Federal learning training method and device, computer equipment and storage medium

Abstract

The invention relates to a federal learning training method and device, computer equipment and a storage medium. The method comprises the following steps: receiving a federal learning training application request sent by each federal node; sending a qualification authentication request of each federated node to an authentication cloud server according to the federated learning identity label and the federated learning qualification validity period of each federated node; determining a target federated node according to a qualification authentication result returned by the authentication cloud server; determining a federal learning deadline according to the federal learning time window of the target federal node; sending the federal learning deadline of the target federal node to the target federal node; the target federal node is used for participating in federal learning training within the federal learning deadline. According to the content, the compliance of all the target federated nodes is guaranteed, the safety of federated learning training is improved, the multi-source heterogeneous data in the metropolitan area network can be aggregated through federated learning training, and the generalization ability and accuracy of the federated learning model are improved.

Description

Technical Field

[0001] This application relates to the field of network and security technology, and in particular to a federated learning training method, apparatus, computer equipment, and storage medium. Background Technology

[0002] A metropolitan area network (MAN) is a network built within a specific area (such as a city) for the purpose of user access, traffic aggregation, and service provision. MAN coverage typically ranges from a few kilometers to tens of kilometers, usually covering a city or a specific area within a city, and connecting various government departments, businesses, schools, and other institutions within the city.

[0003] Metropolitan area networks contain a large number of dispersed institutions, enterprises, and terminals. Their data is characterized by heterogeneity, dispersion, and high sensitivity. In other words, massive amounts of heterogeneous data are scattered among different entities such as government departments, financial institutions, hospitals, enterprises, and individuals. This data involves public safety (such as traffic monitoring and government data), trade secrets (such as financial transaction data and enterprise operation data), and personal privacy (such as medical records and consumption data). Federated learning training is needed to maximize the value of this data.

[0004] However, among related technologies, the security of federated learning training is not high. Summary of the Invention

[0005] Therefore, it is necessary to provide a federated learning training method, apparatus, computer equipment, computer-readable storage medium, and computer program product that can improve the security of federated learning training in response to the above-mentioned technical problems.

[0006] Firstly, this application provides a federated learning training method applied to a supervisory cloud server in a central cloud, which also includes an authentication cloud server. The central cloud is communicatively connected to at least one federated node via a metropolitan area network, including:

[0007] Receive federated learning and training application requests from each federated node; the federated learning and training application requests include each federated node's federated learning identity, the validity period of its federated learning qualification, and the federated learning time window;

[0008] Based on the federated learning identity and the validity period of the federated learning qualification of each federated node, a qualification authentication request for each federated node is sent to the authentication cloud server; the authentication cloud server is used to authenticate the identity and validity period of each federated node and obtain the qualification authentication result.

[0009] Based on the qualification certification results returned by the certification cloud server, the target federated node is determined among the federated nodes; based on the federated learning time window of the target federated node, the federated learning period of the target federated node is determined; the target federated node is the federated node whose qualification certification results indicate that both identity authentication and validity period authentication have passed.

[0010] Send the target federated learning period to the target federated node; the target federated node is used to participate in federated learning training within the federated learning period.

[0011] In one embodiment, the federation learning period of the target federation node is determined based on the federation learning time window of the target federation node, including:

[0012] Obtain the federated learning time window for each target federated node;

[0013] The common time window is obtained by intersecting the federated learning time windows of each target federated node.

[0014] The public time window is used as the federal learning period for each target federated node.

[0015] In one embodiment, each federated node includes a first federated node, which is a federated node in the edge cloud; receiving federated learning training request requests sent by each federated node includes:

[0016] Receive the federated learning and training request sent by the first federated node through the first communication link;

[0017] The first communication link is formed by sequentially connecting the first federated node, the first POP component, the leaf device of the first POD component, the Spine device of the first POD component, the second POP component, and the regulatory cloud server.

[0018] In one embodiment, each federated node includes a second federated node, which is not a federated node in the edge cloud; receiving federated learning training request requests sent by each federated node includes:

[0019] Receive the federated learning and training request from the second federated node sent via the second communication link;

[0020] The second communication link is formed by sequentially connecting the second federated node, the first CPE device, the first OLT device, the leaf device of the first POD component, the Spine device of the first POD component, the second POP component, and the regulatory cloud server.

[0021] In one embodiment, each federated node further includes a third federated node, which is not a federated node in the edge cloud and has a different location in the metropolitan area network than the second federated node; receiving federated learning training request requests sent by each federated node includes:

[0022] Receive federated learning training request requests from third federated nodes sent via a third communication link;

[0023] The third communication link is formed by sequentially connecting the third federated node, the second CPE device, the second OLT device, the leaf device of the second POD component, the Spine device of the second POD component, the Super Spine device of the export functional area component, the Spine device of the first POD component, the second POP component, and the regulatory cloud server.

[0024] In one embodiment, the authentication cloud server is used to receive qualification registration application requests sent by each federated node, to conduct qualification review of each federated node, and after the qualification review is passed, to generate federated learning identity identifiers and federated learning qualification validity periods for each federated node, and to send the federated learning identity identifiers and federated learning qualification validity periods for each federated node to each federated node.

[0025] In one embodiment, the federated learning training method further includes:

[0026] Receive local model parameters sent by each target federated node; the local model parameters are obtained by the target federated node training the initial model parameters of the federated learning model based on its own training samples;

[0027] Based on the local model parameters of each target federated node, global model parameters are generated and sent to each target federated node; each target federated node updates its local model parameters of the federated learning model based on the global model parameters.

