Communication link recommendation method and apparatus for federated learning

Abstract

Provided are a communication link recommendation method and device for federated learning. The communication link recommendation method comprises: obtaining geographical position information of a plurality of computing clusters and a center server; for each computing cluster in the plurality of computing clusters, according to the geographical position information of the computing cluster, the geographical position information of the center server, and link information between a plurality of trusted proxy nodes in a multi-hop network that has been constructed, estimating a reliable communication link of the computing cluster connecting to the center server via at least one trusted proxy node in turn; recommending the estimated reliable communication link of each computing cluster connecting to the center server to each computing cluster, so that each computing cluster uses the recommended reliable communication link to interact with the center server; wherein the plurality of computing clusters are used to perform federated learning on a same model based on local data, and the center server is used to maintain parameters and / or parameter gradients of the same model.

Description

TECHNICAL FIELD

[0001] The present disclosure relates generally to the technical field of machine learning, and more particularly, to a communication link recommendation method and device for federated learning. BACKGROUND

[0002] Federated machine learning (Federated machine learning / Federated Learning), also known as federated learning, joint learning, and alliance learning. Federated machine learning is a machine learning framework that can effectively help multiple data providers to use data and build machine learning models while meeting user privacy protection, data security, and government regulations. As a distributed machine learning paradigm, federated learning can effectively solve the data island problem, allowing participants to jointly model without sharing data, and can technically break down data silos and achieve AI collaboration.

[0003] Generally, federated learning needs to establish multi-party trusted computing based on a wide area network, but the dynamic instability, bandwidth limitation, and heterogeneity of the wide area network will restrict the progress and speed of learning and affect the reliability of the whole process of federated learning. Therefore, the communication link reliability problem of cross-wide-area federated learning needs to be solved. SUMMARY

[0004] An exemplary embodiment of the present disclosure is to provide a communication link recommendation method and device for federated learning, which can recommend reliable communication links for federated learning for computing clusters.

[0005] According to a first aspect of an embodiment of the present disclosure, a communication link recommendation method for federated learning is provided, comprising: obtaining geographical location information of a plurality of computing clusters and a center server; for each computing cluster in the plurality of computing clusters, estimating a reliable communication link of the computing cluster connecting to the center server via at least one trusted proxy node in a multi-hop network in sequence according to the geographical location information of the computing cluster, the geographical location information of the center server, and link information between a plurality of trusted proxy nodes in the multi-hop network that has been constructed; recommending the estimated reliable communication link connecting to the center server to each computing cluster, so that each computing cluster uses the recommended reliable communication link to interact with the center server; wherein the plurality of computing clusters are used to perform federated learning on a same model based on local data, and the center server is used to maintain parameters and / or parameter gradients of the same model.

[0006] Optionally, the step of estimating a reliable communication link for each computing cluster to connect to the central server via at least one trusted proxy node in turn comprises: inputting geographical position information of each computing cluster, geographical position information of the central server, and link information between the plurality of trusted proxy nodes into a pre-trained link recommendation model to obtain a reliable communication link for the computing cluster to connect to the central server output by the link recommendation model.

[0007] Optionally, the step of estimating a reliable communication link for each computing cluster to connect to the central server via at least one trusted proxy node in turn comprises: estimating a reliable communication link for the computing cluster to connect to the central server via at least one trusted proxy node in turn according to geographical position information of each computing cluster, geographical position information of the central server, data volume to be transmitted between the computing cluster and the central server, and link information between the plurality of trusted proxy nodes.

[0008] Optionally, the link information between the plurality of trusted proxy nodes comprises at least one of the following information of a link between any two trusted proxy nodes: whether the link is blocked, blocking degree, link carrying capacity, reliability information, packet loss condition, time delay, transmission rate, link history record, and whether the link is allowed to transmit data based on regional rules.

[0009] Optionally, the communication link recommendation method further comprises: monitoring communication quality of the reliable communication link recommended for each computing cluster to connect to the central server in real time; when it is monitored that the communication quality of the reliable communication link recommended for any one computing cluster to connect to the central server meets a damaged condition, re-estimating a reliable communication link for the computing cluster to connect to the central server via at least one trusted proxy node in turn according to geographical position information of the computing cluster, geographical position information of the central server, and updated link information between the plurality of trusted proxy nodes; recommending the re-estimated reliable communication link for the computing cluster to connect to the central server, so that the computing cluster uses the re-recommended reliable communication link to interact with the central server.

