Distributed service method, device and system based on federated learning

Abstract

The present disclosure relates to a federated learning-based distributed service method, device and system, and a computer storage medium. The distributed service method comprises: determining a plurality of target distributed nodes according to an expected waiting time of a center node, a service identifier of a service required by the center node, and a predicted response time of a plurality of distributed nodes, and a service identifier of a service provided by the plurality of distributed nodes, the expected waiting time and the service identifier of the required service matching the predicted response time and the service identifier of the provided service of the target distributed nodes; sending a service request to the plurality of target distributed nodes, the service request comprising the expected waiting time and the service identifier of the service required by the center node, the service request being used to trigger the plurality of target distributed nodes to execute a federated learning task within the expected waiting time; and receiving service responses from the plurality of target distributed nodes, each service response of each target distributed node comprising an execution result of the federated learning task.

Description

Technical Field

[0001] This disclosure relates to the field of communication technology, and in particular to distributed service methods, apparatus and systems based on federated learning, and computer-storable media. Background Technology

[0002] Federated learning systems are essentially distributed learning systems where multiple distributed nodes collectively provide services to a central node. Unlike traditional distributed learning, the distributed nodes (also known as participants) in federated learning have complete autonomy over their local data and can decide whether and when to join the federated learning process for collaborative modeling. This approach can solve the data silo problem while meeting data privacy, security, and regulatory requirements.

[0003] In each round of joint training in federated learning, after each distributed node completes its own training, it transmits the trained model parameters to the central node under an encrypted mechanism. The central node then performs model fusion and distributes the models to the distributed nodes for the next round of training.

[0004] In related technologies, a distributed node is selected from multiple distributed nodes whose service identifiers that can provide services match the service identifiers required by the central node. This selected distributed node is then used to perform a federated learning task corresponding to the service identifiers required by the central node. Summary of the Invention

[0005] In related technologies, because different target distributed nodes join federated learning at different times, have different datasets and computing resources, the time for different target distributed nodes to complete federated learning tasks is not synchronized. Simply matching the service identifiers of the services required by the central node and the service identifiers of the services that the distributed nodes can provide to determine the target distributed node can easily lead to excessively long actual waiting times for the central node, resulting in wasted resources and decreased federated learning efficiency.

[0006] To address the aforementioned technical problems, this disclosure proposes a solution that can reduce resource waste and improve the efficiency of federated learning.

[0007] According to a first aspect of this disclosure, a distributed service method based on federated learning is provided, comprising: determining a plurality of target distributed nodes from the plurality of distributed nodes based on the expected waiting time of a central node and the service identifier of the service it requires, and the expected response time of a plurality of distributed nodes and the service identifier of the service they can provide, wherein the expected waiting time and the service identifier of the service it requires match the expected response time and the service identifier of the service they can provide; sending a service request to the plurality of target distributed nodes, the service request including the expected waiting time and the service identifier of the service required by the central node, the service request being used to trigger the plurality of target distributed nodes to execute a federated learning task corresponding to the service identifier of the service required by the central node within the expected waiting time; and receiving service responses from the plurality of target distributed nodes, each target distributed node's service response including the execution result of the federated learning task.

[0008] In some embodiments, determining a plurality of target distributed nodes includes: determining a plurality of candidate distributed nodes from the plurality of distributed nodes based on the expected waiting time of the central node and the service identifier of its required service, and the expected response time of the plurality of distributed nodes and the service identifier of its available service, wherein the expected waiting time and the service identifier of the required service match the expected response time and the service identifier of the available service of the candidate distributed nodes; and determining a plurality of target distributed nodes from the plurality of candidate distributed nodes based on the central node's screening strategy.

[0009] In some embodiments, determining multiple candidate distributed nodes includes: determining multiple candidate distributed nodes from the multiple distributed nodes through a service discovery process that interacts with the multiple distributed nodes, wherein the candidate distributed nodes are distributed nodes that decide to join federated learning, the service discovery request in the service discovery process includes the expected waiting time and the service identifier of the required service, and the service discovery response in the service discovery process comes from the candidate distributed nodes and includes the distributed node identifier and the expected response time.

[0010] In some embodiments, determining a plurality of candidate distributed nodes includes: determining a plurality of candidate distributed nodes from the plurality of distributed nodes through a service discovery process interacting with a configuration node, wherein the configuration node stores the expected response time of the plurality of distributed nodes and the service identifiers of the services they can provide, the service discovery request in the service discovery process includes the expected waiting time and the service identifier of the required service, and the service discovery response in the service discovery process comes from the configuration node and includes the distributed node identifier and expected response time of the candidate distributed nodes.

[0011] In some embodiments, determining a plurality of candidate distributed nodes includes: obtaining configuration information of the plurality of distributed nodes, the configuration information including the expected response time of each distributed node and the service identifier of the services it can provide; and determining the distributed nodes whose expected waiting time and the service identifier of the required service match the expected response time and the service identifier of the services it can provide in the configuration information as candidate distributed nodes.

[0012] In some embodiments, the filtering strategy includes: selecting from the plurality of candidate distributed nodes that the difference between the expected response times is within a certain range, as a plurality of target distributed nodes.

[0013] In some embodiments, the distributed service method further includes: before sending a service request to the plurality of target distributed nodes, determining whether the expected waiting time and the expected response time of each target distributed node meet the adjustment conditions for adjusting the expected waiting time; if the adjustment conditions are met, adjusting the expected waiting time, wherein, when adjusting the expected waiting time, the service request includes the adjusted expected waiting time.

