a single device, and improve machine learning efficiency. Summary of the invention
[0004] One purpose of the present disclosure is to provide a configuration scheme for supporting distributed model training on the RAN (Radio Access Network) side, so as to improve the control capability of distributed model training in a wireless communication network.
[0005] According to one aspect of some embodiments of the present disclosure, a configuration method of a terminal is proposed, including: when the RRC (Radio Resource Control) state of the terminal is inactive, the terminal receives first configuration information or second configuration information; and when the RRC state of the terminal is connected, the terminal receives first configuration information, second configuration information or third configuration information, wherein: the first configuration information is used to instruct the terminal to deactivate model training as a distributed node in a training group of a distributed model; the second configuration information is used to instruct the terminal to delete model training as a distributed node in a training group of a distributed model; and the third configuration information is used to instruct the terminal to activate model training as a distributed node in the training group.
[0006] In some embodiments, the first configuration information, the second configuration information and the third configuration information are located in the group configuration information of the training group, and the configuration method also includes: the terminal performs at least one of establishment, update, addition, deletion, activation or deactivation of the distributed node according to the group configuration information.
[0007] In some embodiments, the group configuration information meets at least one of the following conditions: the group configuration information is carried through at least one of DRB (Data Radio Bearer), SRB (Signaling Radio Bearer) or IRB (Intelligent Radio Bearer); the group configuration information is received through at least one of the user plane, control plane, data plane and intelligent plane; or the group configuration information is carried through at least one of system messages, RRC signaling, MAC (Medium Access Control) CE (Control Element) signaling, DCI (Downlink Control Information) signaling and NAS (Non-access stratum) messages.
[0008] In some embodiments, the group configuration information includes at least one of a group identifier, a model identifier, or a central node identifier.
[0009] In some embodiments, the group configuration information also includes an identification of at least one of the nodes, models, or functions whose status has been changed.
[0010] In some embodiments, the group configuration information conforms to at least one of the following: the identification of the node whose state is changed includes at least one of the identification of a deleted distributed node, a list of deleted distributed nodes, the identification of an added distributed node, a list of added distributed nodes, the identification of an activated distributed node, a list of activated distributed nodes, the identification of a deactivated distributed node, or a list of deactivated distributed nodes; the identification of the model whose state is changed includes at least one of the identification of an activated model, a list of activated models, the identification of a deactivated model, or a list of deactivated models; or the identification of the function whose state is changed includes at least one of the identification of an activated function, a list of activated functions, the identification of a deactivated function, or a list of deactivated functions.
[0011] In some embodiments, the group identifier includes a first bit and a second bit, the first bit is a unique identifier of the distributed model training type, and the second bit is a unique identifier of the training group in the same distributed model training type; or the group identifier is a unique identifier of the training group within the same TA (Tracking Area), PLMN (Public Land Mobile Network), 5G base station gNB or CU (Centralized Unit).
[0012] In some embodiments, the configuration method further includes: the terminal receives transmission indication information, and the transmission indication information includes node and path information for uploading, converging or sending information.
[0013] In some embodiments, the central node of the training group is another user terminal, and the transmission indication information includes an identifier of the central node or interface information between terminals.
[0014] In some embodiments, the central node of the training group is a base station, and the transmission indication information includes a base station identifier.
[0015] In some embodiments, the central node of the training group is located in a TA or a base station group, and the distributed nodes include service users of the central node and service users of at least one first base station in the TA or the base station group, and the transmission indication information also includes at least one item of an identifier of the first base station, a link identifier between the central node and the first base station, or routing information between the central node and the first base station.
[0016] In some embodiments, the configuration method also includes at least one of the following: the terminal performs model reasoning based on the local model; the terminal receives model parameter update information from the central node of the training group; or the terminal sends model training result information to the central node of the training group.
[0017] In some embodiments, the configuration method further includes: the terminal configuring the current state according to any one of the first configuration information, the second configuration information or the third configuration information.
[0018] In some embodiments, the terminal suspends training of the model belonging to the training group according to the first configuration information; the terminal exits the training group according to the second configuration information; or the terminal starts model training of the training group according to the third configuration information.
[0019] In some embodiments, the group configuration information carries at least one of the first configuration information, the second configuration information, or the third configuration information through the identifier of at least one of the nodes, models, or functions whose states have been changed; the configuration method also includes: when the identifier of the terminal matches the identifier of the deleted distributed node, or is in the list of the deleted distributed nodes, exiting the training group corresponding to the group identifier in the group configuration information; when the identifier of the terminal matches the identifier of the added distributed node or is in the list of the added distributed nodes, joining the training group corresponding to the group identifier in the group configuration information; when the identifier of the terminal matches the identifier of the activated distributed node or is in the list of the activated distributed nodes Under the following conditions, start model training of the training group corresponding to the group identifier in the group configuration information; when the terminal identifier matches the identifier of the deactivated distributed node or is in the list of deactivated distributed nodes, suspend model training of the training group corresponding to the group identifier in the group configuration information; start training the model corresponding to the identifier of the activated model or the list of activated models in the group configuration information; suspend training the model corresponding to the identifier of the deactivated model or the list of deactivated models in the group configuration information; start training the model corresponding to the identifier of the activated function or the list of activated functions in the group configuration information; or suspend training the model corresponding to the identifier of the deactivated function or the list of deactivated functions in the group configuration information.
[0020] In some embodiments, the configuration method also includes at least one of the following: the terminal receives group activation indication information, wherein the group activation indication information is used to indicate the start of model training belonging to the training group corresponding to the group identifier, and the group activation indication information includes the group identifier; or the terminal receives group deactivation indication information, wherein the group deactivation indication information is used to indicate the suspension of model training belonging to the training group corresponding to the group identifier, and the group deactivation indication information includes the group identifier.
[0021] In some embodiments, the distributed model includes a federated learning model.
[0022] According to one aspect of some embodiments of the present disclosure, a configuration method for a terminal is proposed, including: when the RRC state of the terminal is inactive, the central node of the training group of the distributed model sends first configuration information or second configuration information to the terminal; when the RRC state of the terminal is active, the central node of the training group of the distributed model sends first configuration information, second configuration information or third configuration information to the terminal, wherein: the first configuration information is used to instruct the terminal to deactivate the model training as a distributed node in the training group; the second configuration information is used to instruct the terminal to delete the model training as a distributed node in the training group; and the third configuration information is used to instruct the terminal to activate the model training as a distributed node in the training group.
[0023] In some embodiments, the first configuration information, the second configuration information and the third configuration information are located in the group configuration information of the training group; the configuration method also includes: the central node generates the group configuration information according to at least one of the establishment, update, addition, deletion, activation or deactivation of the distributed node.