[0010] Optionally, the plurality of trusted proxy nodes are constructed on a public cloud; and / or, any two trusted proxy nodes in the plurality of trusted proxy nodes directly establish a link or establish a link based on other routing nodes.

[0011] Optionally, the communication link recommendation method further comprises: for each two computing clusters in the plurality of computing clusters, estimating a reliable communication link in which one of the two computing clusters is connected to the other of the two computing clusters via at least one trusted proxy node in sequence and multiple hops, according to the geographical location information of the two computing clusters and the link information between the plurality of trusted proxy nodes; and recommending the estimated reliable communication link in which the two computing clusters are connected to each other to the two computing clusters, so that the two computing clusters perform data interaction using the recommended reliable communication link in which the two computing clusters are connected to each other.

[0012] Optionally, the step of estimating the reliable communication link in which one of the two computing clusters is connected to the other of the two computing clusters via at least one trusted proxy node in sequence and multiple hops comprises: inputting the geographical location information of the two computing clusters and the link information between the plurality of trusted proxy nodes into a pre-trained link recommendation model to obtain a reliable communication link in which the two computing clusters are connected to each other output by the link recommendation model.

[0013] Optionally, the step of estimating the reliable communication link in which one of the two computing clusters is connected to the other of the two computing clusters via at least one trusted proxy node in sequence and multiple hops comprises: estimating a reliable communication link in which one of the two computing clusters is connected to the other of the two computing clusters via at least one trusted proxy node in sequence and multiple hops, according to the geographical location information of the two computing clusters, the amount of data to be transmitted between the two computing clusters, and the link information between the plurality of trusted proxy nodes.

[0014] Optionally, the communication link recommendation method further comprises: monitoring the communication quality of the reliable communication link in which each two computing clusters are connected to each other in real time; when it is monitored that the communication quality of the reliable communication link in which any two computing clusters are connected to each other meets a damaged condition, estimating a reliable communication link in which one of the two computing clusters is connected to the other of the two computing clusters via at least one trusted proxy node in sequence and multiple hops, according to the geographical location information of the two computing clusters and the updated link information between the plurality of trusted proxy nodes; and recommending the re-estimated reliable communication link in which the two computing clusters are connected to each other to the two computing clusters, so that the two computing clusters perform data interaction using the re-recommended reliable communication link.

[0015] Optionally, the communication link recommendation method further comprises: obtaining link state data reported by the plurality of trusted proxy nodes in real time; updating the link information between the plurality of trusted proxy nodes according to the obtained link state data of the plurality of trusted proxy nodes, so as to estimate the reliable communication link according to the updated link information.

[0016] Optionally, the link state data reported by the trusted agent node comprises: a transmission rate, a bandwidth, and a latency of a link between the trusted agent node and another trusted agent node.

[0017] According to a second aspect of the embodiments of the present disclosure, a communication link recommendation apparatus for federated learning is provided, comprising: a location information obtaining unit configured to obtain geographical location information of a plurality of computing clusters and a center server; a link estimation unit configured to, for each computing cluster in the plurality of computing clusters, estimate a reliable communication link for the computing cluster to connect to the center server via at least one trusted agent node in sequence based on the geographical location information of the computing cluster, the geographical location information of the center server, and link information between a plurality of trusted agent nodes in a multi-hop network that has been constructed; and a link recommendation unit configured to recommend the estimated reliable communication link for each computing cluster to connect to the center server, so that each computing cluster uses the recommended reliable communication link to interact with the center server for data; wherein the plurality of computing clusters are used to perform federated learning on a same model based on local data, and the center server is used to maintain parameters and / or parameter gradients of the same model.

[0018] Optionally, the link estimation unit is configured to input the geographical location information of each computing cluster, the geographical location information of the center server, and the link information between the plurality of trusted agent nodes into a pre-trained link recommendation model to obtain a reliable communication link for the computing cluster to connect to the center server output by the link recommendation model.

[0019] Optionally, the link estimation unit is configured to estimate a reliable communication link for each computing cluster to connect to the center server via at least one trusted agent node in sequence based on the geographical location information of the computing cluster, the geographical location information of the center server, an amount of data to be transmitted between the computing cluster and the center server, and the link information between the plurality of trusted agent nodes.