[0028] The process involves receiving updated local model parameters from each target federated node, generating global model parameters based on the local model parameters of each target federated node, and sending the global model parameters to each target federated node. This process continues until the training termination condition is met, resulting in a trained federated learning model.

[0029] Secondly, this application also provides a federated learning training device for a supervisory cloud server in a central cloud, the central cloud further including an authentication cloud server, the central cloud being communicatively connected to at least one federated node via a metropolitan area network, comprising:

[0030] The training application request receiving module is used to receive federated learning training application requests sent by each federated node. The federated learning training application request includes the federated learning identity identifier, the validity period of the federated learning qualification, and the federated learning time window of each federated node.

[0031] The qualification authentication request sending module is used to send qualification authentication requests from each federated node to the authentication cloud server based on the federated learning identity identifier and the validity period of the federated learning qualification. The authentication cloud server is used to authenticate the identity and validity period of each federated node and obtain the qualification authentication result.

[0032] The target federated node determination module is used to determine the target federated node among the federated nodes based on the qualification certification results returned by the authentication cloud server; and to determine the federated learning period of the target federated node based on the federated learning time window of the target federated node; the target federated node is a federated node whose qualification certification results indicate that both identity authentication and validity period authentication have passed.

[0033] The Federated Learning Period Sending Module is used to send the Federated Learning Period of the target Federated Node to the target Federated Node; the target Federated Node is used to participate in Federated Learning Training within the Federated Learning Period.

[0034] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method steps of the first aspect.

[0035] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method steps of the first aspect.

[0036] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, implements the method steps of the first aspect.

[0037] The aforementioned federated learning training method, apparatus, computer equipment, computer-readable storage medium, and computer program product receive federated learning training application requests from each federated node. These requests include the federated learning identity identifier, validity period of the federated learning qualification, and federated learning time window for each federated node. Based on the federated learning identity identifier and validity period of each federated node, a qualification authentication request is sent to an authentication cloud server. The authentication cloud server performs identity authentication and validity period authentication on each federated node, obtaining a qualification authentication result. Based on the qualification authentication result returned by the authentication cloud server, a target federated node is determined. The federated learning period for the target federated node is determined based on its federated learning time window. The target federated node is defined by the qualification authentication result, indicating that both identity authentication and validity period authentication have passed. The federated learning period for the target federated node is sent to the target federated node. The target federated node participates in federated learning training within the federated learning period. As described above, this application uses a monitoring cloud server to parse the federated learning training application requests of each federated node, generating qualification authentication requests for each node. An authentication cloud server then verifies the identity and validity of these requests, identifying multiple target federated nodes eligible to participate in the federated learning training. This ensures the compliance of each target node and improves the security of the training. Furthermore, based on the federated learning time windows of each target node, the application determines the federated learning period for all nodes. Each target node participates in the training within this period, ensuring that all participants complete their training within a unified timeframe. Federated learning training in a metropolitan area network (MAN) can aggregate multi-source heterogeneous data within the MAN, enabling the trained federated learning model to learn a wider range of data features, thereby improving its generalization ability and accuracy. Attached Figure Description

[0038] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0039] Figure 1 This is a diagram illustrating the application environment of a federated learning training method in one embodiment.

[0040] Figure 2 This is a flowchart illustrating a federated learning training method in one embodiment;

[0041] Figure 3This is a schematic diagram of the training application process for participants in a metropolitan area network federated learning, as shown in one embodiment.

[0042] Figure 4 This is a schematic diagram of the pre-registration process for participants in a metropolitan area network federated learning, as shown in one embodiment.

[0043] Figure 5 This is a schematic diagram of a single POD metropolitan area network federated learning scenario in one embodiment;

[0044] Figure 6 This is a schematic diagram of a multi-POD metropolitan area network federated learning scenario in one embodiment;

[0045] Figure 7 This is a schematic diagram of the metropolitan area network federated learning and training process in one embodiment;

[0046] Figure 8 This is a structural block diagram of a federated learning training device in one embodiment;

[0047] Figure 9 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0048] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0049] It should be noted that the terms "first," "second," etc., used in this application can be used to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish the first element from the second element. The terms "comprising" and "having," and any variations thereof, used in this application, are intended to cover non-exclusive inclusion. The term "multiple" used in this application refers to two or more. The term "and / or" used in this application refers to one of the embodiments, or any combination of multiple embodiments.

[0050] The federated learning training method provided in this application embodiment can be applied to, for example, Figure 1In the application environment shown, at least one federated node 102 communicates with the supervisory cloud server 104 and the authentication cloud server 106 in the central cloud via a metropolitan area network. The supervisory cloud server 104 receives federated learning training application requests from each federated node 102; these requests include the federated learning identity identifier, validity period of the federated learning qualification, and federated learning time window of each federated node 102; based on the federated learning identity identifier and validity period of each federated node 102, it sends a qualification authentication request to the authentication cloud server 106; the authentication cloud server 106 performs identity authentication and validity period authentication on each federated node 102, obtaining a qualification authentication result; based on the qualification authentication result returned by the authentication cloud server 106, it determines the target federated node among the federated nodes 102; based on the federated learning time window of the target federated node, it determines the federated learning period of the target federated node; the target federated node is the federated node whose qualification authentication result indicates that both identity authentication and validity period authentication have passed; the authentication cloud server 104 sends the federated learning period of the target federated node to the target federated node; the target federated node participates in federated learning training within the federated learning period. Federation node 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, drones, low-altitude aircraft, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, smart in-vehicle devices, and projection devices. Portable wearable devices can include smartwatches, smart bracelets, and head-mounted displays. Head-mounted displays can be virtual reality (VR) devices, augmented reality (AR) devices, and smart glasses. Supervisory cloud server 104 and authentication cloud server 106 can be cloud servers providing cloud computing services.