[0024] In some embodiments, the configuration method further includes: the central node sends transmission indication information to the distributed nodes, and the transmission indication information includes node and path information for information uploading, aggregation or sending.
[0025] In some embodiments, the configuration method further includes at least one of the following: the central node sends model parameter update information to the distributed nodes of the training group; or the central node receives model training result information from the distributed nodes of the training group.
[0026] In some embodiments, the configuration method complies with at least one of the following: the distributed model includes a federated learning model; or the central node is at least one of a 5G base station gNB, an OAM (Operation Administration and Maintenance) entity, a LMF (Location Management Function) entity, or a UE (User Equipment).
[0027] According to one aspect of some embodiments of the present disclosure, a terminal is proposed, comprising: an information receiving unit, configured to receive first configuration information or second configuration information when the wireless resource control state of the terminal is inactive; and, when the wireless resource control state of the terminal is in a connected state, receive first configuration information, second configuration information or third configuration information, wherein: the first configuration information is used to instruct the terminal to perform a deactivation or deletion operation of model training as a distributed node in a training group of a distributed model; the second configuration information is used to instruct the terminal to perform a deactivation or deletion operation of model training as a distributed node in a training group of a distributed model; and the third configuration information is used to instruct the terminal to perform activation of model training as a distributed node in the training group.
[0028] In some embodiments, the terminal also includes at least one of the following: a configuration unit, configured to perform at least one of establishment, update, addition, deletion, activation or deactivation of distributed nodes according to the group configuration information, wherein the first configuration information, the second configuration information and the third configuration information are located in the group configuration information of the training group; or a training unit, configured to perform at least one of the following: perform model inference based on the local model; update the local model according to the model parameter update information from the central node of the training group; or send model training result information to the central node of the training group.
[0029] According to one aspect of some embodiments of the present disclosure, a network device is proposed, comprising: an information sending unit, configured to, when the RRC state of the terminal is inactive, the central node of the training group of the distributed model sends first configuration information or second configuration information to the terminal; and when the RRC state of the terminal is active, the central node of the training group of the distributed model sends first configuration information, second configuration information or third configuration information to the terminal, wherein: the first configuration information is used to instruct the terminal to deactivate the model training as a distributed node in the training group; the second configuration information is used to instruct the terminal to delete the model training as a distributed node in the training group; and the third configuration information is used to instruct the terminal to activate the model training as a distributed node in the training group.
[0030] In some embodiments, the network device also includes at least one of the following: a generation unit, configured to generate group configuration information based on at least one of establishment, update, addition, deletion, activation or deactivation of distributed nodes, wherein the first configuration information, the second configuration information and the third configuration information are located in the group configuration information of the training group; or a training control unit, configured to perform at least one of the following: sending model parameter update information to the distributed nodes of the training group; or the central node receiving model training result information from the distributed nodes of the training group.
[0031] According to one aspect of some embodiments of the present disclosure, a network device is provided, comprising: a memory; and a processor coupled to the memory, wherein the processor is configured to execute any one of the terminal configuration methods described above based on instructions stored in the memory.
[0032] According to one aspect of some embodiments of the present disclosure, a non-transitory computer-readable storage medium is provided, on which computer program instructions are stored, and when the instructions are executed by a processor, the steps of any one of the terminal configuration methods described above are implemented.
[0033] According to one aspect of some embodiments of the present disclosure, a network system is proposed, comprising at least one central node of a distributed model training group and multiple distributed nodes of a distributed model training group, wherein the distributed nodes are terminals, and the distributed nodes are configured to execute any one of the terminal configuration methods executed by the terminals described above; and the central node is configured to execute any one of the terminal configuration methods executed by the central node described above.
[0034] In some embodiments, the central node includes at least one of a 5G base station gNB, an operation and maintenance management entity, a location management function entity, or a user terminal. BRIEF DESCRIPTION OF THE DRAWINGS
[0035] The drawings described herein are used to provide a further understanding of the present disclosure and constitute a part of the present disclosure. The illustrative embodiments of the present disclosure and their descriptions are used to explain the present disclosure and do not constitute an improper limitation on the present disclosure. In the drawings:
[0036] Figure 1 The present invention is a flowchart of some embodiments of the terminal configuration method of the present invention.
[0037] Figure 2 The present invention provides flowcharts of other embodiments of the terminal configuration method of the present invention.
[0038] Figure 3 Schematic diagrams of some embodiments of the terminal disclosed herein.
[0039] Figure 4 Schematic diagram of some embodiments of the network device disclosed herein.
[0040] Figure 5 Schematic diagrams of other embodiments of the network device disclosed herein.
[0041] Figure 6 Schematic diagrams of some further embodiments of the network device disclosed herein.
[0042] Figure 7 Schematic diagrams of some embodiments of the network system disclosed herein. DETAILED DESCRIPTION
[0043] The technical solution of the present disclosure is further described in detail below through the accompanying drawings and embodiments.
[0044] The inventors have found that when implementing distributed model training in a wireless communication network, the current RAN side does not support the configuration of distributed model training. In order to implement performance monitoring of distributed model training, RAN needs to support the maintenance of the training group of the distributed model, including the establishment, update, addition, and deletion of group members, so as to improve and ensure the performance of distributed model training.
[0045] In response to the above problems, the present disclosure proposes a terminal configuration method, terminal, network device, system and storage medium to provide a configuration solution for supporting distributed model training on the RAN side and improve the control capability of distributed model training in wireless communication networks.
[0046] In some embodiments, the present disclosure proposes a distributed network architecture, including distributed nodes (or local clients for distributed model training) and central nodes (or servers for distributed model training). The distributed nodes perform model training, and the training results generated by the distributed nodes are aggregated to the central node; the central node can configure the distributed nodes, update the models, and aggregate the training results of each distributed node. In some embodiments, the distributed nodes are terminals. In some embodiments, the central node includes at least one of gNB, OAM, LMF, or UE.
[0047] The present disclosure discloses a flow chart of some embodiments of a terminal configuration method executed by a terminal side capable of serving as a distributed node. Figure 1 shown.
[0048] In step 130, the terminal receives information for configuring the state of the terminal in the training group of the distributed model. In some embodiments, the information may come from the central node of the training group of the distributed model. In some embodiments, the information includes first configuration information, second configuration information, and third configuration information, wherein the first configuration information is used to instruct the terminal to deactivate the model training as a distributed node in the training group of the distributed model; the second configuration information is used to instruct the terminal to perform a deletion operation of the model training as a distributed node in the training group of the distributed model; and the third configuration information is used to instruct the terminal to activate the model training as a distributed node in the training group.