[0014] In some embodiments, the adjustment condition includes the difference between the expected response time of the target distributed node and the expected waiting time being less than or equal to a difference threshold. Adjusting the expected waiting time includes: when the difference between the expected response time of the target distributed node and the expected waiting time is less than or equal to the difference threshold, extending the expected waiting time so that the difference between the expected response time of the target distributed node and the adjusted expected waiting time is greater than the difference threshold.

[0015] In some embodiments, if the actual response time of any target distributed node is longer than the expected waiting time, the service response of the target distributed node may further include timeout information, which is used to guide the adjustment of the expected waiting time for re-executing the federated learning task corresponding to the service identifier.

[0016] In some embodiments, where the service identifier of the available service includes the service identifier of the required service and the expected response time is shorter than the expected waiting time, the expected waiting time and the service identifier of the required service are matched with the expected response time and the service identifier of the available service.

[0017] In some embodiments, the central node includes a Network Data Analysis Function (NWDAF) consumer, and the distributed nodes include NWDAF producers; or the central node includes a Management Data Analysis Service (MDAS) entity, and the distributed nodes include a 5G base station (gNB).

[0018] In some embodiments, where the central node includes a Network Data Analysis Function (NWDAF) consumer and the distributed nodes include NWDAF producers, the configuration node includes a Network Repository Function (NRF) element.

[0019] In some embodiments, the distributed service method further includes: fusing the execution results from the plurality of target distributed nodes for the federated learning task to obtain an aggregated result; and sending the aggregated result to the plurality of target distributed nodes, so that the plurality of distributed nodes can determine whether to execute the federated learning task corresponding to the service identifier again.

[0020] In some embodiments, the federated learning task includes at least one of a federated learning training task and a data analysis task.

[0021] In some embodiments, the distributed service method is executed by the central node.

[0022] According to a second aspect of this disclosure, a distributed service apparatus based on federated learning is provided, comprising: a determining module configured to determine a plurality of target distributed nodes from the plurality of distributed nodes based on the expected waiting time of a central node and the service identifier of the service it requires, and the expected response time of a plurality of distributed nodes and the service identifier of the service they can provide, wherein the expected waiting time and the service identifier of the service it requires match the expected response time and the service identifier of the service they can provide; a sending module configured to send a service request to the plurality of target distributed nodes, the service request including the expected waiting time and the service identifier, the service request being used to trigger the plurality of target distributed nodes to execute a federated learning task corresponding to the service identifier within the expected waiting time; and a receiving module configured to receive service responses from the plurality of target distributed nodes, each target distributed node's service response including the execution result of the federated learning task.

[0023] According to a third aspect of this disclosure, a distributed service apparatus based on federated learning is provided, comprising: a memory; and a processor coupled to the memory, the processor being configured to execute the distributed service method based on federated learning as described in any of the above embodiments based on instructions stored in the memory.

[0024] According to a fourth aspect of this disclosure, a distributed service system based on federated learning is provided, comprising: a central node configured to: determine a plurality of target distributed nodes from the plurality of distributed nodes based on the central node's expected waiting time and the service identifier of the required service, and the expected response time and the service identifier of the available service from the plurality of distributed nodes, wherein the expected waiting time and the service identifier of the required service match the expected response time and the service identifier of the available service from the target distributed nodes; send service requests to the plurality of target distributed nodes, the service requests including the expected waiting time and the service identifier, the service requests being used to trigger the plurality of target distributed nodes to execute a federated learning task corresponding to the service identifier within the expected waiting time; and receive service responses from the plurality of target distributed nodes, each target distributed node's service response including the execution result of the federated learning task.

[0025] In some embodiments, the distributed service system further includes: the plurality of target distributed nodes, each configured to receive a service request from the central node, perform a federated learning task corresponding to the service identifier based on the service request, and generate and send the service response.

[0026] According to a fifth aspect of this disclosure, a computer-storeable medium is provided having computer program instructions stored thereon, which, when executed by a processor, implement the federated learning-based distributed service method described in any of the above embodiments.

[0027] The above embodiments can reduce resource waste and improve the efficiency of federated learning. Attached Figure Description

[0028] The accompanying drawings, which form part of this specification, illustrate embodiments of this disclosure and, together with the specification, serve to explain the principles of this disclosure.

[0029] This disclosure will become clearer with reference to the accompanying drawings and the following detailed description, wherein:

[0030] Figure 1 This is a flowchart illustrating a distributed service method based on federated learning according to some embodiments of the present disclosure;

[0031] Figure 2 This is a block diagram illustrating a distributed service apparatus based on federated learning according to some embodiments of the present disclosure;

[0032] Figure 3 This is a block diagram illustrating a distributed service apparatus based on federated learning according to other embodiments of the present disclosure;

[0033] Figure 4This is a block diagram illustrating a distributed service system based on federated learning according to some embodiments of the present disclosure;

[0034] Figure 5 This is a signaling diagram illustrating a federated learning-based distributed service method according to some embodiments of the present disclosure;

[0035] Figure 6 This is a signaling diagram illustrating a distributed service method based on federated learning according to other embodiments of this disclosure;

[0036] Figure 7 This is a signaling diagram illustrating a distributed service method based on federated learning according to some embodiments of the present disclosure;

[0037] Figure 8 This is a block diagram illustrating a computer system for implementing some embodiments of the present disclosure. Detailed Implementation

[0038] Various exemplary embodiments of the present disclosure will now be described in detail with reference to the accompanying drawings. It should be noted that, unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps set forth in these embodiments do not limit the scope of the present disclosure.