[0020] Optionally, the link information between the plurality of trusted agent nodes comprises at least one of the following information of a link between each two trusted agent nodes: whether to be blocked, a blocking degree, a link carrying capacity, reliability information, a packet loss situation, a latency, a transmission rate, a link history record, and whether to be allowed to establish a link to transmit data based on a regional rule.

[0021] Optionally, the communication link recommendation device further comprises a first link monitoring unit configured to monitor in real time a communication quality of a reliable communication link connecting to the center server recommended to each computing cluster; wherein the link estimation unit is further configured to, when the communication quality of the reliable communication link connecting to the center server recommended to any one computing cluster meets a damaged condition, re-estimate a reliable communication link of the computing cluster connecting to the center server in turn via at least one trusted proxy node according to the geographical position information of the computing cluster, the geographical position information of the center server, and the updated link information between the plurality of trusted proxy nodes; and wherein the link recommendation unit is further configured to recommend the re-estimated reliable communication link connecting to the center server to the computing cluster, so that the computing cluster uses the re-recommended reliable communication link to interact with the center server.

[0022] Optionally, the plurality of trusted proxy nodes are constructed on a public cloud; and / or each two trusted proxy nodes in the plurality of trusted proxy nodes directly establish a link or establish a link based on other routing nodes.

[0023] Optionally, the link estimation unit is further configured to, for each two computing clusters in the plurality of computing clusters, estimate a reliable communication link of one computing cluster in the two computing clusters connecting to another computing cluster in the two computing clusters in turn via at least one trusted proxy node according to the geographical position information of the two computing clusters and the link information between the plurality of trusted proxy nodes; and wherein the link recommendation unit is further configured to recommend the estimated reliable communication link connecting to each other to the two computing clusters, so that the two computing clusters use the recommended reliable communication link connecting to each other to interact with each other.

[0024] Optionally, the link estimation unit is configured to input the geographical position information of the two computing clusters and the link information between the plurality of trusted proxy nodes into a pre-trained link recommendation model to obtain a reliable communication link of the two computing clusters connecting to each other output by the link recommendation model.

[0025] Optionally, the link estimation unit is configured to estimate a reliable communication link of one computing cluster in the two computing clusters connecting to another computing cluster in the two computing clusters in turn via at least one trusted proxy node according to the geographical position information of the two computing clusters, the amount of data to be transmitted between the two computing clusters, and the link information between the plurality of trusted proxy nodes.

[0026] Optionally, the communication link recommendation apparatus further comprises a second link monitoring unit configured to monitor, in real time, a communication quality condition of a reliable communication link recommended to connect each two computing clusters to each other; wherein the link estimation unit is further configured to, when it is monitored that the communication quality condition of the reliable communication link recommended to connect any two computing clusters to each other satisfies a damaged condition, estimate, according to the geographical position information of the two computing clusters and the updated link information between the plurality of trusted proxy nodes, a reliable communication link of one of the two computing clusters to the other of the two computing clusters via at least one trusted proxy node in turn; and wherein the link recommendation unit is further configured to recommend, to the two computing clusters, the re-estimated reliable communication link to connect to each other, so that the two computing clusters use the re-recommended reliable communication link to interact data.

[0027] Optionally, the communication link recommendation apparatus further comprises a state data acquisition unit configured to acquire link state data reported in real time by the plurality of trusted proxy nodes; and a link information updating unit configured to update the link information between the plurality of trusted proxy nodes according to the acquired link state data of the plurality of trusted proxy nodes, so as to estimate a reliable communication link according to the updated link information.

[0028] Optionally, the link state data reported by the trusted proxy node comprises a transmission rate, a bandwidth and a time delay of a link between the trusted proxy node and other trusted proxy nodes.

[0029] According to a third aspect of the embodiments of the present disclosure, a system including at least one computing device and at least one storage device storing instructions is provided, wherein the instructions, when executed by the at least one computing device, cause the at least one computing device to perform the communication link recommendation method for federated learning as described above.

[0030] According to a fourth aspect of the embodiments of the present disclosure, a computer-readable storage medium storing instructions is provided, wherein the instructions, when executed by at least one computing device, cause the at least one computing device to perform the communication link recommendation method for federated learning as described above.

