Federated learning method, device, related equipment, and storage medium

Abstract

The present disclosure discloses a federated learning method, apparatus, related device, and storage medium, wherein the method includes a first network function sending a first request to a network control device corresponding to a type of federated learning node, the first request being used to request optimization of network connections between the federated learning nodes.

Description

[Technical Field]

[0001] (CROSS-REFERENCE TO RELATED APPLICATIONS) This disclosure is based on and claims priority from a Chinese patent application bearing application number 202310118232.6 and filed on February 10, 2023, the entire contents of which are incorporated herein by reference.

[0002] The present disclosure relates to the field of artificial intelligence technology, and in particular to a federated learning method, device, related equipment, and storage medium. [Background technology]

[0003] To solve the problems of data fragmentation and data silos caused by a lack of data samples in a single domain and the inability to share data between different domains, related technologies have proposed the concept of federated learning. Federated learning is a typical example of federated learning, which improves machine learning modeling effectiveness by performing distributed federated modeling without leaking data privacy and keeping proprietary data local. However, the execution efficiency of existing federated learning strategies is low. Summary of the Invention [Problem to be solved by the invention]

[0004] To solve the problems in the related art, embodiments of the present disclosure provide a federated learning method, apparatus, related devices, and storage media. [Means for solving the problem]

[0005] The technical solutions according to the embodiments of the present disclosure are realized as follows: An embodiment of the present disclosure provides a federated learning method applied to a first network function, the method comprising: sending a first request to a network control device corresponding to a type of the federated learning node, the first request being used to request optimization of network connections between the federated learning nodes;

[0006] In the above aspect, transmitting a first request to a network control device corresponding to the type of the federated learning node includes: When the network connection quality between the federated learning nodes is equal to or lower than a predetermined threshold, sending a first request to a network control device corresponding to the type of the federated learning node.

[0007] In the above aspect, the first request carries associated parameters for optimizing the network connection.

[0008] In the above aspect, when the federated learning node is a federated learning node in a carrier network, the network control device is a Software Defined Network (SDN) controller.

[0009] In the above embodiment, when the federated learning node is an external node of the operator network, the network control device is a corresponding connection node of the external network.

[0010] In the above aspect, when the federated learning node is a terminal (UE, User Equipment), the network control device is a policy control function (PCF, Policy Control Function).

[0011] In the above aspect, the first request is an Internet Protocol (IP) address of a federated learning node associated with the network connection to be optimized; and Quality of Service (QoS) requirements for network connections between federated learning nodes.

[0012] In the above aspect, the first request is an identifier of the UE; and the QoS requirements of the UE's bearer flows.

[0013] An embodiment of the present disclosure further provides a federated learning method applied to a network control device, the method comprising: receiving a first request sent from a first network function, the first request being used to request optimizing network connections between federated learning nodes, and the type of the federated learning node and the network control device corresponding to each other; optimizing the network connection based on the received first request.

[0014] In the above aspect, the first request carries associated parameters for optimizing the network connection.

[0015] In the above aspect, when the federated learning node is a federated learning node in an operator network, the network control device is an SDN controller.

[0016] In the above aspect, optimizing the network connection based on the received first request includes: Calculating an optimal route based on the received first request and transmitting the optimal route to associated routing devices, where the optimal route is used to transmit data between federated learning nodes.

[0017] In the above embodiment, when the federated learning node is a UE, the network control device is a PCF.

[0018] In the above aspect, optimizing the network connection based on the received first request includes: This includes sending Policy Control and Charging (PCC) policies to the User Plane Function (UPF) via the Session Management Function (SMF).

[0019] In the above aspect, the PCC policy: Priority of resource allocation; Guaranteed speed and The maximum speed and and a maximum packet loss rate.

[0020] In the above embodiment, when the federated learning node is an external node of the operator network, the network control device is a corresponding connection node of the external network.

[0021] In the above aspect, optimizing the network connection based on the received first request includes: and controlling a network control node corresponding to a corresponding connection node of said external network to configure an associated path.

[0022] In the above aspect, the first request is an IP address of a federated learning node associated with the network connection to be optimized; and QoS requirements for network connections between federated learning nodes.

[0023] In the above aspect, the first request is an identifier of the UE; and the QoS requirements of the UE's bearer flows.

[0024] In the above aspect, the method further comprises: When federated learning is completed, the method includes deleting the first request received.

[0025] An embodiment of the present disclosure further provides an associative learning device including a first sending unit; The first sending unit is configured to send a first request to a network control device corresponding to a type of the federated learning node, and the first request is used to request optimization of the network connection between the federated learning nodes.

[0026] An embodiment of the present disclosure further provides an associative learning device, including a first receiving unit and a first processing unit; the first receiving unit is configured to receive a first request sent from a first network function, the first request being used to request optimizing a network connection between federated learning nodes, and the type of the federated learning node and the network control device correspond to each other; The first processing unit is configured to optimize a network connection based on the received first request.

[0027] An embodiment of the present disclosure further provides a first network function including a first processor and a first communication interface; The first communication interface is configured to send a first request to a network control device corresponding to a type of federated learning node, and the first request is used to request optimization of network connections between the federated learning nodes.

[0028] An embodiment of the present disclosure further provides a network control device including a second processor and a second communication interface; the second communication interface is configured to receive a first request sent from a first network function, the first request being used to request optimizing a network connection between federated learning nodes, and the type of the federated learning node and the network control device correspond to each other; The second processor is configured to optimize the network connection based on the received first request.

[0029] An embodiment of the present disclosure further provides a first network function including a first processor and a first memory for storing a computer program executable by the first processor; Here, the first processor is configured to perform any of the method steps on the first network function side when executing the computer program.

[0030] An embodiment of the present disclosure further provides a network control device including a second processor and a second memory for storing a computer program executable by the second processor; Here, the second processor is configured to execute the steps of any of the methods on the network control device side when executing the computer program.