[0039] At the same time, it should be understood that, for ease of description, the dimensions of the various parts shown in the accompanying drawings are not drawn according to actual scale.

[0040] The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit this disclosure or its application or use.

[0041] Techniques, methods, and equipment known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and equipment should be considered part of the specification.

[0042] In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values.

[0043] It should be noted that similar labels and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be discussed further in subsequent figures.

[0044] Figure 1 This is a flowchart illustrating a distributed service method based on federated learning according to some embodiments of the present disclosure.

[0045] like Figure 1As shown, the federated learning-based distributed service method includes steps S11-S13. In some embodiments, the federated learning-based distributed service method is executed by a central node.

[0046] In step S11, multiple target distributed nodes are determined from the multiple distributed nodes based on the expected waiting time of the central node and the service identifier of its required service, and the expected response time of the multiple distributed nodes and the service identifiers of the services they can provide. The expected waiting time and the service identifier of the required service are matched with the expected response time and the service identifiers of the target distributed nodes. For example, the service identifier is an Analytics ID.

[0047] In some embodiments, where the service identifier of the available service includes the service identifier of the required service and the expected response time is shorter than the expected waiting time, the expected waiting time and the service identifier of the required service are matched with the expected response time and the service identifier of the available service.

[0048] The expected waiting time of the central node represents the anticipated time from when the central node issues a service request to when it receives service responses from each target distributed node. Based on the central node's expected waiting time, the central node expects each target distributed node to receive the service request from the central node, complete its respective federated learning task, and send its service response back to the central node within the expected waiting time. From another perspective, the central node's expected waiting time represents the shortest time from when the central node issues a service request to when it receives service responses from each target distributed node during a round of federated learning. This shortest time refers to the shortest time expected by the central node.

[0049] The estimated response time of a distributed node represents the time it takes for the distributed node to receive a service request from the central node and then send a service response back to the central node. The estimated response time of a distributed node is related to its computing resources; when distributed nodes join the federated learning process at the same time, more computing resources result in a shorter response time. The estimated response time of a distributed node is related to its start time in the federated learning process and its computing resources. Computational resources can be estimated using the time required for a distributed node to complete one round of the federated learning task.

[0050] In some embodiments, multiple target distributed nodes can be determined in the following manner.

[0051] First, based on the expected waiting time of the central node and the service identifier of its required service, and the expected response time of multiple distributed nodes and the service identifiers of the services they can provide, multiple candidate distributed nodes are determined from the multiple distributed nodes. The expected waiting time and the service identifier of the required service are matched with the expected response time and the service identifiers of the services that the candidate distributed nodes can provide.

[0052] Then, based on the central node's selection strategy, multiple target distributed nodes are determined from among the multiple candidate distributed nodes.

[0053] In the above embodiments, target distributed nodes are not only selected based on the expected waiting time of the central node and the expected response time of the distributed nodes, but also the selection strategy of the central node is considered. This provides a further guarantee for each target distributed node to complete the federated learning task within the expected waiting time, further reduces resource waste, and improves the efficiency of federated learning.

[0054] In some embodiments, a service discovery process can be conducted by interacting with multiple distributed nodes to identify multiple candidate distributed nodes. These candidate distributed nodes are those that decide to join the federated learning process. The service discovery request during the service discovery process includes the expected wait time and the service identifier of the requested service. The service discovery response during the service discovery process comes from the candidate distributed nodes and includes the distributed node identifier and the expected response time.

[0055] In some embodiments, candidate distributed nodes can autonomously determine whether to join federated learning based on data security, available computing resources, and remaining time resources. For example, if a candidate distributed node's data security requirement level is higher than a preset threshold, the node decides not to join federated learning; otherwise, it joins. Similarly, if a candidate distributed node's available computing resources are less than a preset threshold, it decides not to join federated learning; otherwise, it joins. Likewise, if a candidate distributed node's remaining time resources are less than a preset threshold, it decides not to join federated learning; otherwise, it joins.

[0056] In other embodiments, a service discovery process can be conducted by interacting with a configuration node to identify multiple candidate distributed nodes from among the distributed nodes. The configuration node stores the estimated response times of the multiple distributed nodes and their service identifiers for the services they can provide. The service discovery request during the service discovery process includes the expected wait time and the service identifier of the requested service. The service discovery response during the service discovery process comes from the configuration node and includes the distributed node identifier and estimated response time of the candidate distributed nodes identified by the configuration node.

[0057] In some embodiments, configuration information for multiple distributed nodes, including the expected response time of each distributed node and the service identifiers of the services it can provide, can be obtained first. Then, distributed nodes whose expected waiting time and the service identifiers of the required services match the expected response time and the service identifiers of the services available in the configuration information are identified as candidate distributed nodes. For example, taking the distributed service method executed by the central node as an example, the configuration information of multiple distributed nodes is pre-configured and stored locally on the central node.

[0058] In some embodiments, the discovery service request and discovery service response described above may also be discovery service subscription and discovery service notification, respectively.

[0059] In some embodiments, the selection strategy for the central node includes selecting multiple candidate distributed nodes from a pool of candidate distributed nodes whose expected response times differ within a certain range, and using these as multiple target distributed nodes. By controlling the difference between the expected response times of the multiple target distributed nodes, the actual response times of each target distributed node can be made closer, further optimizing the use of computing and time resources of each distributed node, further reducing resource waste, and improving the efficiency of federated learning.