[0049] In some embodiments, the distributed model may include a federated learning model, thereby avoiding the transmission of training data between distributed nodes and between distributed nodes and central nodes, thereby reducing the pressure of data transmission and improving data security.
[0050] Depending on the current RRC state of the terminal, the type of information that the terminal can receive for configuring the state of the terminal in the training group of the distributed model is also different. Specifically, step 130 includes step 131 and step 132.
[0051] In step 131, if the current RRC state of the terminal is in an inactive state, the terminal can receive the first configuration information and the second configuration information, that is, the information received by the terminal for configuring the state of the terminal in the training group of the distributed model is any one of the first and second configuration information. When the RRC state of the terminal is in an inactive state, the central node of the training group will not configure the terminal as a distributed node in an active state in the training group, and therefore, the terminal cannot receive the third configuration information.
[0052] In step 132, if the current RRC state of the terminal is a connected state, the terminal can receive the first, second and third configuration information, that is, the information received by the terminal for configuring the state of the terminal in the training group of the distributed model is any one of the first, second or third configuration information. The central node of the training group can use the terminal for distributed model training when the RRC state of the terminal is connected, so the terminal can be configured as a distributed node in the training group in an activated state, and the terminal can receive the third configuration information.
[0053] In some embodiments, if the terminal is in an offline state, the terminal may perform a training operation based on a local model without receiving configuration information from the central node.
[0054] Based on the method in the above embodiment, the RRC state of the terminal can be associated with the state of the terminal as a distributed node in the training group of the distributed model, thereby avoiding the untimely training and data feedback caused by requiring inactive terminals to perform model training, and improving the reliability and operational stability of the distributed model training.
[0055] In some embodiments, Figure 1 As shown in , the terminal configuration method also includes step 140: the terminal configures its current state as a distributed node in the training group according to the received first, second or third configuration information. In some embodiments, if the terminal receives the first configuration information, the terminal switches to the inactive state of the distributed node and suspends the training of the model belonging to the training group indicated in the first configuration information; if the terminal receives the second configuration information, the terminal exits the training group indicated in the second configuration information; if the terminal receives the second configuration information, the terminal switches to the active state of the distributed node and starts the model training of the training group indicated in the third configuration information.
[0056] Based on the method in this embodiment, the terminal can modify its own status according to the information from the central node, including starting model training, pausing model training, and exiting the training group (being offline in the training group), so that the control information from the central node can timely affect the status of the distributed nodes, thereby improving the flexibility, accuracy and efficiency of distributed node control.
[0057] The present disclosure discloses a flow chart of some embodiments of a terminal configuration method that can be performed by a central node that can serve as a distributed model training group. Figure 2 shown.
[0058] In step 220, the central node sends information for configuring the state of the terminal in the training group of the distributed model to the terminal. In some embodiments, the configured terminal is a distributed node in the same training group as the central node. In some embodiments, the information includes first configuration information, second configuration information, and third configuration information, wherein the first configuration information is used to instruct the terminal to perform deactivation of model training as a distributed node in the training group of the distributed model; the second configuration information is used to instruct the terminal to perform a deletion operation of model training as a distributed node in the training group of the distributed model; and the third configuration information is used to instruct the terminal to perform activation of model training as a distributed node in the training group.
[0059] Depending on the current RRC state of the terminal, the types of information that the central node can generate and send for configuring the state of the terminal in the training group of the distributed model are also different. Specifically, step 220 includes step 221 and step 222.
[0060] In step 221, if the current RRC state of the terminal is in an inactive state, the central node can send the first configuration information and the second configuration information to the terminal, that is, the information sent by the central node for configuring the state of the terminal in the training group of the distributed model is any one of the first and second configuration information. When the RRC state of the terminal is in an inactive state, the central node of the training group will not configure the terminal as a distributed node in an active state in the training group, and therefore, the central node cannot send the third configuration information.
[0061] In step 222, if the current RRC state of the terminal is a connected state, the terminal can receive the first, second and third configuration information, that is, the information sent by the central node for configuring the state of the terminal in the training group of the distributed model is any one of the first, second or third configuration information. The central node of the training group can use the terminal for distributed model training when the RRC state of the terminal is connected, so the terminal can be configured as a distributed node in the training group to be in an activated state, and the central node can receive the third configuration information.
[0062] Based on the method in the above embodiment, the RRC state of the terminal can be associated with the state of the terminal as a distributed node in the training group of the distributed model, thereby avoiding the untimely training and data feedback caused by requiring inactive terminals to perform model training, and improving the reliability and operational stability of the distributed model training.
[0063] In some embodiments, Figure 2 As shown, the terminal configuration method executed by the central node may further include step 210 executed before step 220.
[0064] In step 210, the central node generates first configuration information, second configuration information or third configuration information according to the RRC state of the terminal and its own control requirements.
[0065] In some embodiments, the control requirements of the central node may include at least one of activating model training, deactivating model training, and deleting distributed nodes.
[0066] In some embodiments, when the control center needs to activate model training, a terminal whose RRC state is connected is selected from the terminal to generate third configuration information. When the control center needs to deactivate model training or delete a distributed node, it can be selected from terminals whose RRC state is deactivated and connected. For terminals whose RRC state is offline, the control message of the control center is difficult to reach, so the control information for the terminal whose RRC state is offline is not generated.
[0067] Based on the method in the above embodiment, the central node can select a suitable terminal from the terminals to generate control information in combination with the RRC state of the terminal and its own control requirements, thereby ensuring that the control information can be executed by the terminal and improving the control success rate and control efficiency.
[0068] In some embodiments, the first configuration information, the second configuration information, and the third configuration information are located in the group configuration information of the training group. In addition to being able to perform deletion, activation, or deactivation according to the group configuration information, the terminal can also perform at least one of establishment, update, and addition of distributed nodes. In some embodiments, the terminal can obtain the necessary information to become a distributed node, such as model information, according to the group configuration information from the central node, to achieve the establishment of a distributed node; the terminal can update the stored model or model parameters according to the group configuration information from the central node to achieve the update of the distributed node; the terminal can join the training group specified by the central node to achieve the addition of a distributed node.
[0069] Based on the method in this embodiment, the flexibility of the central node in controlling the distributed nodes can be further improved, and the control capability of distributed model training in the wireless communication network can be improved.