[0031] The communication link recommendation method and apparatus for federated learning according to the exemplary embodiments of the present disclosure can recommend a reliable communication link for federated learning for computing clusters, so as to ensure that multiple parties of federated learning can interact data on a stable and reliable communication link, thereby providing support for wide-area federated learning.

[0032] Additional aspects and / or advantages of the general inventive concept will be set forth in part in the description that follows, and in part will be obvious from the description, or can be learned by practice of the general inventive concept. BRIEF DESCRIPTION OF DRAWINGS

[0033] The above and other objects and features of the present disclosure exemplary embodiments will become more apparent from the following description of the exemplary embodiments, taken in conjunction with the accompanying drawings, which illustrate:

[0034] Figure 1 A flowchart illustrating a communication link recommendation method for federated learning according to an exemplary embodiment of the present disclosure is shown;

[0035] Figure 2 An example of a communication link recommendation method for federated learning according to an exemplary embodiment of the present disclosure is shown.

[0036] Figure 3 A block diagram illustrating the structure of a communication link recommendation apparatus for federated learning according to an exemplary embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0037] Reference will now be made in detail embodiments of the present disclosure, examples of which are illustrated in the accompanying drawings, wherein like reference numerals refer to like elements throughout. The embodiments will be explained by referring to the drawings in detail, in order to explain the present disclosure.

[0038] Figure 1 A flowchart illustrating a communication link recommendation method for federated learning according to an exemplary embodiment of the present disclosure is shown. The communication link recommendation method can be performed by a server (e.g., a federated network coordination server), for example.

[0039] Referring to Figure 1 In step S10, geographic location information of a plurality of computing clusters and a center server is acquired.

[0040] The plurality of computing clusters are used for federated learning on a same model based on local data, and the local data of different computing clusters can be different. For example, the local data of different computing clusters can be data provided by different data providers, and each computing cluster can store data of one data provider, so that data of different data providers are stored in different computing clusters and isolated from each other, meeting the requirements of privacy protection and data security. For example, the plurality of computing clusters are used for federated learning on a same model based on local training data, i.e., collaborative training of the same model; in addition, the plurality of computing clusters can also be used for federated learning on a same model based on local prediction data, i.e., collaborative model prediction using the same model.

[0041] As an example, each computing cluster can include at least one computing entity, and the computing entities in the same computing cluster use data of the same type.

[0042] The central server is configured to maintain parameters and / or parameter gradients of the model. As an example, when the central server is configured to maintain parameters of a same model for federated learning (i.e., for a horizontal federated learning scenario), the central server receives, from each computing cluster, a request to read the parameters of the same model, and returns, in response to the request, the latest parameters of the same model; and the central server receives, from each computing cluster, a gradient of the parameters of the same model trained based on local training data using the requested parameters, and updates the parameters of the same model based on the gradients of the parameters of the same model reported by the respective computing clusters. It should be understood that different central servers can be used for federated learning of different models, and the present disclosure does not limit the same.

[0043] As an example, when the federated learning agent service registration information is received, the geographic location information of the plurality of computing clusters registered for federated learning of the same model and the geographic location information of the central server required to be used can be obtained. The plurality of computing clusters need to interact with the central server for data (e.g., for parameter interaction) because they need to jointly train the same model, and in addition, if vertical federated learning is performed, the computing clusters also need to interact with each other for data (e.g., intermediate results need to be sent to other computing clusters during training).

[0044] It should be understood that in federated learning, the data transmitted between computing clusters and between computing clusters and the central server can be encrypted data, which will not be described in detail in the present disclosure.

[0045] In step S20, for each of the plurality of computing clusters, a reliable communication link for the computing cluster to connect to the central server via at least one trusted agent node in turn is estimated according to the geographic location information of the computing cluster, the geographic location information of the central server, and the link information between the plurality of trusted agent nodes in the constructed multi-hop network. That is, for each computing cluster, the most reliable communication link in the multi-hop network for the computing cluster to interact with the central server for data is selected.

[0046] As an example, the reliability of the communication link can be measured by transmission delay and data loss probability. For example, the lower the transmission delay of the communication link, the lower the data loss probability, and the higher the reliability.

[0047] The constructed multi-hop network is an overlay grid formed by the plurality of trusted agent nodes. The trusted agent node is an agent node that can be trusted by each party.