[0060] In some embodiments, if the actual response time of any target distributed node is longer than the expected waiting time, the service response of that target distributed node also includes timeout information. The timeout information is used to guide adjustments to the expected waiting time for re-executing the federated learning task corresponding to the service identifier. By carrying timeout information in the target distributed node that has a timeout response to indicate an adjustment to the expected waiting time, rather than directly removing the target distributed node with a timeout response in the next round of federated learning, it is possible to avoid ignoring distributed nodes that are computationally slower but still have a say in participating in federated learning, especially when extending the expected waiting time appropriately has a low impact on the federated learning task. In this case, the central node can determine whether to adjust the expected waiting time based on its own configuration and the timeout information of the target distributed node.

[0061] In step S12, service requests are sent to multiple target distributed nodes. The service requests include the expected waiting time and the service identifier of the service required by the central node. The service requests trigger the multiple target distributed nodes to execute the federated learning task corresponding to the service identifier of the service required by the central node within the expected waiting time.

[0062] In some embodiments, before sending service requests to multiple target distributed nodes, it is determined whether the expected waiting time and the expected response time of each target distributed node meet the adjustment conditions for adjusting the expected waiting time. If the adjustment conditions are met, the expected waiting time is adjusted. When the expected waiting time is adjusted, the service request includes the adjusted expected waiting time. By combining the expected waiting time and the expected response time to adjust the expected waiting time, it is possible to ensure that each target distributed node completes the federated learning task within the expected waiting time of the central node or the adjusted expected waiting time, thereby further improving resource utilization.

[0063] In some embodiments, the adjustment condition includes the difference between the expected response time and the expected waiting time of the target distributed node being less than or equal to a difference threshold. When the difference between the expected response time and the expected waiting time of the target distributed node is less than or equal to the difference threshold, the expected waiting time is extended so that the difference between the expected response time of the target distributed node and the adjusted expected waiting time is greater than the difference threshold. By allowing the central node to autonomously extend the expected waiting time when the expected response time is close to the expected waiting time, the goal is to ensure that each target distributed node completes the entire federated learning task within the central node's expected waiting time or the adjusted expected waiting time, thereby further improving resource utilization.

[0064] In some embodiments, the federated learning task includes at least one of a federated learning training task and a data analysis task. For example, the data analysis task is an NWDAF analysis task.

[0065] In step S13, service responses are received from multiple target distributed nodes. The service response from each target distributed node includes the execution result of the federated learning task.

[0066] In some embodiments, after receiving service responses from multiple target distributed nodes, the execution results of the federated learning task from the multiple target distributed nodes are merged to obtain an aggregated result; and the aggregated result is sent to the multiple target distributed nodes for the multiple distributed nodes to determine whether to execute the federated learning task corresponding to the service identifier again.

[0067] In some embodiments, the central node includes an NWDAF (Network Data Analytics Function) consumer, and the distributed nodes include NWDAF producers. The NWDAF producer provides data analysis services or federated learning training services to the NWDAF consumer, specifically the network data analysis function as defined in TS23.288. The NWDAF consumer can provide network data analysis functions externally based on the network data analysis results provided by the NWDAF producer.

[0068] For example, assuming the central node includes NWDAF consumers and the distributed nodes include NWDAF producers, the above-mentioned configuration nodes include NRF (Network Repository Function) network elements. When registering with the NRF network element, NWDAF producers can provide information such as NWDAF ID, estimated response time, and computing resources.

[0069] In other embodiments, the central node includes an MDAS (Management Data Analysis Service) entity, and the distributed nodes include gNBs (next generation Node Base stations), i.e., 5G base stations. The MDAS entity needs to collect data to provide management data analysis services, and this data source includes the 5G base stations. The 5G base stations can provide the MDAS entity with raw measurement data or data that has been processed from the raw measurement data. For example, the processing performed by the 5G base stations based on the raw measurement data includes at least one of federated learning training and data analysis.

[0070] In the above embodiments, based on the expected waiting time of the central node and the service identifier of its required services, and the expected response time of multiple distributed nodes and the service identifiers of the services they can provide, a target distributed node whose expected waiting time and the service identifier of its required services match the expected response time and the service identifier of the services it can provide is determined to perform the federated learning task. This allows for a prediction in advance whether the distributed node can complete the corresponding federated learning task within the expected waiting time, thereby reducing resource waste caused by the asynchronous nature of distributed nodes, reducing unnecessary resource consumption during the federated learning process, improving the efficiency of federated learning, and thus initially solving the asynchronous problem of each distributed node.

[0071] Figure 2 This is a block diagram illustrating a distributed service apparatus based on federated learning according to some embodiments of the present disclosure.

[0072] like Figure 2 As shown, the distributed service device 21 based on federated learning includes a determination module 211, a sending module 212, and a receiving module 213.

[0073] The determining module 211 is configured to determine multiple target distributed nodes from the multiple distributed nodes based on the expected waiting time of the central node and the service identifier of its required service, and the expected response time of multiple distributed nodes and the service identifier of their available services. The expected waiting time and the service identifier of the required service are matched with the expected response time and the service identifier of the available services of the target distributed nodes. For example, this is achieved by performing... Figure 1 Step S11 is shown.

[0074] Sending module 212 is configured to send service requests to the plurality of target distributed nodes. The service request includes the expected waiting time and the service identifier. The service request is used to trigger the plurality of target distributed nodes to execute a federated learning task corresponding to the service identifier within the expected waiting time, such as executing... Figure 1 Step S12 is shown.

[0075] The receiving module 213 is configured to receive service responses from the plurality of target distributed nodes, each target distributed node's service response including the execution result of a federated learning task, such as executing... Figure 1 Step S13 is shown.