[0070] In some embodiments, the distributed model training group has a group identifier (group ID), and the group identifier is unique within a cell, a cell group, a TA range, or a PLMN. In some embodiments, the group identifier as a whole is a unique identifier within the same TA, PLMN, gNB, or CU range, so that it is easy to uniquely identify the training group, facilitate subsequent control in units of training groups, and improve the accuracy of control identification. For example, the same terminal can be a distributed node for multiple training groups. The terminal determines the training group for which the configuration operation is targeted based on the group identifier in the received group configuration information, so as to avoid affecting the model training in other training groups and improve the reliability of distributed model training.
[0071] In some embodiments, the format of the group identifier can be: two bits connected in series, including: a first bit for the distributed model training type and a second bit for distinguishing training groups in the same distributed model training type. The group identifier obtained by splicing the first and second bits can uniquely identify the training group. Such a group identifier can uniquely identify the distributed model training group while also facilitating identification of the distributed model training type to which it belongs, thereby improving the convenience of management.
[0072] In some embodiments, the group configuration information includes at least one of a group identifier, a model identifier, or a central node identifier.
[0073] In some embodiments, the terminal determines the training group corresponding to the group configuration information based on the group identifier, thereby improving the accuracy of subsequent configuration and avoiding affecting the model training of other training groups; in some embodiments, the same training group may include one or more distributed models, and the terminal determines the model targeted by the group configuration information based on the model identifier, thereby improving the flexibility of control; in some embodiments, the same training group may include one or more central nodes, and the terminal determines the central node that initiates the control based on the central node identifier, thereby facilitating subsequent tracing of the control.
[0074] In some embodiments, the group configuration information also includes an identifier of at least one of the nodes, models or functions whose states have been changed. In some embodiments, the group configuration information includes an identifier of a changed node, such as an identifier of a deleted distributed node, a list of deleted distributed nodes, an identifier of an added distributed node, a list of added distributed nodes, an identifier of an activated distributed node, a list of activated distributed nodes, an identifier of a deactivated distributed node, or a list of deactivated distributed nodes; the group configuration information includes an identifier of a model whose state has been changed, such as an identifier of an activated model, a list of activated models, an identifier of a deactivated model, or a list of deactivated models; the group configuration information includes an identifier of a function whose state has been changed, such as an identifier of an activated function, a list of activated functions, an identifier of a deactivated function, or at least one of a list of deactivated functions. In some embodiments, different fields for carrying identification information can be set in the group configuration information, and the terminal determines the type of identification and the subsequent configuration method by identifying the field where the identification information in the group configuration information is located.
[0075] Through this method, the central node can initiate control over one or more distributed nodes from the node, model, and function dimensions, thereby improving control flexibility and efficiency.
[0076] In some embodiments, the group configuration information may carry at least one of the first configuration information, the second configuration information, or the third configuration information through an identifier of at least one of a node, a model, or a function whose state has been changed.
[0077] In some embodiments, the group configuration information may carry the first configuration information by at least one of an identifier of a deactivated distributed node, a list of deactivated distributed nodes, an identifier of a deactivated model, a list of deactivated models, an identifier of a deactivated function, or a list of deactivated functions. In some embodiments, when the identifier of the terminal matches the identifier of the deactivated distributed node or is in the list of deactivated distributed nodes, the terminal suspends model training of the training group corresponding to the group identifier in the group configuration information. In some embodiments, the terminal suspends training of the model corresponding to the identifier of the deactivated model or the list of deactivated models in the group configuration information. In some embodiments, the terminal suspends training of the model corresponding to the identifier of the deactivated function or the list of deactivated functions in the group configuration information.
[0078] In some embodiments, the group configuration information may carry the second configuration information through the identifier of the deleted distributed node or the list of the deleted distributed nodes. In some embodiments, when the identifier of the terminal matches the identifier of the deleted distributed node or is in the list of the deleted distributed nodes, the terminal exits the training group corresponding to the group identifier in the group configuration information.
[0079] In some embodiments, the group configuration information may carry the third configuration information by at least one of the identifier of an activated distributed node, a list of activated distributed nodes, an identifier of an activated model, a list of activated models, an identifier of an activated function, or a list of activated functions. In some embodiments, when the identifier of the terminal matches the identifier of the activated distributed node or is in the list of activated distributed nodes, the terminal starts model training of the training group corresponding to the group identifier in the group configuration information. In some embodiments, the terminal starts training the model corresponding to the identifier of the activated model or the list of activated models in the group configuration information. In some embodiments, the terminal starts training the model corresponding to the identifier of the activated function or the list of activated functions in the group configuration information.
[0080] Through the method in the embodiment shown above, the group configuration information can be activated and deactivated from multiple angles of distributed nodes, models, and functions, further improving the flexibility of control; by using the unique identification of nodes, models, and functions, the accuracy of the controlled object can be ensured and the reliability of control can be improved.
[0081] In some embodiments, the group configuration information may include an identifier match of an added distributed node or a list of added distributed nodes, and in this way carries information requiring one or more terminals to become distributed nodes of a specified training group. When the identifier of the terminal matches the identifier of the added distributed node or is in the list of added distributed nodes, the terminal joins the training group corresponding to the group identifier in the group configuration information. Through such a configuration method, the group configuration information can use the identifier of the distributed node to control the terminal as a distributed node in the training group, realize the update of the distributed node in the training group, and further improve the control capability.
[0082] In some embodiments, the group configuration information includes at least one of the following:
[0083] Group ID: The group ID is unique within a TA area, PLMN, gNB or CU;
[0084] Model ID or function ID: One or more models can be trained in a group, and different models have different models / functions;
[0085] Central node ID, link or interface information: for example, cell ID, etc.
[0086] Delete related configuration information: group deletion center distributed node ID, group deletion distributed center node list;
[0087] Add relevant configuration information: add distributed central node ID to the group, add distributed central node list to the group;
[0088] Group activation / deactivation ID;
[0089] Model / function activation ID, model / function activation list, model / function deactivation ID, model / function deactivation list.
[0090] In some embodiments, the group configuration information is transmitted via at least one of a data radio bearer, a control radio bearer, or a smart radio bearer, thereby improving the flexibility of group configuration information transmission.
[0091] In some embodiments, the terminal may receive the group configuration information through at least one of a user plane, a control plane, a data plane, and an intelligent plane, thereby improving the flexibility of transmitting the group configuration information.
[0092] In some embodiments, the group configuration information may be carried by at least one of system messages, RRC signaling, MAC CE signaling, DCI signaling, and NAS messages, thereby improving the flexibility and timeliness of group configuration information transmission.