[0048] As an example, the plurality of trusted agent nodes can be constructed on a public cloud (e.g., a regional cloud service covering the world). In addition, the plurality of trusted agent nodes can be constructed on a global public cloud, an IDC machine room, and an edge device.

[0049] According to embodiments of the present disclosure, the communication link recommendation method can be applied to a cross-public network federated learning scenario (i.e., a global federated learning scenario).

[0050] As an example, each two of the plurality of trusted agent nodes can directly establish a link or establish a link based on other routing nodes. As an example, the other routing nodes can be routing nodes on a public cloud, which can be other types of routing nodes.

[0051] As an example, the link information between the plurality of trusted agent nodes can be updated in real time.

[0052] As an example, the link information between the plurality of trusted agent nodes can include real-time evaluation information of a link between each two of the plurality of trusted agent nodes, and can further include historical evaluation information of the link between each two of the plurality of trusted agent nodes. As an example, the evaluation information of the link can be information for evaluating the transmission condition and transmission performance of the link, such as information for evaluating the blocking condition, carrying capacity, reliability, packet loss condition, delay condition, transmission rate, etc. of the link.

[0053] As an example, the link information between the plurality of trusted agent nodes can include at least one of the following information of a link between each two of the plurality of trusted agent nodes: whether it is blocked (i.e., whether it is currently blocked), the blocking degree (i.e., the current blocking degree), the link carrying capacity (i.e., the link carrying upper limit), the reliability information, the packet loss condition (i.e., the current packet loss condition), the delay (i.e., the current delay condition), the transmission rate (i.e., the current transmission rate), the link history record, and whether it is allowed to establish a link to transmit data based on regional rules. As an example, the reliability information can include the historical disconnection (e.g., due to trusted agent node downtime, etc.) of the link, such as the total duration and / or total number of disconnections. As an example, the link history record can be link historical evaluation information, such as whether it is often blocked, often delayed, etc.

[0054] As an example, the communication link recommendation method according to the exemplary embodiments of the present disclosure can further include: obtaining link state data reported in real time by the plurality of trusted agent nodes; updating the link information between the plurality of trusted agent nodes according to the obtained link state data of the plurality of trusted agent nodes, to estimate a reliable communication link according to the updated link information.

[0055] As an example, the link state data reported by the trusted agent node can include the transmission rate, bandwidth, and delay of the link between the trusted agent node and other trusted agent nodes.

[0056] As an example, the link information between the plurality of trusted agent nodes can be maintained in a matrix form, each row of the matrix corresponding to a trusted agent node, each column of the matrix corresponding to a trusted agent node, and each element in the matrix being the link information of the link between the trusted agent node corresponding to the row in which the element is located and the trusted agent node corresponding to the column in which the element is located.

[0057] As an example, the reliable communication link through which each computing cluster connects to the center server in turn via at least one trusted agent node can be estimated according to the geographic location information of each computing cluster, the geographic location information of the center server, the amount of data to be transmitted between the computing cluster and the center server (for example, the upper limit of the amount of data to be transmitted each time), and the link information between the plurality of trusted agent nodes.

[0058] As an example, the geographic location information of each computing cluster, the geographic location information of the center server, and the link information between the plurality of trusted agent nodes can be input into a pre-trained link recommendation model to obtain the reliable communication link through which the computing cluster connects to the center server output by the link recommendation model. As an example, the link recommendation model can be updated periodically.

[0059] Further, as an example, the geographic location information of each computing cluster, the geographic location information of the center server, the amount of data to be transmitted between the computing cluster and the center server, and the link information between the plurality of trusted agent nodes can be input into a pre-trained link recommendation model to obtain the reliable communication link through which the computing cluster connects to the center server output by the link recommendation model.

[0060] In step S30, the estimated reliable communication link through which each computing cluster connects to the center server is recommended to each computing cluster, so that each computing cluster uses the recommended reliable communication link to interact with the center server.

[0061] In addition, as an example, the communication link recommendation method for federated learning according to the example embodiments of the present disclosure can further include: monitoring the communication quality of the reliable communication link through which each computing cluster connects to the center server in real time; when it is monitored that the communication quality of the reliable communication link through which any one of the computing clusters connects to the center server satisfies a damaged condition, re-estimating the reliable communication link through which the computing cluster connects to the center server in turn via at least one trusted agent node according to the geographic location information of the computing cluster, the geographic location information of the center server, and the updated link information between the plurality of trusted agent nodes; recommending the re-estimated reliable communication link through which the computing cluster connects to the center server to the computing cluster, so that the computing cluster uses the re-recommended reliable communication link to interact with the center server.