[0076] Figure 3 This is a block diagram illustrating a distributed service apparatus based on federated learning according to other embodiments of the present disclosure.

[0077] like Figure 3 As shown, the federated learning-based distributed service apparatus 31 includes a memory 311 and a processor 312 coupled to the memory 311. The memory 311 is used to store instructions for executing embodiments of the federated learning-based distributed service method. The processor 312 is configured to execute the federated learning-based distributed service method in any of the embodiments of this disclosure based on the instructions stored in the memory 311.

[0078] Figure 4 This is a block diagram illustrating a distributed service system based on federated learning according to some embodiments of the present disclosure.

[0079] like Figure 4 As shown, the distributed service system 4 based on federated learning includes a central node 41.

[0080] The central node 41 is configured to execute the federated learning-based distributed service method in any of the embodiments of this disclosure.

[0081] In some embodiments, the federated learning-based distributed service system 4 further includes multiple target distributed nodes 42. Each target distributed node 42 is configured to receive service requests from the central node 41, perform a federated learning task corresponding to the service identifier based on the service request, and generate and send a service response.

[0082] In some embodiments, the central node 41 is configured as follows:

[0083] First, based on the expected waiting time of the central node and the service identifier of its required service, and the expected response time of multiple distributed nodes and the service identifiers of the services they can provide, multiple candidate distributed nodes are determined from the multiple distributed nodes. The expected waiting time and the service identifier of the required service are matched with the expected response time and the service identifiers of the services that the candidate distributed nodes can provide.

[0084] Then, based on the central node's selection strategy, multiple target distributed nodes are determined from among the multiple candidate distributed nodes.

[0085] Taking the process of a central node discovering multiple candidate distributed nodes through interaction with multiple distributed nodes as an example, the distributed service system 4 based on federated learning includes multiple distributed nodes. The central node is configured to send service discovery requests to multiple distributed nodes. Each distributed node is configured to, upon receiving a service discovery request from the central node, determine whether its corresponding expected response time and the service identifier of the service it can provide match the central node's expected waiting time and the service identifier of the service required by the central node, and then determine whether to join the federated learning process.

[0086] The expected response time and the service identifier of the services it can provide are matched with the expected waiting time and the service identifier of the services required by the central node. Distributed nodes that decide to join federated learning are configured to send a discovery service response to the central node. The discovery service response includes the distributed node identifier of the distributed node sending the discovery service response and the expected response time.

[0087] The central node is configured to identify candidate distributed nodes based on the distributed node identifier in the discovery service response.

[0088] Taking the service discovery process, in which a central node identifies multiple candidate distributed nodes through interaction with a configuration node, as an example, a federated learning-based distributed service system includes a configuration node. The configuration node is configured to store the estimated response times of multiple distributed nodes and their service identifiers. The central node is configured to send service discovery requests to the configuration node. The configuration node is configured to identify candidate distributed nodes based on the estimated response times and service identifiers of the multiple distributed nodes it stores, and send a service discovery response to the central node. The service discovery request in the service discovery process includes the expected waiting time and the service identifier of the desired service. The service discovery response in the service discovery process comes from the configuration node and includes the distributed node identifier and estimated response time of the candidate distributed nodes.

[0089] Taking the example where the central node stores configuration information for multiple distributed nodes locally, including the estimated response time of each distributed node and the service identifier of the services it can provide, the central node is configured to first obtain the configuration information for multiple distributed nodes, including the estimated response time of each distributed node and the service identifier of the services it can provide, and then determine the distributed nodes whose expected waiting time and the service identifier of the required service match the expected response time and the service identifier of the services provided in the configuration information as candidate distributed nodes. For example, taking the distributed service method executed by the central node as an example, the configuration information of multiple distributed nodes is pre-configured and stored locally on the central node.

[0090] Figure 5 This is a signaling diagram illustrating a federated learning-based distributed service method according to some embodiments of the present disclosure.

[0091] like Figure 5 As shown, the central node sends discovery service requests to distributed nodes 1, 2, and 3 respectively. The discovery service request includes the central node's expected waiting time and the service identifier of the service required by the central node.

[0092] After receiving a discovery service request from the central node, distributed nodes 1, 2, and 3 determine whether their expected response time and the service identifier of the service they can provide match the central node's expected waiting time and the service identifier of the service required by the central node, and then determine whether they decide to join the federated learning.

[0093] For example, the expected response times and service identifiers of distributed nodes 1, 2, and 3 match the expected waiting time and service identifiers of the services required by the central node, and all three decide to join federated learning.

[0094] Distributed nodes 1, 2, and 3 each send a discovery service response to the central node. The discovery service response includes the distributed node identifiers of distributed nodes 1, 2, and 3, and the estimated response time.

[0095] After receiving the discovery service responses from distributed nodes 1, 2, and 3, the central node determines distributed nodes 1, 2, and 3 as candidate distributed nodes.

[0096] The central node determines multiple target distributed nodes from candidate distributed nodes 1, 2, and 3 based on its selection strategy.

[0097] In some embodiments, distributed nodes 1, 2, and 3 are all target distributed nodes.

[0098] The central node sends service requests to distributed nodes 1, 2, and 3 respectively. Each service request includes the expected waiting time and the service identifier of the service required by the central node. These service requests trigger distributed nodes 1, 2, and 3 to execute the federated learning task corresponding to the service identifier of the service required by the central node within the expected waiting time.

[0099] After receiving a service request from the central node, distributed nodes 1, 2, and 3 collect data on their respective distributed nodes corresponding to the service identifiers required by the central node, and use the collected data to perform the corresponding federated learning tasks.