[0031] An embodiment of the present disclosure further provides a storage medium on which a computer program is stored, which, when executed by a processor, implements steps of any of the methods on the first network function side or steps of any of the methods on the network control device side. [Effects of the Invention]

[0032] In accordance with the federated learning method, apparatus, related device, and storage medium according to the embodiments of the present disclosure, a first network function sends a first request to a network control device corresponding to a type of federated learning node, the first request being used to request optimization of the network connection between the federated learning nodes, and the network control device receives the first request sent from the first network function and optimizes the network connection based on the received first request. In the above technical solution, the type of federated learning node and the network control device correspond to each other, and during the federated learning process, the network control device dynamically optimizes the network connection between the federated learning nodes, thereby improving the quality of the network connection between the federated learning nodes, further improving the efficiency of data transmission between the federated learning nodes, and improving the execution efficiency of the federated learning. [Brief explanation of the drawings]

[0033] [Figure 1] FIG. 1 is an exemplary diagram illustrating a federated learning system architecture applied to an embodiment of the present disclosure. [Figure 2] 1 is a schematic diagram illustrating an implementation flow of a federated learning method according to an embodiment of the present disclosure. [Figure 3] 1 is a schematic diagram illustrating an implementation flow of a federated learning method according to an embodiment of the present disclosure. [Figure 4] FIG. 1 is a schematic diagram of an interaction flow of a federated learning method according to an embodiment of the present disclosure. [Figure 5]FIG. 1 is a schematic diagram of an interaction flow of a federated learning method according to an embodiment of the present disclosure. [Figure 6] FIG. 1 is a schematic diagram of an interaction flow of a federated learning method according to an embodiment of the present disclosure. [Figure 7] FIG. 1 is a schematic diagram illustrating the configuration of a federated learning device according to an embodiment of the present disclosure. [Figure 8] FIG. 1 is a schematic diagram illustrating the configuration of a federated learning device according to an embodiment of the present disclosure. [Figure 9] FIG. 2 is a structural schematic diagram of a first network function according to an embodiment of the present disclosure; [Figure 10] FIG. 1 is a structural schematic diagram of a network control device according to an embodiment of the present disclosure. DETAILED DESCRIPTION OF THE INVENTION

[0034] Federated learning is a subset or paradigm of distributed machine learning (ML), and is also known as distributed federated learning. Currently, support for federated learning has been introduced in the network intelligence application scene. That is, intelligent computing nodes in the network are distributed and federated learning is supported between the intelligent computing nodes. An intelligent computing node can be understood as a node with computing or data processing capabilities.

[0035] In the conventional technology solutions, optimization points for the federated learning implementation mechanism are mainly focused on the federated learning algorithms such as node selection, parameter aggregation, and parameter compression, but in the solution where the network supports federated learning, there is no method to combine the advantages of the network to optimize the data transmission efficiency between the federated learning nodes. Furthermore, since transmission efficiency is an important factor affecting the execution efficiency of federated learning, this leads to low execution efficiency of the existing federated learning solutions.

[0036] Based on this, in each embodiment of the present disclosure, the first network function sends a first request to a network control device corresponding to the type of the federated learning node, the first request is used to request optimization of the network connection between the federated learning nodes, and the network control device receives the first request sent from the first network function and optimizes the network connection based on the received first request. In the above technical solution, the type of the federated learning node and the network control device correspond to each other, and during the federated learning process, the network control device dynamically optimizes the network connection between the federated learning nodes, thereby improving the network connection quality between the federated learning nodes, further improving data transmission efficiency between the federated learning nodes, and improving the execution efficiency of the federated learning.

[0037] The present disclosure will now be described in more detail with reference to the drawings and examples.

[0038] First, to better explain the federated learning method according to an embodiment of the present disclosure, FIG. 1 illustrates an example of a federated learning system architecture applicable to an embodiment of the present disclosure. In the system architecture illustrated in FIG. 1, a consumer using an ML service can request an ML service from a first network function (NF) in an ML underlay network. The NF in the ML underlay network can be understood as a basic network function of an operator. The first network function mainly provides correlation services for federated learning. The provided services include receiving federated learning tasks announced by consumers using the ML service, analyzing the federated learning tasks, determining federated learning nodes, distributing the federated learning tasks to the federated learning nodes, monitoring the network connection status between the federated learning nodes, and sending a first request to a network control device. The first request is used to request optimization of the network connection between the federated learning nodes. The federated learning node can be understood as a network node or an intelligent computing node that executes the federated learning task or participates in federated learning. The network node or the intelligent computing node can be a training server function or a data server function. A federated learning task may be understood as a distributed federated learning task or federated learning request.

[0039] The first network function may be an artificial intelligence (AI) service orchestration function in an operator network, or an ML service orchestration function in an operator network, such as a machine learning function orchestration (MLFO) service function in an operator network. Consumers of ML services include internal consumers and external consumers, and external consumers interact with NFs in the ML underlay network through the outward-facing public interface of the ML underlay network. Internal consumers include NFs in the ML underlay network, and external consumers include consumers of ML services in external systems or operator external networks.

[0040] The federated learning node may include a parameter aggregation node (DJL server), a distributed execution node (DJL client), and a distributed data processing node (Data service). Here, DJL stands for Distributed Joint Learning. The first network function performs overall orchestration control for both the parameter aggregation node and the distributed execution node, and they must register with the first network function in advance. The parameter aggregation node may be understood as a federated learning service node (Joint learning server). The distributed execution node may be understood as a client node (Training client) or client execution node. The distributed data processing node is associated with the client node, and is also called a data processing node or data service node. A training service function within the operator network can function as a parameter aggregation node and / or a distributed execution node.

[0041] The parameter aggregation node and the distributed execution node may be located inside the operator network, or may be located in at least one of a UE, an application function (AF), and an external enterprise IT system, or may be located in an operator external network such as another operator network. The other operator network may include an International Mobile Telecommunications (IMT) network, such as an IMT-2020 network. That is, the parameter aggregation node and the distributed execution node may be located inside the operator network or outside the operator network (a node in the operator external network). The parameter aggregation node and the distributed execution node located in the operator external network can access the ML underlay network via the external public interface of the ML underlay network and perform signaling interactions with the first network function.

[0042] The network control device may include at least one of a controller responsible for lower-layer routing devices (routers), a PCF, and a corresponding connection node of an external network. The external network may be understood as an operator's external network. The controller responsible for the lower-layer routing devices includes a Software Defined Network (SDN) controller. The SDN controller and the router jointly configure the lower-layer transmission network of the operator's network and provide network connections for network nodes within the operator's network, where the network nodes include federated learning nodes.