[0093] In some embodiments, the distributed model training group can also activate or deactivate training in units of training groups. In some embodiments, the terminal receives group activation indication information, and the group activation indication information is used to indicate the start of model training belonging to the training group corresponding to the group identifier, and the group activation indication information includes the group identifier. According to the group activation indication information, the terminal starts the training of all models in the training group corresponding to the group identifier. In some embodiments, the terminal receives group deactivation indication information, and the group deactivation indication information is used to indicate the suspension of model training belonging to the training group corresponding to the group identifier, and the group deactivation indication information includes the group identifier. According to the group activation indication information, the terminal suspends the training of all models in the training group corresponding to the group identifier.
[0094] Through this method, training can be activated or deactivated in units of training groups, which improves control flexibility while also improving the efficiency of model start and pause control.
[0095] In some embodiments, the configuration method of the terminal further includes: the central node sends transmission indication information to the terminal, and the terminal receives the transmission indication information. The transmission indication information includes node and path information for information uploading, aggregation or sending. Through such a method, the path for interaction between the distributed nodes and the central node can be configured, thereby improving the controllability and reliability of information interaction in distributed model training.
[0096] In some embodiments, when the central node of the training group is a user terminal other than the distributed node of the current training group, the transmission indication information includes the identifier of the central node or the interface information between the terminals. In some embodiments, when the central node of the training group is a base station, the transmission indication information includes the base station identifier. Through such a method, the transmission path between the distributed node and the central node can be clarified.
[0097] In some embodiments, when the central node of the training group is a base station, and the base station is located in a TA or a base station group, and the distributed nodes include service users of the central node and service users of at least one other base station in the TA or the base station group, the transmission indication information also includes at least one of the identifier of the other base station, the link identifier between the central node and the first base station, or the routing information between the central node and the other base station. Through such a method, a path can be provided for the distributed node and the central node to communicate using other base stations in the area or base station group, thereby improving the flexibility of interaction between the distributed node and the central node.
[0098] In some embodiments, in the configuration method of the terminal disclosed in the present invention, the distributed model can be a model for calculating the positioning problem. In some embodiments, it can also be any distributed model used to reduce the bit error rate, reduce interference, increase channel capacity and information transmission rate, reduce energy consumption and delay, and improve frequency band utilization.
[0099] Through this method, the computing power and data collection capabilities of each terminal in the network can be utilized to solve problems in wireless communication technology based on distributed machine learning models, thereby improving the intelligence and communication quality of wireless communications.
[0100] In some embodiments, the distributed model can be a model for multiple purposes such as distributed positioning, distributed beam management, distributed CSI enhancement, and distributed wireless resource management, thereby reducing the transmission of user-location-related data between the air interface and the network element interface, protecting the security and privacy of user positioning-related data, reducing the transmission overhead related to the air interface and model training, and increasing the generalization and robustness of the model.
[0101] In some embodiments, the training group of the distributed model of the present disclosure has a group identifier, which is unique within a cell, a cell group, a TA range, or a PLMN, and the format of the group identifier is any one of the following two:
[0102] 1. Two parts of bits are connected in series, including: a distributed model training type bit, and a bit used to distinguish the same type. For example, an M-bit binary bit string is used to indicate the type of distributed model training, such as horizontal federation, vertical federation, federated transfer learning, single-task distributed learning, multi-task distributed learning, etc.; an N-bit binary bit string is used to distinguish different models in the same type of distributed model training, for example: multiple models trained using the same type of distributed model using different gradient descent methods, different optimization objectives or different data sets, where M and N are positive integers greater than or equal to 1 respectively;
[0103] 2. Bits used to distinguish different distributed model training groups. In some embodiments, the identifiers of different training groups may be globally unique or unique within a certain area.
[0104] In some embodiments, the group configuration information may be sent to the terminal by the central node, or may be sent to the terminal by other devices on the network side except the central node. In some embodiments, the signaling carrying the group configuration information may include at least one of a system message, an RRC signaling, a MAC CE signaling, a DCI signaling, and a NAS message.
[0105] In some embodiments, the group configuration information is transmitted via at least one of DRB, SRB, and IRB, wherein DRB and SRB are two types of radio bearers defined in relevant standards. In some embodiments, IRB is used to carry the transmission of model-related information related to AI (Artificial Intelligence) / ML (Machine Learning), or for the transmission of model training / reasoning-related information.
[0106] In some embodiments, the distributed node has at least one of the following states in the training group: active, inactive, and offline. The active state indicates that the distributed node participates in the distributed model training, and the local model of the distributed node can be used for model reasoning. The inactive state indicates that the distributed node does not participate in the distributed model training temporarily, and the local model of the distributed node can be used for model reasoning. The offline state indicates that the distributed node is unreachable, cannot participate in the distributed model training, and cannot perform model reasoning.
[0107] In some embodiments, the state of a distributed node in a training group has the following relationship with the RRC state of the distributed node:
[0108] The RRC state of a distributed node with different model identifiers / functional identifiers may be the same as or different from the state of the distributed node in the training group to which it belongs. For example, when the distributed node is in the RRC connected state, the state of the distributed node in the training group to which it belongs may be one of activated, inactivated, and offline; when the distributed node is in the RRC Inactive state, the state of the distributed node in the training group to which it belongs may be one of inactivated and offline; when the distributed node is in the RRC idle state, the state of the distributed node in the training group to which it belongs may be offline. For example, when the trained model has the function of cell access / cell reselection, when the distributed node is in the RRC idle state, the state of the distributed node in the group to which it belongs cannot be activated.
[0109] In some embodiments, the training group of the distributed model supports the uploading (aggregation) of relevant information from distributed nodes to the central node, as well as the node and path indication of information sent from the central node to the distributed nodes. In some embodiments, this can be carried by transmission indication information.
[0110] In some embodiments, the transmission indication information includes at least one of a central node ID, link or interface information, which is used to indicate the nodes and paths for uploading, aggregating, and sending model-related information in distributed model training.
[0111] In some embodiments, when the central node is a UE, it includes UE ID or inter-UE interface information, such as PC5 link identifier.
[0112] In some embodiments, when the central node is a base station, it includes the central node base station ID. When the central node gNB is in a TA area or a gNB group, and the distributed node UE includes the service users of the central gNB and some or all of the users served by other second gNBs in the TA area or gNB group where the central gNB is located, it is necessary to include the second base station ID, or the Xn link ID, or the corresponding routing information between the central gNB and the second gNB, for example: RAN node ID, TNLaddress of the Xn-C plane SCTP connection.