[0100] After completing the federated learning task, distributed nodes 1, 2, and 3 each send a service response to the central node. The service response includes the execution result of the federated learning task. In some embodiments, if a distributed node times out during the federated learning task completion process, the service response may also include a timeout message. The timeout message may include information such as the duration by which the actual response time exceeded the expected waiting time.

[0101] Figure 6 This is a signaling diagram illustrating a distributed service method based on federated learning according to other embodiments of this disclosure.

[0102] like Figure 6 As shown, the central node sends discovery service requests to the configuration nodes corresponding to distributed nodes 1, 2, and 3, respectively. The discovery service request includes the central node's expected waiting time and the service identifier of the service required by the central node. The configuration nodes store the expected response times of distributed nodes 1, 2, and 3, as well as the service identifiers of the services they can provide.

[0103] After receiving a discovery service request from the central node, the configuration node determines whether the expected response time and the service identifier of the services that the distributed nodes 1, 2, and 3 can provide match the expected waiting time and the service identifier of the services required by the central node, and identifies the matching distributed nodes as candidate distributed nodes.

[0104] For example, the expected response times of distributed nodes 1, 2, and 3, as well as the service identifiers of the services they can provide, are matched with the expected waiting time of the central node and the service identifiers of the services required by the central node.

[0105] The configuration node sends a discovery service response to the central node. The discovery service response includes the distributed node identifiers of distributed nodes 1, 2, and 3, and the estimated response time.

[0106] After receiving the discovery service response from the configuration node, the central node determines distributed nodes 1, 2, and 3 in the discovery service response as candidate distributed nodes.

[0107] The central node determines multiple target distributed nodes from candidate distributed nodes 1, 2, and 3 based on its selection strategy.

[0108] In some embodiments, distributed nodes 1, 2, and 3 are all target distributed nodes.

[0109] The central node sends service requests to distributed nodes 1, 2, and 3 respectively. Each service request includes the expected waiting time and the service identifier of the service required by the central node. These service requests trigger distributed nodes 1, 2, and 3 to execute the federated learning task corresponding to the service identifier of the service required by the central node within the expected waiting time.

[0110] After receiving a service request from the central node, distributed nodes 1, 2, and 3 each collect data on their respective distributed nodes corresponding to the service identifiers of the services required by the central node, and then use the collected data to perform the corresponding federated learning tasks. For example, after receiving a service request from the central node, distributed nodes 1, 2, and 3 can first determine whether to join the federated learning process. If they decide to join, they collect data on their distributed nodes corresponding to the service identifiers of the services required by the central node, and then use the collected data to perform the corresponding federated learning tasks.

[0111] After completing the federated learning task, distributed nodes 1, 2, and 3 each send a service response to the central node. The service response includes the execution result of the federated learning task. In some embodiments, if a distributed node times out during the federated learning task completion process, the service response may also include a timeout message. The timeout message may include information such as the duration by which the actual response time exceeded the expected waiting time.

[0112] Figure 7 This is a signaling diagram illustrating a distributed service method based on federated learning according to some embodiments of the present disclosure.

[0113] like Figure 7 As shown, the central node stores the estimated response times of distributed nodes 1, 2, and 3, as well as the service identifiers of the services they can provide.

[0114] The central node determines whether the expected response time and the service identifiers of the distributed nodes 1, 2, and 3 match the central node's expected waiting time and the service identifiers of the services required by the central node.

[0115] For example, the expected response times and service identifiers of distributed nodes 1, 2, and 3 match the expected waiting time and service identifiers of the services required by the central node. The central node then identifies distributed nodes 1, 2, and 3 as candidate distributed nodes.

[0116] The central node determines multiple target distributed nodes from candidate distributed nodes 1, 2, and 3 based on its selection strategy.

[0117] In some embodiments, distributed nodes 1, 2, and 3 are all target distributed nodes.

[0118] The central node sends service requests to distributed nodes 1, 2, and 3 respectively. Each service request includes the expected waiting time and the service identifier of the service required by the central node. These service requests trigger distributed nodes 1, 2, and 3 to execute the federated learning task corresponding to the service identifier of the service required by the central node within the expected waiting time.

[0119] After receiving a service request from the central node, distributed nodes 1, 2, and 3 each collect data on their respective distributed nodes corresponding to the service identifiers of the services required by the central node, and then use the collected data to perform the corresponding federated learning tasks. For example, after receiving a service request from the central node, distributed nodes 1, 2, and 3 can first determine whether to join the federated learning process. If they decide to join, they collect data on their distributed nodes corresponding to the service identifiers of the services required by the central node, and then use the collected data to perform the corresponding federated learning tasks.

[0120] After completing the federated learning task, distributed nodes 1, 2, and 3 each send a service response to the central node. The service response includes the execution result of the federated learning task. In some embodiments, if a distributed node times out during the federated learning task completion process, the service response may also include a timeout message. The timeout message may include information such as the duration by which the actual response time exceeded the expected waiting time.

[0121] The above Figures 5-7The central node in this system is, for example, an NWDAF consumer, and the distributed nodes are NWDAF producers. In this case, the configured nodes are NRF network elements. For example, Figures 5-7 The central node is the MDAS entity, and the distributed nodes are gNBs.