[0043] An embodiment of the present disclosure provides a federated learning method applied to a first network function in the system architecture of FIG. 1, and as shown in FIG. 2, the method includes the following step 201:

[0044] In step 201, a first request is sent to a network control device corresponding to a type of federated learning node, where the first request is used to request optimization of network connections between federated learning nodes.

[0045] Here, the first network function receives a federated learning task published by a consumer using the ML service, determines a federated learning node based on the received federated learning task, determines a type of the federated learning node, determines a network control device corresponding to the type of the federated learning node, and sends a first request to the determined network control device. The first request can be understood as a request for optimizing a network connection. The first network function can send the first request to a network controller corresponding to the type of some or all of the federated learning nodes, where the network controller corresponding to the type of some or all of the federated learning nodes can be a network controller corresponding to the type of federated learning nodes associated with the network connection to be optimized.

[0046] The implementation process of determining a federated learning node based on a received federated learning task can be to analyze the received federated learning task to obtain an analysis result, and select a federated learning node from the intelligent computing nodes based on the analysis result. Here, when analyzing the received federated learning task, the needs of the federated learning task and the characteristics of the data belonging to it can be analyzed, and the data characteristics of the intelligent computing node can be analyzed, and the distance between the location of the intelligent computing node and the consumer using the ML service can be analyzed.

[0047] The type of a federated learning node can be set according to actual circumstances, and there is a correspondence or association relationship between the type of federated learning node and a network control device. A network control device corresponding to a type of federated learning node can be understood as a network control device responsible for controlling the network connection of that type of federated learning node.

[0048] Furthermore, since network connection control between different types of federated learning nodes is handled by different types of network control devices, by sending the first request to a network control device corresponding to the type of federated learning node without sending the first request to all network control devices in the network, the signaling overhead of the first network function can be reduced, and resources of the network controller unrelated to the network connection to be optimized can be saved, and the success rate of optimizing the network connection can be improved.

[0049] In one embodiment, the type of the federated learning node includes at least one of a first type, a second type, and a third type. Here, the first type is used to indicate that the federated learning node is a federated learning node within an operator network, and includes at least one of a parameter aggregation node, a distributed execution node, and a distributed data processing node, and the federated learning node within the operator network can be understood as the training service function and / or the data service function within the operator network in FIG. 1. The second type is used to indicate that the federated learning node is a UE. The third type is used to indicate that the federated learning node is an external node of the operator network, and the external node of the operator network includes a parameter aggregation node and / or a distributed execution node.

[0050] During the federated learning process, if the nodes participating in the federated learning may change due to factors such as the state of the federated learning node, the network state, or user needs, the first network function needs to reselect or adjust the nodes that perform the federated learning task, and if the federated learning nodes change, the network connections between the federated learning nodes may also change, so the first network function needs to re-determine the first request. Based on this, in one embodiment, the method further comprises: If the federated learning node changes, the first requirement is determined anew.

[0051] Here, the first network function is state information of the federated learning node; The network status and Location information of the federated learning node; The data characteristics of the federated learning node; and the needs information of the federated learning task, a decision can be made as to whether to modify or adjust the federated learning node.

[0052] When the federated learning nodes are changed or adjusted, the network connections between the federated learning nodes are determined anew based on the changed or adjusted federated learning nodes, and the first request is determined anew based on the newly determined network connections between the federated learning nodes.

[0053] Here, the status information of the federated learning node may be used to indicate whether the federated learning node is available or whether a failure has occurred. It is understood that the availability of the network status may be used to indicate whether the network connection status is stable or whether an abnormality exists. The data characteristics of the federated learning node may be understood as the data type, data processing method, etc. supported by the federated learning node. The needs information of the federated learning task may include the capacity needs and / or data processing needs of the intelligent computing node.

[0054] In order to save the signaling overhead of the first network function and the data processing resources of the network control device, when the network connection quality between the federated learning nodes is poor, the first network function can request the network control device to optimize the network connection between the federated learning nodes. Accordingly, in one embodiment, sending a first request to a network control device corresponding to the type of the federated learning node includes: When the network connection quality between the federated learning nodes is equal to or lower than a predetermined threshold, sending a first request to a network control device corresponding to the type of the federated learning node.

[0055] Here, the first network function detects the quality of the network connection between the federated learning nodes performing the distributed federated learning task, and if the quality of the network connection between the federated learning nodes is below a predetermined threshold, the network connection between the federated learning nodes needs to be optimized, and the first network function sends a first request to a network control device corresponding to the type of the federated learning node, where the network connection quality can be measured or evaluated by indicators such as packet loss rate and / or delay.

[0056] If the network connection quality between the federated learning nodes is greater than a predetermined threshold, the first request does not have to be sent to a network control device corresponding to the type of the federated learning node.

[0057] To improve the efficiency and success rate of the network controller optimizing the network connection, in one embodiment, the first request carries relevant parameters for optimizing the network connection.

[0058] Since network connection control between different types of federated learning nodes is handled by different types of network control devices, in order to accurately optimize the network connection to be optimized, in one embodiment, when the federated learning node is a federated learning node in an operator network, the network control device is an SDN controller.

[0059] In order to be able to accurately optimize the network connection to be optimized, in one embodiment, when the federated learning node is a UE, the network control device is a PCF.

[0060] In order to be able to accurately optimize the network connection to be optimized, in one embodiment, when the federated learning node is an external node of the operator network, the network control device is a corresponding connection node of the external network.

[0061] Here, the external network can be understood as an operator external network, and the corresponding connection node of the external network includes at least one of an AF, an IT system of an external enterprise, and an MLFO service function of another operator network (MLFO of an external network).

[0062] If the federated learning node is a federated learning node within an operator network or an external node of the operator network, the first request: an IP address of a federated learning node associated with the network connection to be optimized; and QoS requirements for network connections between federated learning nodes.

[0063] Here, if the network connection to be optimized is associated with a federated learning node in an operator network, the first network function sends a first request to the SDN controller, since the SDN controller is responsible for controlling network connections between federated learning nodes in the operator network.

[0064] When the network connection to be optimized is associated with an external node of the operator network (a federated learning node of the operator external network), the network connection control of the external node of the operator network is handled by the corresponding connection node of the external network, and the network control device connected to correspond to the first network function is the corresponding connection node of the external network, so the first network function sends a first request to the corresponding connection node of the external network via a predetermined interface. Note that whether the network connection of the operator external network can be optimized depends on the function of the operator external network.