[0062] As an example, the communication quality condition can include: delay, packet loss, disconnection, and the like. As an example, when the delay exceeds a delay threshold, it can be determined that the impaired condition is met; when packet loss occurs, it can be determined that the impaired condition is met; when the link is disconnected, it can be determined that the impaired condition is met.

[0063] As an example, the communication link recommendation method for federated learning according to the exemplary embodiments of the present disclosure can further include: for each of the plurality of computing clusters, estimating a reliable communication link (i.e., a reliable communication link connecting each other) in which one of the two computing clusters is connected to the other of the two computing clusters via at least one trusted proxy node in turn, according to the geographical location information of the two computing clusters, and the link information between the plurality of trusted proxy nodes; and then recommending the estimated reliable communication link connecting each other to the two computing clusters, so that the two computing clusters use the recommended reliable communication link connecting each other for data interaction.

[0064] As an example, the reliable communication link in which one of the two computing clusters is connected to the other of the two computing clusters via at least one trusted proxy node in turn can be estimated according to the geographical location information of the two computing clusters, the amount of data to be transmitted between the two computing clusters, and the link information between the plurality of trusted proxy nodes.

[0065] As an example, the geographical location information of the two computing clusters and the link information between the plurality of trusted proxy nodes can be input into a pre-trained link recommendation model to obtain a reliable communication link connecting each other of the two computing clusters output by the link recommendation model. Further, the geographical location information of the two computing clusters, the amount of data to be transmitted between the two computing clusters, and the link information between the plurality of trusted proxy nodes can be input into the pre-trained link recommendation model to obtain a reliable communication link connecting each other of the two computing clusters output by the link recommendation model.

[0066] As an example, the communication link recommendation method for federated learning according to the exemplary embodiments of the present disclosure can further include: monitoring the communication quality condition of the reliable communication link connecting each other recommended to each of the two computing clusters in real time; when it is monitored that the communication quality condition of the reliable communication link connecting each other recommended to any two computing clusters meets the impaired condition, estimating a reliable communication link in which one of the two computing clusters is connected to the other of the two computing clusters via at least one trusted proxy node in turn according to the geographical location information of the two computing clusters and the updated link information between the plurality of trusted proxy nodes; and recommending the re-estimated reliable communication link connecting each other to the two computing clusters, so that the two computing clusters use the re-recommended reliable communication link for data interaction.

[0067] Figure 2An example of a communication link recommendation method for federated learning according to an example embodiment of the present disclosure is shown.

[0068] As shown in Figure 2 , a plurality of trusted proxy nodes (i.e., trusted proxies in Figure 2 ) can be built in global area cloud services, IDC machine rooms, edge devices, etc., and the trusted proxy nodes can report proxy network data (e.g., transmission rate, bandwidth, latency, etc.) to a federated network coordination center in real time (as shown in step 1 in Figure 2 ); the federated network coordination center can construct a connectivity matrix of trusted proxy nodes available to each computing cluster of federated learning based on the proxy network data reported by the trusted proxy nodes in real time, and the federated network coordination engine can perform deduction and diagnosis of multi-hop network agents on this basis (as shown in step 2 in Figure 2 ); when the computing cluster of federated learning (i.e., the cluster in Figure 2 ) performs proxy service pairing registration (as shown in step 3 in Figure 2 ), the federated network coordination engine can train and recommend a list of access link points (e.g., sorted by reliability) through dimensions such as blocking, link carrying, reliability, latency, history, regional regulations, etc. (as shown in step 4 in Figure 2 ); the trusted proxy nodes start multi-hop interconnection based on trusted overlay to provide reliable communication links for the computing cluster (as shown in step 5 in Figure 2 ); once the communication link quality of the computing cluster is monitored to be impaired, dynamic adjustment of the link is performed (as shown in step 6 in Figure 2 ). Each row of the connectivity matrix can correspond to a trusted proxy node, and each column can correspond to a trusted proxy node. Each element in the matrix is the link information of the link between the trusted proxy node corresponding to the row and the trusted proxy node corresponding to the column.