[0113] In some embodiments, the training group of the distributed model supports deleting one or more distributed nodes in the training group. In some embodiments, by deleting relevant configuration information (group deleted distributed node identifier, group deleted distributed node list), the distributed node with the distributed node identifier is deleted from the group to stop participating in the model training of the current training group.
[0114] In some embodiments, the training group of the distributed model supports adding one or more distributed nodes in the distributed model training. In some embodiments, by adding relevant configuration information (group adding distributed node identifier, group adding distributed node list), a distributed node with a distributed node identifier is added or newly added from the training group to participate in the group's model training.
[0115] In some embodiments, the training group of the distributed model supports activation / deactivation of the distributed model training. In some embodiments, the group activation identifier indicates to start all training work of the group with the corresponding group identifier; the group deactivation identifier indicates to suspend all training work of the group with the corresponding group identifier.
[0116] In some embodiments, the training group of the distributed model supports activation / deactivation of the training of a single or multiple models or functions of the distributed model training group. In some embodiments, the activation / deactivation of the training of a single or multiple models or functions in the group is indicated by a model or function activation identifier or a model or function activation list; the deactivation of the training of a single or multiple models or functions in the group is indicated by a model or function deactivation identifier or a model or function deactivation list.
[0117] Schematic diagrams of some embodiments of the terminal 31 of the present disclosure are as follows Figure 3 As shown, the terminal 31 can be used as a distributed node in the training group of the distributed model training.
[0118] The terminal 31 includes an information receiving unit 311, which can receive the first configuration information or the second configuration information when the radio resource control state of the terminal is inactive, and can receive the first configuration information, the second configuration information or the third configuration information when the radio resource control state of the terminal is connected. In some embodiments, the first configuration information is used to instruct the terminal to perform a deactivation or deletion operation of model training as a distributed node in a training group of a distributed model; the second configuration information is used to instruct the terminal to perform a deactivation or deletion operation of model training as a distributed node in a training group of a distributed model; and the third configuration information is used to instruct the terminal to perform activation of model training as a distributed node in the training group.
[0119] Such a terminal can associate the RRC state of the terminal with the state of the terminal as a distributed node in the training group of the distributed model, thereby avoiding untimely training and data feedback caused by inactive terminals being required to perform model training, and improving the reliability and operational stability of distributed model training.
[0120] In some embodiments, the terminal further includes a configuration unit 312, which can perform at least one of establishment, update, addition, deletion, activation or deactivation of a distributed node according to the group configuration information, wherein the first configuration information, the second configuration information and the third configuration information are located in the group configuration information of the training group. In some embodiments, the group configuration information can be any one of the above-mentioned.
[0121] Such a terminal can modify its own status according to the information from the central node, including starting model training, pausing model training, and exiting the training group (offline in the training group). It can also execute the establishment, update, and addition of distributed nodes, so that the control information from the central node can timely affect the status of the distributed nodes, thereby improving the flexibility, accuracy and efficiency of distributed node control.
[0122] In some embodiments, the terminal also includes a training unit 313, which can perform at least one of the following: perform model reasoning based on the local model; update the local model based on model parameter update information from the central node of the training group; or send model training result information to the central node of the training group.
[0123] Such terminals can serve as distributed nodes in the training group. Based on the configuration of the central node, they can use their own data and computing power to perform distributed model training, thereby improving the efficiency of model training in wireless communication networks and reducing the requirements for the capabilities of machine learning equipment. In addition, the distributed model can be a federated learning model, which further avoids the transmission of original data between nodes and improves data security.
[0124] Schematic diagrams of some embodiments of the network device 42 of the present disclosure are as follows Figure 4 As shown. The network device 42 can serve as a central node in the training group of the distributed model training. In some embodiments, the network device can be at least one of a 5G base station gNB, an operation and maintenance management entity OAM, a location management function entity LMF, or a user terminal UE.
[0125] The network device includes an information sending unit 43, which can send the first configuration information or the second configuration information to the terminal when the radio resource control RRC state of the terminal is inactive, and send the first configuration information, the second configuration information or the third configuration information to the terminal when the radio resource control RRC state of the terminal is active. In some embodiments, the first configuration information is used to instruct the terminal to deactivate the model training as a distributed node in the training group; the second configuration information is used to instruct the terminal to delete the model training as a distributed node in the training group; and the third configuration information is used to instruct the terminal to activate the model training as a distributed node in the training group.
[0126] Such network equipment can associate the RRC state of the terminal with the state of the terminal as a distributed node in the training group of the distributed model, thereby avoiding the untimely training and data feedback caused by requiring inactive terminals to perform model training, and improving the reliability and operational stability of the distributed model training.
[0127] In some embodiments, Figure 4 As shown, the network device also includes a generation unit 422, which can generate group configuration information according to at least one of establishment, update, addition, deletion, activation or deactivation of the distributed node, wherein the first configuration information, the second configuration information and the third configuration information are located in the group configuration information of the training group.
[0128] Such network equipment can require the terminal to modify its own status according to demand, including starting model training, pausing model training, and exiting the training group (offline in the training group). It can also execute the establishment, update, and addition of distributed nodes, thereby improving the flexibility, accuracy, and efficiency of distributed node control.
[0129] In some embodiments, Figure 4 As shown, the network device may further include a training control unit 41, capable of performing at least one of the following: sending model parameter update information to the distributed nodes of the training group; or the central node receiving model training result information from the distributed nodes of the training group.
[0130] Such a network device can serve as the central node in the training group, control the model parameters and status of the distributed nodes, and summarize the training results based on each distributed node, thereby improving the efficiency of model training in the wireless communication network; in addition, the distributed model can be a federated learning model, thereby further avoiding the transmission of original data between nodes and improving data security.
[0131] A structural diagram of an embodiment of the network device disclosed in the present invention is shown in FIG. Figure 5 As shown. The network device includes a memory 501 and a processor 502. Among them: the memory 501 can be a disk, a flash memory or any other non-volatile storage medium. The memory is used to store the instructions in the corresponding embodiment of the terminal configuration method executed by the terminal or the central node above. The processor 502 is coupled to the memory 501 and can be implemented as one or more integrated circuits, such as a microprocessor or a microcontroller. The processor 502 is used to execute the instructions stored in the memory, which can improve the reliability and operational stability of the training of the distributed model.
[0132] In one embodiment, it is also possible to Figure 6 As shown, the network device 600 includes a memory 601 and a processor 602. The processor 602 is coupled to the memory 601 via a BUS 603. The network device 600 can also be connected to an external storage device 605 via a storage interface 604 to call external data, and can also be connected to a network or another computer system (not shown) via a network interface 606. No further detailed description will be given here.