[0065] The first request carries at least one of an IP address of a federated learning node associated with the network connection to be optimized and a QoS requirement for the network connection between the federated learning nodes, where the IP address of the federated learning node is used by the network control device to determine the network connection to be optimized. The federated learning node associated with the network connection to be optimized includes at least one of a parameter aggregation node, a distributed execution node, and a distributed data processing node. The QoS requirement for the network connection includes at least one of a maximum delay, a minimum bandwidth, a maximum packet loss rate, and a maximum jitter.

[0066] When the federated learning node is a UE, in one embodiment, the first request is: an identifier of the UE; and the QoS requirements of the UE's bearer flows.

[0067] Here, if the network connection to be optimized is associated with a UE, the first network function sends a first request to a PCF responsible for user plane control of the UE. In this case, the first request can be understood as a user plane optimization request. The first request carries at least one of a UE identifier and QoS requirements of the UE's bearer flow. The UE identifier includes a Subscription Permanent Identifier (SUPI) and / or an International Mobile Subscriber Identity (IMSI). The QoS requirements of the UE's bearer flow include at least one of a maximum delay, a maximum packet loss rate, and a maximum jitter. The UE's bearer flow can be understood as a UE's Protocol Data Unit (PDU) session channel, also referred to as a PDU session channel.

[0068] Corresponding to the above, an embodiment of the present disclosure further provides a federated learning method applied to a network control device, and as shown in FIG. 3, the method includes the following steps 301 and 302:

[0069] In step 301, a first request sent from a first network function is received. The first request is used to request optimization of network connections between federated learning nodes, and the types of the federated learning nodes and the network control devices correspond to each other.

[0070] In step 302, the network connection is optimized based on the received first request.

[0071] Here, the network control device determines the network connection to be optimized based on the received first request, and optimizes the network connection to be optimized.

[0072] To improve the efficiency of optimizing the network connection, in one embodiment, the first request is accompanied by relevant parameters for optimizing the network connection.

[0073] In one embodiment, when the federated learning node is a federated learning node in an operator network, the network control device is an SDN controller.

[0074] In one embodiment, when the federated learning node is an external node of the operator network, the network control device is a corresponding access node of the external network.

[0075] In one embodiment, when the federated learning node is a UE, the network control device is a PCF.

[0076] When the network control device is an SDN controller or a corresponding connection node of an external network, in one embodiment, the first request is: an IP address of a federated learning node associated with the network connection to be optimized; and QoS requirements for network connections between federated learning nodes.

[0077] Here, the network control device can determine the federated learning node based on the IP address of the federated learning node associated with the network connection to be optimized, and determine the network connection associated with the federated learning node to obtain the network connection to be optimized.

[0078] When the network control device is a PCF, in one embodiment, the first request is: an identifier of the UE; and the QoS requirements of the UE's bearer flows.

[0079] If the federated learning node associated with the network connection to be optimized is a federated learning node in an operator network, the network control device is an SDN controller, and the first network function sends a first request to the SDN controller. Based on this, to optimize the network connection between the federated learning nodes in the operator network, in one embodiment, optimizing the network connection based on the received first request includes: Calculating an optimal route based on the received first request and transmitting the optimal route to associated routing devices, where the optimal route is used to transmit data between federated learning nodes.

[0080] Here, when the SDN controller receives the first request, it calculates an optimal route between the federated learning nodes based on the information attached to the received first request, and transmits the optimal route to a routing device associated with the optimal route, so that the associated routing device transmits relevant data of the federated learning node based on the received optimal route, thereby improving data transmission efficiency.

[0081] For example, the SDN controller determines the network connections to be optimized based on the IP addresses of the federated learning nodes associated with the network connections to be optimized, and calculates the optimal route between the federated learning nodes associated with the network connections to be optimized based on the network topology structure of the operator network and the IP addresses of the federated learning nodes and / or the QoS requirements of the network connections between the federated learning nodes.

[0082] If the federated learning node associated with the network connection to be optimized is a UE, the network control device is a PCF, and the first network function sends a first request to the PCF, so as to optimize the network connection associated with the UE based on the received first request. In one embodiment, optimizing the network connection based on the received first request includes: This includes sending the PCC policy to the UPF via the SMF.

[0083] Here, the PCF determines a PCC policy based on the information attached to the first request and sends the PCC policy to the SMF, so that the SMF forwards the received PCC policy to the UPF, and the UPF optimizes the transmission policy of the UE's associated data based on the received PCC policy to improve the transmission efficiency of the UE's associated data. Here, the PCC policy may be pre-configured or may be determined by the PCF based on the QoS requirements of the UE's bearer flow attached to the first request.

[0084] In one embodiment, the PCC policy is: Priority of resource allocation; Guaranteed speed and The maximum speed and and a maximum packet loss rate.

[0085] If the federated learning node associated with the network connection to be optimized is an external node of the operator network, the network control device is a corresponding connection node of the external network, and the first network function sends a first request to the corresponding connection node of the external network. Accordingly, to optimize the network connection of the external node of the operator network, in one embodiment, optimizing the network connection based on the received first request includes: and controlling a network control node corresponding to a corresponding connection node of said external network to configure an associated path.

[0086] Here, in order to improve the efficiency of data transmission between external nodes of an operator network, when a corresponding connection node of the external network receives a first request sent from a first network function, the corresponding connection node of the external network can forward the first request to a network control node corresponding to the corresponding connection node of the external network, and can control the network control node to configure a data transmission path between associated federated learning nodes based on information attached to the first request.

[0087] When the federated learning is completed, in order to save the storage resources of the network control device, in one embodiment, the method further comprises: When federated learning is completed, the method includes deleting the first request received.

[0088] Here, when federated learning is completed, the network controller no longer needs to pay attention to the network connection between the federated learning nodes, and deletes the first request that it received.

[0089] The present disclosure will now be explained in more detail with reference to application examples and interaction flow diagrams.

[0090] [Application example 1] The federated learning node associated with the network connection to be optimized is a federated learning node in the operator network, and the network control device is an SDN controller. As shown in Figure 4, the federated learning method includes the following steps 1 to 8.