[0051] In one embodiment, such as Figure 2 As shown, a federated learning training method is provided. This embodiment illustrates the application of this method to a supervisory cloud server in a central cloud. The central cloud also includes an authentication cloud server, and the central cloud communicates with at least one federated node via a metropolitan area network. In this embodiment, the method includes the following steps:

[0052] Step S210: Receive the federated learning training application requests sent by each federated node; the federated learning training application requests include the federated learning identity identifier of each federated node, the validity period of the federated learning qualification, and the federated learning time window.

[0053] Centralized cloud refers to a centralized data center that provides large-scale computing, storage, and network resources, and is suitable for non-real-time, long-cycle data processing and complex business decision-making scenarios.

[0054] The central cloud deploys a monitoring cloud server and an authentication cloud server. The monitoring cloud server is mainly responsible for aggregating and distributing model parameters for the federated learning model, while the authentication cloud server is mainly responsible for the qualification authentication of federated nodes, including identity authentication and validity period authentication.

[0055] The federated learning identity is used to identify and verify the identity of federated nodes during federated learning training. Each federated node's federated learning identity must be globally unique. Specifically, the federated learning identity can be a string.

[0056] The validity period of a federated learning qualification refers to the valid time during which a federated node is eligible for federated learning training. For example, the validity period of a federated learning qualification may be one week or one month.

[0057] The federated learning time window is determined by the federated nodes themselves and refers to the time period during which the federated nodes participate in federated learning training. For example, the federated learning time window is from March 1, 2025 to March 5, 2025.

[0058] In this embodiment, each federated node stores its own federated learning identity ID and federated learning qualification validity period P, and designates a federated learning time window T. Each federated node generates a federated learning training application request based on its own federated learning identity ID, federated learning qualification validity period P, and designated federated learning time window T, and sends the federated learning training application request to the monitoring cloud server in the central cloud via the metropolitan area network.

[0059] Step S220: Based on the federated learning identity identifier and the validity period of the federated learning qualification of each federated node, send the qualification authentication request of each federated node to the authentication cloud server; the authentication cloud server is used to authenticate the identity and validity period of each federated node and obtain the qualification authentication result.

[0060] The qualification certification result includes qualification certification passed and qualification certification failed. Qualification certification passed means that both identity certification and validity period certification are passed. Qualification certification failed means that identity certification and / or validity period certification are failed.

[0061] In this embodiment, the monitoring cloud server parses the federated learning training application requests sent by each federated node to obtain the federated learning identity identifier, the validity period of the federated learning qualification, and the federated learning time window of each federated node. Based on the federated learning identity identifier and the validity period of the federated learning qualification of each federated node, a qualification authentication request for each federated node is generated and sent to the authentication cloud server.

[0062] The authentication cloud server authenticates each federated node based on its federated learning identity identifier, obtaining an authentication result. This result includes successful and unsuccessful authentication. The authentication cloud server also authenticates each federated node based on its federated learning eligibility validity period, obtaining a validity period authentication result. This result includes successful and unsuccessful validity period authentication.

[0063] The authentication cloud server sends the qualification authentication results of each federated node to the regulatory cloud server.

[0064] Step S230: Based on the qualification certification results returned by the certification cloud server, determine the target federated node among the federated nodes; based on the federated learning time window of the target federated node, determine the federated learning period of the target federated node; the target federated node is the federated node whose qualification certification results represent that both identity authentication and validity period authentication have passed.

[0065] In this embodiment, the monitoring cloud server receives the qualification certification results from each federated node. If the qualification certification of a federated node fails, the federated learning training application process is terminated, and the federated node is notified. Conversely, if the qualification certification of a federated node passes, the monitoring cloud server will aggregate the federated learning time windows T of all federated nodes (target federated nodes) that have initiated federated learning training applications and passed qualification certification, and calculate a federated learning period T0 that is feasible for all federated nodes.

[0066] Step S240: Send the target federated learning period to the target federated node; the target federated node is used to participate in federated learning training within the federated learning period.

[0067] In this embodiment of the application, the supervisory cloud server sends the federated learning period T0 to each qualified federated node (target federated node), and each target federated node participates in federated learning training within the federated learning period T0.

[0068] For a better understanding of steps S210~S240 above, please refer to [link / reference]. Figure 3This document provides a schematic diagram of the application process for federated learning participants in a metropolitan area network (MAN). The process involves a participant submitting a request to the supervisory cloud server C, which includes their federated learning identity ID, the validity period of their federated learning qualification P, and the federated learning time window T. The federated learning time window T is determined by the participant and represents the time period for which they will participate in the federated learning. The supervisory cloud server C receives the submitted federated learning identity ID and validity period P and initiates a qualification authentication request to the authentication cloud server A. The authentication cloud server A queries its database; if it finds that the participant's federated learning identity ID is already registered and their qualification is valid, the authentication is successful; otherwise, it fails. The authentication cloud server A sends the authentication result back to the supervisory cloud server C. The supervisory cloud server C receives the authentication result; if the participant's authentication fails, the application process is terminated, and the participant is notified. Conversely, if a participant in the federated learning program passes the qualification assessment, the supervisory cloud server C will aggregate the federated learning time windows T of all federated learning participants who have initiated training applications and passed the qualification assessment, and calculate a federated learning period T0 that is feasible for all participants. The supervisory cloud server C will then send the federated learning period T0 to each qualified federated learning participant.