[0133] In this embodiment, by storing data instructions in a memory and then processing the instructions through a processor, the reliability and operational stability of the training of the distributed model can be improved.
[0134] In another embodiment, a computer-readable storage medium stores thereon computer program instructions, which, when executed by a processor, implement the steps of the method in the corresponding embodiment of the configuration method of the terminal executed by the terminal or the central node. It should be understood by those skilled in the art that the embodiments of the present disclosure may be provided as methods, devices, or computer program products. Therefore, the present disclosure may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present disclosure may take the form of a computer program product implemented on one or more computer-usable non-transient storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0135] Schematic diagrams of some embodiments of the network system disclosed in the present invention are as follows Figure 7As shown. The network system includes at least one central node 72 of a distributed model training group and multiple distributed nodes 711-71n of distributed model training groups, wherein n is an integer greater than 1, the distributed nodes are terminals, and the distributed nodes can execute any of the terminal configuration methods executed by the terminals described above; and the central node can execute any of the terminal configuration methods executed by the central node described above.
[0136] In some embodiments, the central node includes at least one of a 5G base station gNB, an operation and maintenance management entity OAM, a location management function entity LMF, or a user terminal UE.
[0137] Such a network system can associate the RRC state of the terminal with the state of the terminal as a distributed node in the training group of the distributed model, thereby avoiding the untimely training and data feedback caused by requiring the inactive terminal to perform model training, and improving the reliability and operational stability of the distributed model training. In addition, the mechanism of the network system disclosed in the present invention can be applied to various solutions such as distributed deployment, centralized deployment, or hybrid deployment of computing resources on the RAN side in the future, which is conducive to promotion and application.
[0138] The present disclosure is described with reference to flowcharts and / or block diagrams of methods, devices (systems) and computer program products according to embodiments of the present disclosure. It should be understood that each process and / or block in the flowchart and / or block diagram and the combination of processes and / or blocks in the flowchart and / or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0139] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.
[0140] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.
[0141] So far, the present disclosure has been described in detail. In order to avoid obscuring the concept of the present disclosure, some details known in the art are not described. Based on the above description, those skilled in the art can fully understand how to implement the technical solution disclosed here.
[0142] The method and apparatus of the present disclosure may be implemented in many ways. For example, the method and apparatus of the present disclosure may be implemented by software, hardware, firmware, or any combination of software, hardware, and firmware. The above order of steps for the method is for illustration only, and the steps of the method of the present disclosure are not limited to the order specifically described above, unless otherwise specifically stated. In addition, in some embodiments, the present disclosure may also be implemented as a program recorded in a recording medium, which includes machine-readable instructions for implementing the method according to the present disclosure. Therefore, the present disclosure also covers a recording medium storing a program for executing the method according to the present disclosure.
[0143] It should be noted that the terms "first", "second", etc. in the specification, claims, and drawings of the present disclosure are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products, or devices.
[0144] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present disclosure rather than to limit it. Although the present disclosure has been described in detail with reference to the preferred embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present disclosure can still be modified or some technical features can be replaced by equivalents without departing from the spirit of the technical solution of the present disclosure, which should be included in the scope of the technical solution for protection of the present disclosure.
Claims
1. A terminal configuration method, comprising: When the radio resource control state of the terminal is inactive, the terminal receives first configuration information or second configuration information; and When the radio resource control state of the terminal is a connected state, the terminal receives first configuration information, second configuration information or third configuration information, in, The first configuration information is used to instruct the terminal to deactivate the execution of model training as a distributed node in a training group of a distributed model; The second configuration information is used to instruct the terminal to perform a deletion operation of model training as a distributed node in a training group of a distributed model; and The third configuration information is used to instruct the terminal to activate as a distributed node in the training group to perform model training.
2. The configuration method according to claim 1, wherein: The first configuration information, the second configuration information and the third configuration information are located in the group configuration information of the training group, The configuration method further includes: the terminal executing at least one of establishing, updating, adding, deleting, activating or deactivating a distributed node according to the group configuration information.
3. The configuration method according to claim 2, wherein: The group configuration information meets at least one of the following requirements: The group configuration information is carried out through at least one of a data radio bearer, a control radio bearer or a smart radio bearer; The group configuration information is received via at least one of a user plane, a control plane, a data plane, and an intelligent plane; or The group configuration information is carried by at least one of a system message, a radio resource control signaling, a media access control MAC control element CE signaling, a downlink control information signaling, and a non-access layer message.
4. The configuration method according to claim 2 or 3, wherein: The group configuration information includes at least one of a group identifier, a model identifier or a central node identifier.
5. The configuration method according to claim 4, wherein: The group configuration information also includes an identifier of at least one of the nodes, models or functions whose status has been changed.
6. The configuration method according to claim 5, wherein: The group configuration information meets at least one of the following requirements: The identifier of the node whose state is changed includes at least one of an identifier of a deleted distributed node, a list of deleted distributed nodes, an identifier of an added distributed node, a list of added distributed nodes, an identifier of an activated distributed node, a list of activated distributed nodes, an identifier of a deactivated distributed node, or a list of deactivated distributed nodes; The identification of the model whose state is changed includes at least one of an identification of an activated model, a list of activated models, an identification of a deactivated model, or a list of deactivated models; or The identifier of the function whose status is changed includes at least one of an identifier of an activated function, a list of activated functions, an identifier of a deactivated function, or a list of deactivated functions.
7. The configuration method according to claim 4, wherein: The group identifier includes a first bit and a second bit, the first bit is a unique identifier of the distributed model training type, and the second bit is a unique identifier of the training group in the same distributed model training type; or the group identifier is a unique identifier of the training group within the same tracking area, public land mobile network, 5G base station or control plane unit.
8. The configuration method according to claim 1, further comprising: The terminal receives transmission indication information, wherein the transmission indication information includes node and path information for information uploading, aggregation or sending.
9. The configuration method according to claim 8, wherein: The central node of the training group is other user terminals, and the transmission indication information includes the identifier of the central node or the interface information between the terminals.
10. The configuration method according to claim 8, wherein: The central node of the training group is a base station, and the transmission indication information includes a base station identifier.
11. The configuration method according to claim 10, wherein: The central node of the training group is located in a tracking area TA or a base station group, and the distributed nodes include service users of the central node and service users of at least one first base station in the TA or the base station group, and the transmission indication information also includes at least one item of the identifier of the first base station, the link identifier between the central node and the first base station, or the routing information between the central node and the first base station.
12. The configuration method according to any one of claims 1 to 3, further comprising at least one of the following: The terminal performs model reasoning according to the local model; The terminal receives model parameter update information from the central node of the training group; or The terminal sends model training result information to the central node of the training group.