[0091] In step 1, a first network function receives a federated learning task published by a consumer using an ML service.

[0092] Here, the federated learning task can also be understood as a federated learning request. The first network function can be an MLFO of the operator network.

[0093] In step 2, the first network function determines a federated learning node based on the received federated learning task.

[0094] Here, the first network function analyzes the received federated learning task to obtain an analysis result, and selects a federated learning node from the intelligent computing nodes based on the analysis result, for example, it can select a parameter aggregation node, a distributed execution node, and a distributed data processing node to participate in the federated learning.

[0095] Here, when analyzing the received federated learning task, the needs of the federated learning task and the characteristics of the associated data can be analyzed, the characteristics of the data of the intelligent computing node can be analyzed, and the distance between the location of the intelligent computing node and the consumer using the ML service can be analyzed.

[0096] wherein the first network function further comprises: state information of the federated learning node; The network status and Location information of the federated learning node; The data characteristics of the federated learning node; The federated learning nodes can be determined or adjusted again based on at least one of the federated learning task needs information.

[0097] In step 3, the first network function sends a federated learning task to the determined federated learning node.

[0098] Here, the federated learning node performs federated learning based on the received federated learning task.

[0099] In step 4, the first network function detects the network connection quality between the federated learning nodes.

[0100] In step 5, if the network connection quality between the federated learning nodes is below a predetermined threshold, and the federated learning node associated with the network connection to be optimized is a federated learning node in the operator network, the first network function sends a first request to the SDN controller.

[0101] Wherein, if the network connection quality between the federated learning nodes is greater than a predetermined threshold, the first network function may not send the first request, where the first request includes: an IP address of a federated learning node associated with the network connection to be optimized; and QoS requirements for network connections between the federated learning nodes, the QoS requirements including at least one of maximum delay, minimum bandwidth, maximum packet loss rate, and maximum jitter.

[0102] Here, the first network function can further determine the first requirement anew when the federated learning node changes.

[0103] In step 6, the SDN controller receives the first request and calculates an optimal route based on the received first request.

[0104] In one embodiment, once federated learning is complete, the SDN controller may also delete the received first request.

[0105] In step 7, the SDN controller sends the optimal route to the associated routing device.

[0106] In step 8, the routing device transmits data between the federated learning nodes based on the received optimal route.

[0107] [Application example 2] The federated learning node associated with the network connection to be optimized is the UE, and the network control device is the PCF. As shown in Figure 5, the federated learning method includes the following steps 1 to 8.

[0108] In step 1, a first network function receives a federated learning task published by a consumer using an ML service.

[0109] For the process of implementing steps 1 to 4 in FIG. 5, please refer to the related explanation in Application Example 1, and the explanation will be omitted here.

[0110] In step 2, the first network function determines a federated learning node based on the received federated learning task.

[0111] In step 3, the first network function sends a federated learning task to the determined federated learning node.

[0112] In step 4, the first network function detects the network connection quality between the federated learning nodes.

[0113] In step 5, if the network connection quality between the federated learning nodes is below a predetermined threshold, and the federated learning node associated with the network connection to be optimized is a UE, the first network function sends a first request to the PCF.

[0114] Here, the first requirement is: an identifier of the UE; and the QoS requirements of the UE's bearer flows.

[0115] Wherein, the UE identifier includes a SUPI and / or an IMSI, and the QoS requirements of the UE bearer flow include at least one of a maximum delay, a maximum packet loss rate, and a maximum jitter.

[0116] In step 6, the PCF receives the first request sent from the first network function and determines a PCC policy based on the received first request.

[0117] Here, the PCC policy is: Priority of resource allocation; Guaranteed speed and The maximum speed and and a maximum packet loss rate.

[0118] In one embodiment, once federated learning is complete, the PCF may also delete the first received request.

[0119] In step 7, the PCF sends the determined PCC policy to the SMF.

[0120] In step 8, the SMF forwards the received PCC policy to the UPF, so that the UPF can optimize the transmission policy of the UE's associated data based on the received PCC policy.

[0121] [Application Example 3] The federated learning node associated with the network connection to be optimized is an external node of the operator network, and the network control device is a corresponding connection node of the external network. As shown in Figure 6, the federated learning method includes the following steps 1 to 6.

[0122] In step 1, a first network function receives a federated learning task published by a consumer using an ML service.

[0123] For the process of implementing steps 1 to 4 in FIG. 6, please refer to the related explanation in Application Example 1, and the explanation will be omitted here.

[0124] In step 2, the first network function determines a federated learning node based on the received federated learning task.

[0125] In step 3, the first network function sends a federated learning task to the determined federated learning node.

[0126] In step 4, the first network function detects the network connection quality between the federated learning nodes.

[0127] In step 5, if the network connection quality between the federated learning nodes is below a predetermined threshold, and the federated learning node associated with the network connection to be optimized is an external node of the operator network, the first network function sends a first request to a corresponding connection node of the external network.

[0128] Here, the corresponding connection node of the external network can be the MLFO of the external network.

[0129] The first requirement is: an IP address of a federated learning node associated with the network connection to be optimized; and QoS requirements for network connections between the federated learning nodes, the QoS requirements including at least one of maximum delay, minimum bandwidth, maximum packet loss rate, and maximum jitter.

[0130] In step 6, the corresponding connection node of the external network controls the network control node corresponding to the corresponding connection node of the external network to configure a relevant path according to the received first request.

[0131] Here, in order to improve the efficiency of data transmission between external nodes of the operator network, the corresponding connection node of the external network can forward the received first request to a corresponding network control node, and can instruct the network control node to configure a data transmission path between the associated federated learning nodes based on the information attached to the first request.

[0132] In accordance with the federated learning method, apparatus, related device, and storage medium according to the embodiments of the present disclosure, a first network function sends a first request to a network control device corresponding to a type of federated learning node, the first request being used to request optimization of the network connection between the federated learning nodes, and the network control device receives the first request sent from the first network function and optimizes the network connection based on the received first request. In the above technical solution, the type of federated learning node and the network control device correspond to each other, and during the federated learning process, the network control device dynamically optimizes the network connection between the federated learning nodes, thereby improving the quality of the network connection between the federated learning nodes, further improving the efficiency of data transmission between the federated learning nodes, and improving the execution efficiency of the federated learning.