[0122] Network data analysis functionality is defined in TS 23.288. It is a network function within 5GC that supports data collection from Network Functions (NFs), Application Functions (AFs), and Operations Administration and Maintenance (OAM) systems, and provides analytical information to these systems. In multiple NWDAF deployment scenarios, when an NWDAF service consumer requests analysis services that require joint provision by multiple NWDAF producers, these producers can collaboratively provide the analysis service to the consumer. In this scenario, the NWDAF service consumer acts as the central node, and the multiple NWDAFs providing services to the consumer act as distributed nodes. When registering with the NRF network element, the distributed node NWDAFs can provide information such as their NWDAF ID, estimated response time, and computing resources.

[0123] Figure 8 This is a block diagram illustrating a computer system for implementing some embodiments of the present disclosure.

[0124] like Figure 8 As shown, the computer system 80 can be represented in the form of a general computing device. The computer system 80 includes a memory 810, a processor 820, and a bus 800 connecting different system components.

[0125] The memory 810 may include, for example, system memory, non-volatile storage media, etc. The system memory may store, for example, an operating system, applications, a boot loader, and other programs. The system memory may include volatile storage media, such as random access memory (RAM) and / or cache memory. The non-volatile storage media may store, for example, instructions for executing at least one of the corresponding embodiments of the federated learning-based distributed service method. Non-volatile storage media include, but are not limited to, disk storage, optical storage, flash memory, etc.

[0126] The processor 820 can be implemented using a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete hardware components such as discrete gates or transistors. Accordingly, each module, such as the decision module and the determination module, can be implemented by executing instructions in the central processing unit (CPU) memory to perform the corresponding steps, or by implementing dedicated circuitry to perform the corresponding steps.

[0127] Bus 800 can use any of the various bus architectures. For example, bus architectures include, but are not limited to, the Industry Standard Architecture (ISA) bus, the Micro Channel Architecture (MCA) bus, and the Peripheral Component Interconnect (PCI) bus.

[0128] The computer system 80 may also include an input / output interface 830, a network interface 840, and a storage interface 850. These interfaces 830, 840, and 850, as well as the memory 810 and processor 820, can be connected via a bus 800. The input / output interface 830 provides a connection interface for input / output devices such as a monitor, mouse, and keyboard. The network interface 840 provides a connection interface for various networked devices. The storage interface 850 provides a connection interface for external storage devices such as floppy disks, USB flash drives, and SD cards.

[0129] Various aspects of this disclosure are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus, and computer program products according to embodiments of this disclosure. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations thereof, can be implemented by computer-readable program instructions.

[0130] These computer-readable program instructions are provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable device to produce a machine, such that execution of the instructions by the processor produces means for implementing the functions specified in one or more boxes of the flowchart and / or block diagram.

[0131] These computer-readable program instructions may also be stored in a computer-readable storage medium. These instructions cause a computer to work in a particular manner to produce an article of manufacture, including instructions that implement the functions specified in one or more boxes in a flowchart and / or block diagram.

[0132] This disclosure may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects.

[0133] The distributed service method, apparatus, system, and computer-storable medium based on federated learning described in the above embodiments can reduce resource waste and improve the efficiency of federated learning.

[0134] This concludes the detailed description of the distributed service method, apparatus, system, and computer-storable medium based on federated learning according to this disclosure. To avoid obscuring the concept of this disclosure, some details known in the art have not been described. Those skilled in the art can fully understand how to implement the technical solutions disclosed herein based on the above description.

Claims

1. A distributed service method based on federated learning, comprising: Based on the expected waiting time of the central node and the service identifier of its required service, and the expected response time of multiple distributed nodes and the service identifier of the services they can provide, multiple target distributed nodes are determined from the multiple distributed nodes, wherein the expected waiting time and the service identifier of the required service match the expected response time and the service identifier of the services they can provide of the target distributed nodes. A service request is sent to the plurality of target distributed nodes. The service request includes the expected waiting time and the service identifier of the service required by the central node. The service request is used to trigger the plurality of target distributed nodes to execute a federated learning task corresponding to the service identifier of the service required by the central node within the expected waiting time. Receive service responses from the plurality of target distributed nodes, wherein the service response from each target distributed node includes the execution result of the federated learning task. Wherein, the expected waiting time characterizes the expected time from when the central node issues the service request to when it receives the service response from each of the target distributed nodes, and the estimated response time characterizes the estimated time from when the distributed node receives the service request from the central node to when it sends the service response back to the central node.

2. The distributed service method based on federated learning according to claim 1, wherein, Identifying multiple target distributed nodes includes: Based on the expected waiting time of the central node and the service identifier of its required service, and the expected response time of multiple distributed nodes and the service identifier of the services they can provide, multiple candidate distributed nodes are determined from the multiple distributed nodes, and the expected waiting time and the service identifier of the required service are matched with the expected response time and the service identifier of the services they can provide. Based on the filtering strategy of the central node, multiple target distributed nodes are determined from the multiple candidate distributed nodes.

3. The distributed service method based on federated learning according to claim 2, wherein, The identified candidate distributed nodes include: Through the service discovery process interacting with the multiple distributed nodes, multiple candidate distributed nodes are determined from the multiple distributed nodes. The candidate distributed nodes are the distributed nodes that decide to join federated learning. The service discovery request in the service discovery process includes the expected waiting time and the service identifier of the required service. The service discovery response in the service discovery process comes from the candidate distributed nodes and includes the distributed node identifier and the expected response time.

4. The distributed service method based on federated learning according to claim 2, wherein, The identified candidate distributed nodes include: Through a service discovery process that interacts with the configuration node, multiple candidate distributed nodes are determined from the plurality of distributed nodes. The configuration node stores the expected response time of the plurality of distributed nodes and the service identifiers of the services they can provide. The service discovery request in the service discovery process includes the expected waiting time and the service identifier of the required service. The service discovery response in the service discovery process comes from the configuration node and includes the distributed node identifier and expected response time of the candidate distributed node.