[0069] The aforementioned federated learning training method involves receiving federated learning training application requests from each federated node. These requests include the federated learning identity identifier, validity period of the federated learning qualification, and federated learning time window for each node. Based on the federated learning identity identifier and validity period of each node, a qualification authentication request is sent to an authentication cloud server. The authentication cloud server performs identity authentication and validity period authentication on each federated node, obtaining the qualification authentication result. Based on the qualification authentication result returned by the authentication cloud server, a target federated node is determined. The federated learning period for the target federated node is determined based on its federated learning time window. The target federated node is defined by the qualification authentication result indicating that both identity authentication and validity period authentication have passed. The federated learning period for the target federated node is sent to the target federated node. The target federated node participates in federated learning training within the specified period. As described above, this application uses a monitoring cloud server to parse the federated learning training application requests of each federated node, generating qualification authentication requests for each node. An authentication cloud server then verifies the identity and validity of these requests, identifying multiple target federated nodes eligible to participate in the federated learning training. This ensures the compliance of each target node and improves the security of the training. Furthermore, based on the federated learning time windows of each target node, the application determines the federated learning period for all nodes. Each target node participates in the training within this period, ensuring that all participants complete their training within a unified timeframe. Federated learning training in a metropolitan area network (MAN) can aggregate multi-source heterogeneous data within the MAN, enabling the trained federated learning model to learn a wider range of data features, thereby improving its generalization ability and accuracy.

[0070] In one embodiment, the authentication cloud server is used to receive qualification registration application requests sent by each federated node, conduct qualification review of each federated node, generate federated learning identity identifiers and federated learning qualification validity periods for each federated node after the qualification review is passed, and send the federated learning identity identifiers and federated learning qualification validity periods for each federated node to each federated node.

[0071] In this embodiment of the application, before each federated node sends a federated learning and training application request to the supervisory cloud server, each federated node needs to complete the pre-registration process for federated learning and training.

[0072] Specifically, please refer to Figure 4This paper presents a schematic diagram of the pre-registration process for participants in a metropolitan area network (MAN) federated learning program. Each federated node submits a registration request for federated learning training eligibility to an authentication cloud server. The authentication cloud server verifies the eligibility of each federated node; if the verification is successful, the registration application is approved, generating a globally unique federated learning identity ID and a federated learning eligibility validity period P specific to that node. The authentication cloud server then sends the federated learning identity ID and the validity period P back to the federated node and stores these two identifiers in its own database. Each federated node receives and stores its own federated learning identity ID and the validity period P.

[0073] This application embodiment uses an authentication cloud server to perform identity authentication and validity period authentication on the qualification authentication requests of each federated node, thereby determining multiple target federated nodes that can participate in federated learning and training, and ensuring the compliance of the target federated nodes.

[0074] In one embodiment, each federation node includes a first federation node, which is a federation node in the edge cloud.

[0075] Receive federated learning and training request requests from each federated node, including:

[0076] Step S212: Receive the federated learning training request from the first federated node sent through the first communication link;

[0077] The first communication link is formed by sequentially connecting the first federated node, the first POP component, the leaf device of the first POD component, the Spine device of the first POD component, the second POP component, and the regulatory cloud server.

[0078] Edge cloud refers to the deployment of computing power to edge nodes closer to users or terminal devices, primarily used in scenarios with high real-time requirements, short-cycle data processing, and localized decision-making.

[0079] The POD (Point of Delivery) component is a metropolitan area network matrix that enables the carrying of fixed-line, mobile, and cloud services. It adopts a Spine-Leaf networking architecture, with Spine and Leaf devices deployed in pairs within the POD component. This allows for 2H / 2B / 2C user access and traffic routing to / from the cloud. The configuration of the POD component is based on factors such as the actual user scale, service scale, and edge cloud deployment within the metropolitan area network.

[0080] The Spine device is the core backbone switch of the POP component, responsible for high-speed forwarding of traffic between Leaf devices, enabling full Mesh interconnection. The Leaf device is the service access switch of the POP component, responsible for the access and traffic aggregation of terminal devices (such as home broadband, enterprise leased lines, and IoT devices).

[0081] Among them, the POP (Point of Presence) component enables the POD component to connect with the cloud resource pool, and through dedicated Leaf devices, it enables rapid access to central cloud, edge cloud, third-party cloud, etc.

[0082] In the embodiments of this application, please refer to Figure 5 This paper presents a schematic diagram of a single POD metropolitan area network federated learning scenario. Specifically, the first communication link between federated learning participant A (the first federated node) and the central cloud is as follows: Federated learning participant A (first federated node) - POP (first POP component) - POD 1 Leaf (leaf device of the first POD component) - POD1 Spine (spine device of the first POD component) - POP (second POP component) - central cloud. Based on this first communication link, the first federated node sends the federated learning training request to the monitoring cloud server in the central cloud.