[0133] To realize the federated learning method of the embodiment of the present disclosure, the embodiment of the present disclosure further provides a federated learning device disposed in the first network function, which includes a first sending unit 701, as shown in FIG.

[0134] The first sending unit 701 is configured to send a first request to a network control device corresponding to a type of federated learning node, and the first request is used to request optimizing network connections between the federated learning nodes.

[0135] In one embodiment, the first sending unit 701 is specifically configured to send a first request to a network control device corresponding to the type of the federated learning node when the network connection quality between the federated learning nodes is below a predetermined threshold.

[0136] In one embodiment, the first request carries associated parameters for optimizing the network connection.

[0137] In one embodiment, when the federated learning node is a federated learning node in an operator network, the network control device is an SDN controller.

[0138] In one embodiment, when the federated learning node is an external node of the operator network, the network control device is a corresponding access node of the external network.

[0139] In one embodiment, when the federated learning node is a UE, the network control device is a PCF.

[0140] In one embodiment, the first request is: an IP address of a federated learning node associated with the network connection to be optimized; and QoS requirements for network connections between federated learning nodes.

[0141] In one embodiment, the first request is: an identifier of the UE; and the QoS requirements of the UE's bearer flows.

[0142] In practical application, the first sending unit 701 can be realized by a combination of a processor and a communication interface in a federated learning device.

[0143] Although the above embodiments have been described with reference to the division of each program module as an example when the associative learning device performs associative learning, in actual applications, the above processes can be completed by allocating them to different program modules as needed, i.e., the internal structure of the device can be divided into different program modules to complete all or part of the above-described processes. Furthermore, the associative learning device according to the above embodiments is based on the same concept as the associative learning method embodiments, and specific implementation processes can be achieved by referring to the method embodiments, and therefore will not be described here.

[0144] To realize the federated learning method of the embodiment of the present disclosure, the embodiment of the present disclosure further provides a federated learning device provided in the network control equipment, and as shown in FIG. 8, the device includes a first receiving unit 801 and a first processing unit 802.

[0145] The first receiving unit 801 is configured to receive a first request sent from a first network function, the first request is used to request optimizing network connections between federated learning nodes, and the types of the federated learning nodes and the network control devices correspond to each other.

[0146] The first processing unit 802 is configured to optimize a network connection based on the received first request.

[0147] In one embodiment, the first request carries associated parameters for optimizing the network connection.

[0148] In one embodiment, when the federated learning node is a federated learning node in an operator network, the network control device is an SDN controller.

[0149] In one embodiment, the first processing unit 802 is specifically configured to calculate an optimal route based on the received first request and send the optimal route to an associated routing device, where the optimal route is used to transmit data between federated learning nodes.

[0150] In one embodiment, when the federated learning node is a UE, the network control device is a PCF.

[0151] In one embodiment, the first processing unit 802 is specifically configured to send a PCC policy to a UPF via an SMF.

[0152] In one embodiment, the PCC policy is: Priority of resource allocation; Guaranteed speed and The maximum speed and and a maximum packet loss rate.

[0153] In one embodiment, when the federated learning node is an external node of the operator network, the network control device is a corresponding access node of the external network.

[0154] In one embodiment, the first processing unit 802 is specifically configured to control a network control node corresponding to a corresponding access node of the external network to configure a related path.

[0155] In one embodiment, the first request is: an IP address of a federated learning node associated with the network connection to be optimized; and QoS requirements for network connections between federated learning nodes.

[0156] In one embodiment, the first request is: an identifier of the UE; and the QoS requirements of the UE's bearer flows.

[0157] In one embodiment, the first processing unit 802 is further configured to delete the received first request when federated learning is completed.

[0158] In practical application, the first receiving unit 801 and the first processing unit 802 can be realized by a combination of a processor and a communication interface in the federated learning device, or can be realized by a processor in the federated learning device.

[0159] Although the above embodiments have been described with reference to the division of each program module as an example when the associative learning device performs associative learning, in actual applications, the above processes can be completed by allocating them to different program modules as needed, i.e., the internal structure of the device can be divided into different program modules to complete all or part of the above-described processes. Furthermore, the associative learning device according to the above embodiments is based on the same concept as the associative learning method embodiments, and specific implementation processes can be achieved by referring to the method embodiments, and therefore will not be described here.

[0160] Based on the hardware implementation of the above program modules, in order to realize the method on the first network function side of the embodiment of the present disclosure, the embodiment of the present disclosure further provides a first network function, and as shown in FIG. 9, the first network function 900 includes a first communication interface 901 and a first processor 902.

[0161] The first communication interface 901 can exchange information with other network nodes; The first processor 902 is connected to the first communication interface 901, and is used to realize information interaction with other network nodes, and to execute the methods related to one or more technical solutions on the first network function side when executing a computer program, which is stored in the first memory 903.

[0162] Specifically, the first communication interface 901 is configured to send a first request to a network control device corresponding to the type of the federated learning node, and the first request is used to request optimization of the network connection between the federated learning nodes.

[0163] In one embodiment, the first communication interface 901 is specifically configured to send a first request to a network control device corresponding to the type of the federated learning node when the network connection quality between the federated learning nodes is below a predetermined threshold.

[0164] In one embodiment, the first request carries associated parameters for optimizing the network connection.

[0165] In one embodiment, when the federated learning node is a federated learning node in an operator network, the network control device is an SDN controller.

[0166] In one embodiment, when the federated learning node is an external node of the operator network, the network control device is a corresponding access node of the external network.

[0167] In one embodiment, when the federated learning node is a UE, the network control device is a PCF.

[0168] In one embodiment, the first request is: an IP address of a federated learning node associated with the network connection to be optimized; and QoS requirements for network connections between federated learning nodes.

[0169] In one embodiment, the first request is: an identifier of the UE; and the QoS requirements of the UE's bearer flows.

[0170] The specific processing steps of the first processor 902 and the first communication interface 901 can be understood by referring to the above method.

[0171] Of course, in actual application, each component in the first network function 900 is coupled via a bus system 904. It is understood that the bus system 904 is used to realize communication between these components. In addition to a data bus, the bus system 904 further includes a power bus, a control bus, and a status signal bus. However, for clarity, various buses are referred to as the bus system 904 in FIG. 9.