5. The distributed service method based on federated learning according to claim 2, wherein, The identified candidate distributed nodes include: Obtain the configuration information of the plurality of distributed nodes, the configuration information including the estimated response time of each distributed node and the service identifier of the services it can provide; Distributed nodes that match the expected waiting time and the service identifier of the required service with the expected response time and the service identifier of the available service in the configuration information are identified as candidate distributed nodes.

6. The distributed service method based on federated learning according to claim 2, wherein, The screening strategy includes: From the plurality of candidate distributed nodes, select multiple candidate distributed nodes whose expected response times are within the range of the difference range, and use them as multiple target distributed nodes.

7. The distributed service method based on federated learning according to claim 1 further includes: Before sending a service request to the multiple target distributed nodes, it is determined whether the expected waiting time and the expected response time of each target distributed node meet the adjustment conditions for adjusting the expected waiting time. If the adjustment conditions are met, the expected waiting time is adjusted, wherein, when the expected waiting time is adjusted, the service request includes the adjusted expected waiting time.

8. The distributed service method based on federated learning according to claim 7, wherein, The adjustment condition includes that the difference between the expected response time of the target distributed node and the expected waiting time is less than or equal to a difference threshold, and adjusting the expected waiting time includes: If the difference between the expected response time of the target distributed node and the expected waiting time is less than or equal to the difference threshold, the expected waiting time is extended so that the difference between the expected response time of the target distributed node and the adjusted expected waiting time is greater than the difference threshold.

9. The distributed service method based on federated learning according to claim 1, wherein, If the actual response time of any target distributed node is longer than the expected waiting time, the service response of the target distributed node also includes timeout information, which is used to guide the adjustment of the expected waiting time for re-executing the federated learning task corresponding to the service identifier.

10. The distributed service method based on federated learning according to any one of claims 1-5, wherein, If the service identifier available for service includes the service identifier of the required service and the expected response time is shorter than the expected waiting time, the expected waiting time and the service identifier of the required service are matched with the expected response time and the service identifier available for service.

11. The distributed service method based on federated learning according to claim 1, wherein, The central node includes NWDAF consumers with network data analysis capabilities, and the distributed nodes include NWDAF producers; or The central node includes the Management Data Analysis Service (MDAS) entity, and the distributed nodes include 5G base stations (gNBs).

12. The distributed service method based on federated learning according to claim 4, wherein, In the case where the central node includes a Network Data Analysis Function (NWDAF) consumer and the distributed nodes include NWDAF producers, the configuration node includes a Network Repository Function (NRF) network element.

13. The distributed service method based on federated learning according to claim 1 further includes: The execution results from the multiple target distributed nodes for the federated learning task are merged to obtain the aggregated result; The aggregation result is sent to the plurality of target distributed nodes, which then determine whether to execute the federated learning task corresponding to the service identifier again.

14. The distributed service method based on federated learning according to claim 1, wherein, The federated learning task includes at least one of the federated learning training task and the data analysis task.

15. The distributed service method based on federated learning according to claim 1, wherein, The distributed service method is executed by the central node.

16. A distributed service device based on federated learning, comprising: The determination module is configured to determine multiple target distributed nodes from the multiple distributed nodes based on the expected waiting time of the central node and the service identifier of its required service, and the expected response time of the multiple distributed nodes and the service identifier of its available service. The expected waiting time and the service identifier of the required service match the expected response time and the service identifier of the available service of the target distributed node. The sending module is configured to send a service request to the plurality of target distributed nodes. The service request includes the expected waiting time and the service identifier. The service request is used to trigger the plurality of target distributed nodes to execute a federated learning task corresponding to the service identifier within the expected waiting time. The receiving module is configured to receive service responses from the plurality of target distributed nodes, wherein the service response from each target distributed node includes the execution result of the federated learning task. Wherein, the expected waiting time characterizes the expected time from when the central node issues the service request to when it receives the service response from each of the target distributed nodes, and the estimated response time characterizes the estimated time from when the distributed node receives the service request from the central node to when it sends the service response back to the central node.

17. A distributed service device based on federated learning, comprising: Memory; as well as A processor coupled to the memory, the processor being configured to execute the federated learning-based distributed service method as described in any one of claims 1 to 15, based on instructions stored in the memory.

18. A distributed service system based on federated learning, comprising: The central node is configured as follows: Based on the expected waiting time of the central node and the service identifier of its required service, and the expected response time of multiple distributed nodes and the service identifier of the services they can provide, multiple target distributed nodes are determined from the multiple distributed nodes, wherein the expected waiting time and the service identifier of the required service match the expected response time and the service identifier of the services they can provide of the target distributed nodes. A service request is sent to the plurality of target distributed nodes. The service request includes the expected waiting time and the service identifier. The service request is used to trigger the plurality of target distributed nodes to execute a federated learning task corresponding to the service identifier within the expected waiting time. Receive service responses from the plurality of target distributed nodes, wherein the service response from each target distributed node includes the execution result of the federated learning task.

19. The distributed service system based on federated learning according to claim 18, further comprising: The plurality of target distributed nodes are each configured to receive service requests from the central node, execute federated learning tasks corresponding to the service identifier based on the service requests, and generate and send the service response.

20. A computer-storeable medium having stored thereon computer program instructions that, when executed by a processor, implement the distributed service method based on federated learning as described in any one of claims 1 to 15.

Images