[0083] This application embodiment establishes a communication link between the first federated node and the regulatory cloud server through the joint cooperation of the first POP component, the first POD component, and the second POP component, thereby realizing the communication connection between the first federated node and the regulatory cloud server.

[0084] In one embodiment, each federation node includes a second federation node that is not a federation node in the edge cloud.

[0085] Receive federated learning and training request requests from each federated node, including:

[0086] Step S214: Receive the federated learning training request from the second federated node sent through the second communication link;

[0087] The second communication link is formed by sequentially connecting the second federated node, the first CPE device, the first OLT device, the leaf device of the first POD component, the Spine device of the first POD component, the second POP component, and the regulatory cloud server.

[0088] CPE (Customer Premises Equipment), also known as client equipment, refers to devices located at the client's location that can access network services. These devices include routers, modems, and wireless gateways, which can convert broadband signals (such as 4G, 5G, and fiber optic) into local signals suitable for devices such as computers and smart appliances.

[0089] Among them, the OLT (Optical Line Terminal) equipment is located at the core of the optical fiber access network. It is responsible for the conversion of optical signals to electrical signals and manages the access, control and data transmission of user-end equipment (such as ONU / ONT).

[0090] In the embodiments of this application, please refer to Figure 5 The second communication link between Federated Learning Participant B (the second federated node) and the central cloud is as follows: Federated Learning Participant B (the second federated node) - CPE (the first CPE device) - POD 1 OLT (the first OLT device) - POD 1 Leaf (the leaf device of the first POD component) - POD 1 Spine (the Spine device of the first POD component) - POP (the second POP component) - Central Cloud. Based on this second communication link, the second federated node sends the federated learning training request to the supervisory cloud server in the central cloud.

[0091] This application embodiment establishes a communication link between the second federated node and the regulatory cloud server through the joint cooperation of the first CPE device, the first OLT device, the first POD component, and the second POP component, thus realizing the communication connection between the second federated node and the regulatory cloud server.

[0092] In one embodiment, each federated node also includes a third federated node, which is not a federated node in the edge cloud and is located in a different position in the metropolitan area network than the second federated node.

[0093] Receive federated learning and training request requests from each federated node, including:

[0094] Step S216: Receive the federated learning training request from the third federated node sent through the third communication link.

[0095] The third communication link is formed by sequentially connecting the third federated node, the second CPE device, the second OLT device, the leaf device of the second POD component, the Spine device of the second POD component, the Super Spine device of the export functional area component, the Spine device of the first POD component, the second POP component, and the regulatory cloud server.

[0096] The export functional area components include Super Spine and B-Leaf devices, which are responsible for connecting to the backbone network, IDC network, business platform / core network and other third-party networks, and uniformly guiding the traffic between multiple PODs as well as the traffic entering and leaving the backbone network and IDC.

[0097] In the embodiments of this application, please refer to Figure 6 This paper presents a schematic diagram of a multi-POD metropolitan area network (MAN) federated learning scenario. Similar to the single-POD MAN federated learning scenario, both POD 1 and POD 2 connect to the central cloud and edge cloud via POPs, and each POD's central cloud deploys a supervisory cloud server C and an authentication cloud server A, respectively. However, in a real multi-POD MAN federated learning scenario, to avoid chaotic interaction processes between network elements, only one set of supervisory cloud server C and authentication cloud server A with the strongest overall performance in traffic forwarding and data computation will be activated. As an example, Figure 6 The demonstration shows a scenario where the supervisory cloud server C and the authentication cloud server A in POD 1 are enabled, while in POD 2, the supervisory cloud server C and the authentication cloud server A are in a dormant state and not active. Figure 6 The information is provided in the text.

[0098] The third communication link between Federated Learning Participant C (the third federated node) and the central cloud is as follows: Federated Learning Participant C (the third federated node) - CPE (the second CPE device) - POD 2 OLT (the second OLT device) - POD 2 Leaf (the leaf device of the second POD component) - POD 2 Spine (the Spine device of the second POD component) - Super Spine (the Super Spine device of the export functional area component) - POD 1 Spine (the Spine device of the first POD component) - POP (the second POP component) - Central Cloud. Based on this third communication link, the third federated node sends the federated learning training request to the supervisory cloud server in the central cloud.

[0099] This application embodiment establishes a communication link between the third federation node and the regulatory cloud server through the joint cooperation of the second CPE device, the second OLT device, the second POD component, the export functional area component, the first POD component, and the second POP component, thus realizing the communication connection between the third federation node and the regulatory cloud server.

[0100] In one embodiment, determining the federation learning period of the target federation node based on its federation learning time window includes:

[0101] Step S232: Obtain the federated learning time window for each target federated node.

[0102] In this embodiment, each target federated node specifies its own federated learning time window. For example, target federated node 1 specifies federated learning window T1, target federated node 2 specifies federated learning window T2, ..., target federated node n specifies federated learning window Tn.

[0103] Step S234: Intersect the federated learning time windows of each target federated node to obtain a common time window.

[0104] In this embodiment, assuming the federated learning window T1 specified by target federated node 1 is 03:01~03:05, and the federated learning window T2 specified by target federated node 2 is 03:03~03:07, then the common time window of target federated node 1 and target federated node 2 is 03:03~03:05. The common time window of all target federated nodes is obtained by performing the intersection processing described above on the federated learning time windows of all target federated nodes.