13. The method according to any one of claims 1 to 3, further comprising: The terminal configures a current state according to any one of the first configuration information, the second configuration information, or the third configuration information.
14. The configuration method according to claim 1, wherein: The terminal suspends training of the model belonging to the training group according to the first configuration information; The terminal exits the training group according to the second configuration information; or The terminal starts model training of the training group according to the third configuration information.
15. The configuration method according to claim 6, wherein: The group configuration information carries at least one of the first configuration information, the second configuration information, or the third configuration information through an identifier of at least one of a node, a model, or a function whose state has been changed; The configuration method further includes: In a case where the identifier of the terminal matches the identifier of the deleted distributed node or is in the list of the deleted distributed nodes, exiting the training group corresponding to the group identifier in the group configuration information; In a case where the identifier of the terminal matches the identifier of the added distributed node or is in the list of the added distributed nodes, adding the training group corresponding to the group identifier in the group configuration information; In a case where the identifier of the terminal matches the identifier of an activated distributed node or is in the list of activated distributed nodes, starting model training of the training group corresponding to the group identifier in the group configuration information; In a case where the identifier of the terminal matches the identifier of the deactivated distributed node or is in the list of the deactivated distributed nodes, suspending model training of the training group corresponding to the group identifier in the group configuration information; Start training a model corresponding to the identifier of the activated model or the list of activated models in the group configuration information; Suspending training of a model corresponding to an identifier of a deactivated model or a list of deactivated models in the group configuration information; Start training a model corresponding to the identifier of the activated function or the list of activated functions in the group configuration information; or The training of the model corresponding to the identifier of the deactivated function or the list of the deactivated functions in the group configuration information is suspended.
16. The configuration method according to claim 1, further comprising at least one of the following: The terminal receives group activation indication information, wherein: The group activation indication information is used to indicate the start of model training belonging to the training group corresponding to the group identifier, and the group activation indication information includes the group identifier; or The terminal receives group deactivation indication information, wherein the group deactivation indication information is used to instruct to suspend model training belonging to a training group corresponding to a group identifier, and the group deactivation indication information includes the group identifier.
17. The configuration method according to any one of claims 1 to 3, wherein: The distributed model includes a federated learning model.
18. A terminal configuration method, comprising: When the radio resource control state of the terminal is inactive, the central node of the training group of the distributed model sends the first configuration information or the second configuration information to the terminal; When the radio resource control state of the terminal is activated, the central node of the training group of the distributed model sends the first configuration information, the second configuration information or the third configuration information to the terminal, in: The first configuration information is used to instruct the terminal to deactivate the model training as a distributed node in the training group; The second configuration information is used to instruct the terminal to perform a deletion operation of model training as a distributed node in the training group; and The third configuration information is used to instruct the terminal to activate as a distributed node in the training group to perform model training.
19. The configuration method according to claim 18, wherein: The first configuration information, the second configuration information and the third configuration information are located in the group configuration information of the training group; The configuration method further includes: the central node generating the group configuration information according to at least one of establishment, update, addition, deletion, activation or deactivation of the distributed nodes.
20. The configuration method according to claim 18, further comprising: The central node sends transmission indication information to the distributed nodes, wherein the transmission indication information includes node and path information for information uploading, aggregation or sending.
21. The configuration method according to claim 19, further comprising at least one of the following: The central node sends model parameter update information to the distributed nodes of the training group; or The central node receives model training result information from the distributed nodes of the training group.
22. The configuration method according to claim 19, wherein: The configuration method meets at least one of the following requirements: The distributed model includes a federated learning model; or The central node is at least one of a 5G base station gNB, an operation and maintenance management entity, a location management function entity or a user terminal.
23. A terminal, comprising: an information receiving unit, configured to receive the first configuration information or the second configuration information when the radio resource control state of the terminal is inactive; and, when the radio resource control state of the terminal is a connected state, receiving first configuration information, second configuration information or third configuration information, in: The first configuration information is used to instruct the terminal to perform a deactivation or deletion operation of model training as a distributed node in a training group of a distributed model; The second configuration information is used to instruct the terminal to perform a deactivation or deletion operation of model training as a distributed node in a training group of a distributed model; and The third configuration information is used to instruct the terminal to activate as a distributed node in the training group to perform model training.
24. The terminal according to claim 23, further comprising at least one of the following: A configuration unit is configured to perform at least one of establishing, updating, adding, deleting, activating or deactivating distributed nodes according to the group configuration information, wherein: The first configuration information, the second configuration information and the third configuration information are located in the group configuration information of the training group; or The training unit is configured to perform at least one of the following: Perform model inference based on the local model; updating the local model according to the model parameter update information from the central node of the training group; or Send model training result information to the central node of the training group.
25. A network device comprising: The information sending unit is configured to, when the radio resource control state of the terminal is inactive, the central node of the training group of the distributed model sends the first configuration information or the second configuration information to the terminal; and when the radio resource control state of the terminal is active, the central node of the training group of the distributed model sends the first configuration information, the second configuration information or the third configuration information to the terminal, in: The first configuration information is used to instruct the terminal to deactivate the model training as a distributed node in the training group; The second configuration information is used to instruct the terminal to perform a deletion operation of model training as a distributed node in the training group; and The third configuration information is used to instruct the terminal to activate as a distributed node in the training group to perform model training.
26. The network device according to claim 25, further comprising at least one of the following: A generating unit is configured to generate group configuration information according to at least one of establishment, update, addition, deletion, activation or deactivation of a distributed node, wherein: The first configuration information, the second configuration information and the third configuration information are located in the group configuration information of the training group; or A training control unit is configured to perform at least one of the following: Sending model parameter update information to the distributed nodes of the training group; or The central node receives model training result information from the distributed nodes of the training group.
27. A network device comprising: Memory; as well as A processor coupled to the memory, the processor being configured to execute the method according to any one of claims 1 to 22 based on instructions stored in the memory.
28. A non-transitory computer-readable storage medium having computer program instructions stored thereon, which, when executed by a processor, implement the steps of the method of any one of claims 1 to 22.
29. A network system comprising at least one central node of a distributed model training group and a plurality of distributed nodes of distributed model training groups, wherein: The distributed node is a terminal, and the distributed node is configured to execute the method according to any one of claims 1 to 17; and The central node is configured to execute the method according to any one of claims 18 to 22.
30. The system of claim 29, wherein: The central node includes at least one of a 5G base station gNB, an operation and maintenance management entity, a location management function entity or a user terminal.