[0172] In embodiments of the present disclosure, the first memory 903 stores various types of data to support the operation of the first network function 900. Examples of this data include any computer programs for running on the first network function 900.

[0173] The method according to the embodiment of the present disclosure may be implemented in or by the first processor 902. The first processor 902 may be an integrated circuit chip having signal processing capabilities. In the implementation process, each step of the method may be completed by an integrated logic circuit in hardware or instructions in software form in the first processor 902. The first processor 902 may be a general-purpose processor, a digital signal processor (DSP), or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component, or the like. The first processor 902 may implement or execute each method, step, and logic block diagram according to the embodiment of the present disclosure. The general-purpose processor may be a microprocessor or any conventional processor, for example. The steps of the method according to the embodiment of the present disclosure may be combined and embodied as being executed directly by a hardware decoding processor or by a combination of hardware and software modules in the decoding processor. The software module can be located in a storage medium, which is located in the first memory 903, and the first processor 902 reads the information in the first memory 903 and combines its hardware to complete the steps of the above method.

[0174] In an example embodiment, the first network function 900 may be implemented by one or more Application Specific Integrated Circuits (ASICs), DSPs, Programmable Logic Devices (PLDs), Complex Programmable Logic Devices (CPLDs), Field-Programmable Gate Arrays (FPGAs), general-purpose processors, controllers, Micro Controller Units (MCUs), microprocessors, or other electronic elements to perform the above methods.

[0175] Based on the hardware implementation of the above program modules, in order to realize the method on the network control device side of the embodiment of the present disclosure, the embodiment of the present disclosure further provides a network control device, and as shown in FIG. 10, the network control device 1000 includes a second communication interface 1001 and a second processor 1002.

[0176] The second communication interface 1001 can exchange information with other network nodes; The second processor 1002 is connected to the second communication interface 1001, and is used to realize information interaction with other network nodes, and to execute the methods related to one or more technical solutions on the network control device side when executing a computer program, and the computer program is stored in the second memory 1003.

[0177] Specifically, the second communication interface 1001 is configured to receive a first request sent from a first network function, the first request is used to request optimizing the network connection between the federated learning nodes, and the type of the federated learning node and the network control device correspond to each other.

[0178] The second processor 1002 is configured to optimize a network connection based on the received first request.

[0179] In one embodiment, the first request carries associated parameters for optimizing the network connection.

[0180] In one embodiment, when the federated learning node is a federated learning node in an operator network, the network control device is an SDN controller.

[0181] In one embodiment, the second processor 1002 is specifically configured to calculate an optimal route based on the received first request and send the optimal route to an associated routing device, where the optimal route is used to transmit data between federated learning nodes.

[0182] In one embodiment, when the federated learning node is a UE, the network control device is a PCF.

[0183] In one embodiment, the second processor 1002 is specifically configured to send a PCC policy to the UPF via the SMF.

[0184] In one embodiment, the PCC policy is: Priority of resource allocation; Guaranteed speed and The maximum speed and and a maximum packet loss rate.

[0185] In one embodiment, when the federated learning node is an external node of the operator network, the network control device is a corresponding access node of the external network.

[0186] In one embodiment, the second processor 1002 is specifically configured to control a network control node corresponding to a corresponding access node of the external network to configure a related path.

[0187] In one embodiment, the first request is: an IP address of a federated learning node associated with the network connection to be optimized; and QoS requirements for network connections between federated learning nodes.

[0188] In one embodiment, the first request is: an identifier of the UE; and the QoS requirements of the UE's bearer flows.

[0189] In one embodiment, the second processor 1002 is further configured to delete the received first request when the federated learning is completed.

[0190] The specific processing steps of the second processor 1002 and the second communication interface 1001 can be understood by referring to the above method.

[0191] Of course, in actual application, each component in the network control device 1000 is coupled via a bus system 1004. It is understood that the bus system 1004 is used to realize communication between these components. The bus system 1004 includes a power bus, a control bus, and a status signal bus in addition to a data bus. However, for clarity, various buses are referred to as the bus system 1004 in FIG. 10.

[0192] In an embodiment of the present disclosure, the second memory 1003 stores various types of data to support the operation of the network control device 1000. Examples of this data include any computer programs for running on the network control device 1000.

[0193] The method according to the embodiment of the present disclosure may be implemented within or realized by the second processor 1002. The second processor 1002 may be an integrated circuit chip having signal processing capabilities. In the implementation process, each step of the method may be completed by an integrated logic circuit in hardware or instructions in software form within the second processor 1002. The second processor 1002 may be a general-purpose processor, a DSP, or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component, or the like. The second processor 1002 may implement or execute each method, step, and logic block diagram according to the embodiment of the present disclosure. The general-purpose processor may be a microprocessor or any conventional processor, for example. The steps of the method according to the embodiment of the present disclosure may be combined and embodied directly as execution by a hardware decoding processor or as execution by a combination of hardware and software modules within the decoding processor. The software module can be located in a storage medium, which is located in the second memory 1003, and the second processor 1002 reads the information in the second memory 1003 and combines its hardware to complete the steps of the above method.

[0194] In an exemplary embodiment, the network control device 1000 may be implemented by one or more ASICs, DSPs, PLDs, CPLDs, FPGAs, general-purpose processors, controllers, MCUs, microprocessors, or other electronic devices to perform the above methods.

[0195] It should be understood that the memory (first memory 903, second memory 1003) of the embodiments of the present disclosure may be volatile or nonvolatile memory, or may include both volatile and nonvolatile memory. Here, nonvolatile memory may be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), ferromagnetic random access memory (FRAM), flash memory, magnetic surface memory, optical disk, or compact disc read-only memory (CD-ROM). Magnetic surface memory may be magnetic disk memory or magnetic tape memory. Volatile memory may be random access memory (RAM) used as an external cache. By way of example and not limitation, many forms of RAM are available.For example, the memory may be a static random access memory (SRAM), a synchronous static random access memory (SSRAM), a dynamic random access memory (DRAM), a synchronous dynamic random access memory (SDRAM), a double data rate synchronous dynamic random access memory (DDRSDRAM), an enhanced synchronous dynamic random access memory (ESDRAM), a synchronous link dynamic random access memory (SLDRAM), or a direct memory bus random access memory (DRRAM). Memory as described in embodiments of the present disclosure is intended to include, but is not limited to, these and any other suitable types of memory.