[0105] Step S236: Use the public time window as the federal learning period for each target federated node.

[0106] In this embodiment of the application, by determining the common time window of all target federated nodes, the common time window is used as the federated learning period of each target federated node, so that each target federated node can participate in federated learning training within the same time period.

[0107] In one embodiment, the federated learning training method further includes:

[0108] Step S250: Receive the local model parameters sent by each target federated node; the local model parameters are obtained by the target federated node training the initial model parameters of the federated learning model based on its own training samples.

[0109] For specific application scenarios, different federated learning models can be trained. Specifically, in the medical field, different hospitals possess their own patients' medical records, examination reports, and other private data. Through metropolitan area network (MAN) federated learning, multiple hospitals can jointly train disease diagnosis and prediction models without sharing the original patient data. In the financial field, banks, payment platforms, and consumer finance companies hold user transaction data across various dimensions, such as transfer records, consumption locations, and credit delinquency status. This data contains a large amount of user privacy information. Through MAN federated learning, financial institutions can train anti-fraud models locally using their own data.

[0110] In this embodiment of the application, each target federated node trains the federated learning model locally based on its own training samples and sends the trained local model parameters to the monitoring cloud server.

[0111] Step S260: Generate global model parameters based on the local model parameters of each target federated node, and send the global model parameters to each target federated node; each target federated node updates the local model parameters of the federated learning model based on the global model parameters.

[0112] In this embodiment, the monitoring cloud server averages or weights the local model parameters of each target federated node to obtain the global model parameters. The monitoring cloud server then sends the global model parameters to each target federated node.

[0113] Each target federated node calculates the loss function value based on the global model parameters, and then performs gradient descent and backpropagation on the global model parameters based on the loss function value to obtain the updated local model parameters.

[0114] Step S270: Receive the updated local model parameters sent by each target federated node, return the steps of generating global model parameters based on the local model parameters of each target federated node, and send the global model parameters to each target federated node, until the training termination condition is met, and obtain the trained federated learning model.

[0115] Among them, the conditions for ending training include, but are not limited to, the convergence of the loss function and the completion of the training time for the federated learning period.

[0116] In this embodiment, each target federated node sends the updated local model parameters to the monitoring cloud server. The monitoring cloud server recalculates the new global model parameters, distributes the new global model parameters to each target federated node, and continues to update the local model parameters of each target federated node until the training termination condition is met, thus obtaining the trained federated learning model.

[0117] For a better understanding of steps S250~S270 above, please refer to [link / reference]. Figure 7 This paper presents a schematic diagram of a metropolitan area network (MAN) federated learning training process. Each qualified federated learning participant initiates local training based on its own samples at the start of the federated learning period T0, generating local training parameters. Following the bidirectional interaction path between the federated learning participant and the central cloud, the participant encrypts and uploads its local training parameters to the monitoring cloud server C. The monitoring cloud server C aggregates the local training parameters to generate global training parameters. Following the bidirectional interaction path between the federated learning participant and the central cloud, the monitoring cloud server C distributes the global training parameters to each participant. The next round of MAN federated learning iterations begins, continuing until the loss function calculated by each participant converges, or the federated learning period T0 terminates.

[0118] The above-mentioned federated learning training method can be applied to both single-POP and multi-POP metropolitan area network federated learning scenarios, and has the following beneficial effects:

[0119] 1. Improve model performance and generalization ability: Metropolitan area networks (MANs) have a wide coverage area, and the data from different regions and institutions are diverse and heterogeneous. MAN federated learning can aggregate multi-source heterogeneous data within the MAN. Each participant trains its model locally, and the monitoring cloud server integrates the model parameters from all parties through a specific algorithm, enabling the final model to learn a wider range of data features, thereby improving the model's generalization ability and accuracy.

[0120] 2. Protecting Data Privacy and Security: In a metropolitan area network environment, many participants, such as government departments, enterprises, and medical institutions, hold a large amount of sensitive data. Federated learning adopts a localized training and parameter exchange mechanism. Each participant trains its model locally using its own data, and then uploads the model parameters (such as gradients or weights) to the monitoring cloud server or exchanges them with other participants. The original data does not leave the local machine throughout the process, greatly reducing the risk of data leakage. In addition, the pre-registration and qualification authentication mechanism of the authentication server ensures the compliance of each federated learning participant.

[0121] 3. Optimize resource utilization and reduce costs: In metropolitan area networks, traditional centralized learning methods require a large amount of raw data to be transmitted to the monitoring cloud server for training. This not only places extremely high demands on network bandwidth and incurs high communication costs, but may also lead to data transmission delays. In the federated learning model, each participant completes model training locally, only needing to upload a small number of model parameters, significantly reducing the amount of data transmission and saving network bandwidth resources and communication costs.

[0122] It should be understood that although the steps in the flowcharts of the above embodiments are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the above embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages in other steps. It is understood that the steps in different embodiments can be freely combined as needed, and all non-contradictory solutions formed by such combinations are within the scope of protection of this application.

[0123] Based on the same inventive concept, this application also provides a federated learning training apparatus for implementing the federated learning training method described above. The solution provided by this apparatus is similar to the solution described in the above method; therefore, the specific limitations in one or more federated learning training apparatus embodiments provided below can be found in the limitations of the federated learning training method described above, and will not be repeated here.

[0124] In one exemplary embodiment, please refer to Figure 8 A federated learning training apparatus is provided for a supervisory cloud server in a central cloud, the central cloud further including an authentication cloud server, the central cloud being communicatively connected to at least one federated node via a metropolitan area network, comprising:

[0125] The training application request receiving module 810 is used to receive federated learning training application requests sent by each federated node; the federated learning training application request includes the federated learning identity identifier, the validity period of the federated learning qualification, and the federated learning time window of each federated node;

[0126] The qualification authentication request sending module 820 is used to send qualification authentication requests from each federated node to the authentication cloud server based on the federated learning identity identifier and the validity period of the federated learning qualification of each federated node; the authentication cloud server is used to perform identity authentication and validity period authentication on each federated node to obtain the qualification authentication result.

[0127] The target federated node determination module 830 is used to determine the target federated node among the federated nodes based on the qualification certification results returned by the certification cloud server; and to determine the federated learning period of the target federated node based on the federated learning time window of the target federated node; the target federated node is a federated node whose qualification certification results represent that both identity authentication and validity period authentication have passed.

[0128] The Federated Learning Period Sending Module 840 is used to send the Federated Learning Period of the target Federated Node to the target Federated Node; the target Federated Node is used to participate in Federated Learning Training within the Federated Learning Period.

[0129] In one embodiment, determining the federation learning period of the target federation node based on its federation learning time window includes:

[0130] Obtain the federated learning time window for each target federated node;

[0131] The common time window is obtained by intersecting the federated learning time windows of each target federated node.

[0132] The public time window is used as the federal learning period for each target federated node.

[0133] In one embodiment, each federated node includes a first federated node, which is a federated node in the edge cloud; receiving federated learning training request requests sent by each federated node includes:

[0134] Receive the federated learning and training request sent by the first federated node through the first communication link;

[0135] The first communication link is formed by sequentially connecting the first federated node, the first POP component, the leaf device of the first POD component, the Spine device of the first POD component, the second POP component, and the regulatory cloud server.

[0136] In one embodiment, each federated node includes a second federated node, which is not a federated node in the edge cloud; receiving federated learning training request requests sent by each federated node includes:

[0137] Receive the federated learning and training request from the second federated node sent via the second communication link;

[0138] The second communication link is formed by sequentially connecting the second federated node, the first CPE device, the first OLT device, the leaf device of the first POD component, the Spine device of the first POD component, the second POP component, and the regulatory cloud server.

[0139] In one embodiment, each federated node further includes a third federated node, which is not a federated node in the edge cloud and has a different location in the metropolitan area network than the second federated node; receiving federated learning training request requests sent by each federated node includes:

[0140] Receive federated learning training request requests from third federated nodes sent via a third communication link;

[0141] The third communication link is formed by sequentially connecting the third federated node, the second CPE device, the second OLT device, the leaf device of the second POD component, the Spine device of the second POD component, the Super Spine device of the export functional area component, the Spine device of the first POD component, the second POP component, and the regulatory cloud server.

[0142] In one embodiment, the authentication cloud server is used to receive qualification registration application requests sent by each federated node, conduct qualification review of each federated node, generate federated learning identity identifiers and federated learning qualification validity periods for each federated node after the qualification review is passed, and send the federated learning identity identifiers and federated learning qualification validity periods for each federated node to each federated node.

[0143] In one embodiment, the federated learning training method further includes:

[0144] Receive local model parameters sent by each target federated node; the local model parameters are obtained by the target federated node training the initial model parameters of the federated learning model based on its own training samples;

[0145] Based on the local model parameters of each target federated node, global model parameters are generated and sent to each target federated node; each target federated node updates its local model parameters of the federated learning model based on the global model parameters.

[0146] The process involves receiving updated local model parameters from each target federated node, generating global model parameters based on the local model parameters of each target federated node, and sending the global model parameters to each target federated node. This process continues until the training termination condition is met, resulting in a trained federated learning model.

[0147] The modules in the aforementioned federated learning training device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can invoke and execute the operations corresponding to each module.

[0148] In one exemplary embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 9As shown, the computer device includes a processor, memory, input / output interfaces, a communication interface, a display unit, and an input device. The processor, memory, and input / output interfaces are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The input / output interfaces are used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, Near Field Communication (NFC), or other technologies. When the computer program is executed by the processor, it implements a federated learning training method. The display unit is used to form a visually visible image and can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.

[0149] Those skilled in the art will understand that Figure 9 The structures shown are merely block diagrams of some structures related to the present application and do not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than shown in the figures, or combine certain components, or have different component arrangements. In one embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, which, when executed by the processor, causes the processor to perform the steps of the aforementioned federated learning training method. The steps of the federated learning training method described here may be steps from one of the federated learning training methods in the various embodiments described above.

[0150] In one embodiment, a computer-readable storage medium is provided, storing a computer program that, when executed by a processor, causes the processor to perform the steps of the federated learning training method described above. The steps of the federated learning training method described here may be steps from one of the federated learning training methods in the various embodiments described above.

[0151] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, causes the processor to perform the steps of the federated learning training method described above. The steps of the federated learning training method described here may be steps from one of the federated learning training methods in the various embodiments described above.