[0196] In an exemplary embodiment, the embodiment of the present disclosure further provides a storage medium, which is a computer storage medium, specifically a computer-readable storage medium. For example, a first memory 903 may be provided for storing a computer program, which may be executed by the first processor 902 of the first network function 900 to complete the steps of the first network function-side method. Also, for example, a second memory 1003 may be provided for storing a computer program, which may be executed by the second processor 1002 of the network control device 1000 to complete the steps of the network control device-side method. The computer-readable storage medium may be a memory such as FRAM, ROM, PROM, EPROM, EEPROM, Flash Memory, magnetic surface memory, optical disk, or CD-ROM.

[0197] Note that "first," "second," etc. do not necessarily describe a particular order or priority, but are merely used to distinguish between similar objects.

[0198] The term "and / or" herein is simply a relation describing related objects and represents the presence of a triple relationship, for example, A and / or B can represent A being present alone, A and B being present together, and B being present alone. Also, the term "at least one" herein represents any one of a plurality or any combination of at least two of a plurality, for example, including at least one of A, B, and C can represent any one or more elements selected from the set consisting of A, B, and C.

[0199] In addition, the technical solutions described in the embodiments of the present disclosure may be arbitrarily combined if they do not conflict with each other.

[0200] The above descriptions are merely preferred embodiments of the present disclosure, and are not intended to limit the protection scope of the present disclosure.

Claims

1. A federated learning method applied to a first network function, comprising: A federated learning method, comprising: sending a first request to a network control device corresponding to a type of the federated learning node, the first request being used to request optimization of network connections between the federated learning nodes.

2. Sending a first request to a network control device corresponding to a type of the federated learning node includes: The method of claim 1 , further comprising: sending a first request to a network control device corresponding to a type of the federated learning node when a network connection quality between the federated learning nodes is below a predetermined threshold.

3. 3. The method of claim 1, wherein the first request carries associated parameters for optimizing the network connection.

4. The method according to claim 1 , wherein if the federated learning node is a federated learning node in an operator network, the network control device is a software-defined network SDN controller.

5. The method according to claim 1 , wherein if the federated learning node is an external node of the operator network, the network control device is a corresponding connection node of the external network.

6. The method according to any one of claims 1 to 3, wherein if the federated learning node is a terminal UE, the network control equipment is a Policy Control Function PCF.

7. The first request is: an Internet Protocol IP address of a federated learning node associated with the network connection to be optimized; 6. The method according to claim 3, further comprising at least one of: a quality of service (QoS) requirement for network connections between federated learning nodes;

8. The first request is: an identifier of the UE; and 7. The method of claim 6, wherein the QoS requirements of the UE's bearer flows are attached to the QoS requirements of the UE.

9. A federated learning method applied to a network control device, comprising: receiving a first request sent from a first network function, the first request being used to request optimizing network connections between federated learning nodes, and the type of the federated learning node and the network control device corresponding to each other; optimizing the network connection based on the received first request.

10. 10. The method of claim 9, wherein the first request carries associated parameters for optimizing the network connection.

11. 11. The method according to claim 9 or 10, wherein if the federated learning node is a federated learning node in an operator network, the network control device is an SDN controller.

12. optimizing a network connection based on the received first request 12. The method of claim 11, further comprising: calculating an optimal route based on the received first request; and transmitting the optimal route to associated routing equipment, wherein the optimal route is used to transmit data between federated learning nodes.

13. The method according to claim 9 or 10, wherein when the federated learning node is a UE, the network control device is a PCF.

14. optimizing a network connection based on the received first request 14. The method of claim 13, comprising sending policy control and charging PCC policies to a user plane function UPF via a session management function SMF.

15. The PCC policy is Priority of resource allocation; Guaranteed speed and The maximum speed and and a maximum packet loss rate.

16. 11. The method according to claim 9 or 10, wherein if the federated learning node is an external node of the operator network, the network control device is a corresponding connection node of the external network.

17. optimizing a network connection based on the received first request 17. The method of claim 16, comprising controlling a network control node corresponding to a corresponding connection node of the external network to perform configuration of an associated path.

18. The first request is: the IP address of the federated learning node associated with the network connection to be optimized; 18. The method of claim 10, wherein the method is accompanied by at least one of: a QoS requirement for network connections between federated learning nodes;

19. The first request is: an identifier of the UE; and 16. The method according to any one of claims 13 to 15, wherein the method is accompanied by at least one of the following: a QoS requirement of a bearer flow of the UE;

20. The method of claim 9 , further comprising deleting the received first request when federated learning is completed.

21. An associative learning device including a first transmitting unit, A federated learning device, wherein the first sending unit is configured to send a first request to a network control device corresponding to a type of federated learning node, and the first request is used to request optimization of network connections between the federated learning nodes.

22. An associative learning device including a first receiving unit and a first processing unit, the first receiving unit is configured to receive a first request sent from a first network function, the first request being used to request optimizing a network connection between federated learning nodes, and the type of the federated learning node and the network control device correspond to each other; The first processing unit is configured to optimize a network connection based on the received first request.

23. a first network function including a first processor and a first communication interface; A first network function, wherein the first communication interface is configured to send a first request to a network control device corresponding to a type of federated learning node, the first request being used to request optimization of network connections between the federated learning nodes.

24. a network control device including a second processor and a second communication interface, the second communication interface is configured to receive a first request sent from a first network function, the first request being used to request optimizing a network connection between the federated learning nodes, and the type of the federated learning node and the network control device correspond to each other; The second processor is configured to optimize a network connection based on the received first request.

25. a first network function including a first processor and a first memory for storing a computer program executable on the first processor, A first network function, wherein the first processor is configured to perform the steps of the method of any one of claims 1 to 8 when executing the computer program.

26. A network control device including a second processor and a second memory for storing a computer program executable by the second processor, 21. A network control appliance, wherein the second processor is configured to perform the steps of the method of any one of claims 9 to 20 when executing the computer program.

27. A storage medium having a computer program stored thereon, the computer program implementing the steps of the method according to any one of claims 1 to 8 or the steps of the method according to any one of claims 9 to 20 when executed by a processor.

Images