[0152] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0153] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0154] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0155] The above embodiments are merely illustrative of several implementation methods of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A federated learning training method, characterized in that, The method is applied to a supervision cloud server in a center cloud, the center cloud further comprises an authentication cloud server, the center cloud is in communication connection with at least one federal node through a metropolitan area network, and the method comprises the following steps: Receiving a federal learning training application request sent by each federal node; the federal learning training application request comprises a federal learning identity, a federal learning qualification validity period and a federal learning time window of each federal node; According to the federal learning identity and the federal learning qualification validity period of each federal node, sending a qualification authentication request of each federal node to the authentication cloud server; the authentication cloud server is used for identity authentication and validity period authentication of each federal node, and a qualification authentication result is obtained; According to the qualification authentication result returned by the authentication cloud server, a target federal node in each federal node is determined; according to the federal learning time window of the target federal node, the federal learning period of the target federal node is determined; the target federal node is a federal node whose identity authentication and validity period authentication are both passed in the qualification authentication result; Sending the federal learning period of the target federal node to the target federal node; the target federal node is used for participating in federal learning training within the federal learning period.

2. The federated learning training method of claim 1, wherein, The federal learning period of the target federal node is determined according to the federal learning time window of the target federal node, comprising: Obtaining the federal learning time window of each target federal node; Processing the federal learning time window of each target federal node by intersection to obtain a public time window; The public time window is used as the federal learning period of each target federal node.

3. The federated learning training method of claim 1, wherein, Each federal node comprises a first federal node, and the first federal node belongs to a federal node in an edge cloud; The federal learning training application request sent by each federal node comprises: Receiving the federal learning training application request of the first federal node sent through a first communication link; The first communication link is a communication link formed by the first federal node, a first POP component, a leaf device of a first POD component, a Spine device of the first POD component, a second POP component and the supervision cloud server in sequence.

4. The federated learning training method of claim 1, wherein, Each federal node comprises a second federal node, and the second federal node does not belong to a federal node in an edge cloud; The federal learning training application request sent by each federal node comprises: Receiving the federal learning training application request of the second federal node sent through a second communication link; The second communication link is a communication link formed by the second federal node, a first CPE device, a first OLT device, a leaf device of a first POD component, a Spine device of the first POD component, a second POP component and the supervision cloud server in sequence.

5. The federated learning training method according to claim 4, characterized in that, Each federal node further comprises a third federal node, the third federal node does not belong to a federal node in an edge cloud, and the position of the third federal node in the metropolitan area network is different from that of the second federal node; The receiving federal node sends a federal learning training application request, including: Receiving the federal learning training application request of the third federal node sent through the third communication link; Wherein, the third communication link is a communication link formed by the third federal node, the second CPE device, the second OLT device, the leaf device of the second POD component, the Spine device of the second POD component, the Super Spine device of the export function area component, the Spine device of the first POD component, the second POP component and the supervision cloud server connected in turn.

6. The federated learning training method of claim 1, wherein, The authentication cloud server is used for receiving the qualification registration application request sent by each federal node, and performing qualification audit on each federal node, and after the qualification audit is passed, generating the federal learning identity and the federal learning qualification validity period of each federal node, and sending the federal learning identity and the federal learning qualification validity period of each federal node to each federal node.

7. The federated learning training method according to any one of claims 1 to 6, characterized in that, Also includes: Receiving the local model parameters sent by each target federal node; The local model parameters are obtained by training the initial model parameters of the federal learning model based on the training sample of the target federal node; According to the local model parameters of each target federal node, generate global model parameters, and send the global model parameters to each target federal node; each target federal node updates the local model parameters of the federal learning model according to the global model parameters; Receiving the updated local model parameters sent by each target federal node, returning the step of generating global model parameters according to the local model parameters of each target federal node, and sending the global model parameters to each target federal node, until the training end condition is met, and obtaining the trained federal learning model.

8. A federated learning training apparatus, comprising: The supervision cloud server in the center cloud, the center cloud further includes an authentication cloud server, the center cloud is communicated and connected with at least one federal node through a metropolitan area network, and the device comprises: A training application request receiving module is used for receiving the federal learning training application request sent by each federal node; the federal learning training application request includes the federal learning identity, the federal learning qualification validity period and the federal learning time window of each federal node; The qualification authentication request sending module is used for sending the qualification authentication request of each federal node to the authentication cloud server according to the federal learning identity and the federal learning qualification validity period of each federal node; the authentication cloud server is used for identity authentication and validity period authentication of each federal node, and obtains the qualification authentication result; The target federal node determination module is used for determining the target federal node in each federal node according to the qualification authentication result returned by the authentication cloud server; the federal learning time window of the target federal node is determined according to the federal learning time window of the target federal node; the target federal node is a federal node whose identity authentication and validity period authentication are both passed as represented by the qualification authentication result; The authentication cloud server is used for receiving the qualification registration application request sent by each federal node, and performing qualification audit on each federal node, and after the qualification audit is passed, generating the federal learning identity and the federal learning qualification validity period of each federal node, and sending the federal learning identity and the federal learning qualification validity period of each federal node to each federal node. A federal learning period sending module is configured to send a federal learning period of the target federal node to the target federal node, and the target federal node is configured to participate in federal learning training within the federal learning period. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-8 when the computer program is executed by the processor. The computer program, when executed by the processor, implements the steps of the method of any one of claims 1 to 7.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program, when executed by the processor, implements the steps of the method of any one of claims 1 to 7.

11. A computer program product comprising a computer program, characterized in that, The computer program, when executed by the processor, implements the steps of the method of any one of claims 1 to 7. The computer program, when executed by the processor, implements the steps of the method of any one of claims 1 to 7.

Images