[0069] According to the embodiments of the present disclosure, the link reliability problem of cross-wide area and cross-region federated learning is solved, and the overall process reliability and stability of global federated learning are improved; the access link points for federated learning parties can be recommended, and the federated learning parties can perform multi-hop interconnection based on trusted overlay for the respective recommended access link points; once the communication link quality is monitored to be impaired, dynamic adjustment is performed to maintain the reliability of the multi-party communication link and shield the underlying link changes.

[0070] Figure 3 A structural block diagram of a communication link recommendation device for federated learning according to an example embodiment of the present disclosure is shown.

[0071] As shown in Figure 3As shown, the communication link recommendation apparatus for federated learning according to the exemplary embodiments of the present disclosure comprises a location information acquisition unit 10, a link estimation unit 20, and a link recommendation unit 30.

[0072] Specifically, the location information acquisition unit 10 is configured to acquire geographic location information of a plurality of computing clusters and a center server.

[0073] The plurality of computing clusters are configured to perform federated learning for a same model based on local data, and the center server is configured to maintain parameters and / or parameter gradients of the same model.

[0074] The link estimation unit 20 is configured to, for each of the plurality of computing clusters, estimate a reliable communication link for the computing cluster to connect to the center server via at least one trusted proxy node in a multi-hop manner according to the geographic location information of the computing cluster, the geographic location information of the center server, and link information between a plurality of trusted proxy nodes in a constructed multi-hop network.

[0075] The link recommendation unit 30 is configured to recommend the estimated reliable communication link for each computing cluster to connect to the center server, so that each computing cluster uses the recommended reliable communication link to interact with the center server for data.

[0076] As an example, the link estimation unit 20 can be configured to input the geographic location information of each computing cluster, the geographic location information of the center server, and the link information between the plurality of trusted proxy nodes into a pre-trained link recommendation model to obtain a reliable communication link for the computing cluster to connect to the center server output by the link recommendation model.

[0077] As an example, the link estimation unit 20 can be configured to estimate a reliable communication link for each computing cluster to connect to the center server via at least one trusted proxy node in a multi-hop manner according to the geographic location information of the computing cluster, the geographic location information of the center server, an amount of data to be transmitted between the computing cluster and the center server, and the link information between the plurality of trusted proxy nodes.

[0078] As an example, the link information between the plurality of trusted proxy nodes can include at least one of the following information of a link between each two trusted proxy nodes: whether to block, blocking degree, link carrying capacity, reliability information, packet loss condition, latency, transmission rate, link history record, and whether to allow to establish a link to transmit data based on regional rules.

[0079] As an example, the communication link recommendation apparatus can further comprise a first link monitoring unit (not shown) configured to monitor in real time a communication quality of a reliable communication link recommended to each computing cluster to connect to the central server. Accordingly, the link estimation unit 20 can be further configured to, when it is monitored that the communication quality of the reliable communication link recommended to any one of the computing clusters to connect to the central server satisfies a damaged condition, re-estimate a reliable communication link of the computing cluster to connect to the central server in turn via at least one trusted proxy node based on the geographical location information of the computing cluster, the geographical location information of the central server, and the updated link information between the plurality of trusted proxy nodes; and the link recommendation unit 30 can be further configured to recommend the re-estimated reliable communication link of the computing cluster to connect to the central server to the computing cluster, so that the computing cluster uses the re-recommended reliable communication link to interact with the central server.

[0080] As an example, the plurality of trusted proxy nodes can be constructed on a public cloud.

[0081] As an example, each two of the plurality of trusted proxy nodes can directly establish a link or establish a link based on other routing nodes.

[0082] As an example, the link estimation unit 20 can be further configured to, for each two of the plurality of computing clusters, estimate a reliable communication link of one of the two computing clusters to connect to the other of the two computing clusters in turn via at least one trusted proxy node based on the geographical location information of the two computing clusters and the link information between the plurality of trusted proxy nodes; and the link recommendation unit 30 can be further configured to recommend the estimated reliable communication link of the two computing clusters to connect to each other to the two computing clusters, so that the two computing clusters use the recommended reliable communication link to interact with each other.

[0083] As an example, the link estimation unit 20 can be configured to input the geographical location information of the two computing clusters and the link information between the plurality of trusted proxy nodes into a pre-trained link recommendation model to obtain a reliable communication link of the two computing clusters to connect to each other output by the link recommendation model.

[0084] As an example, the link estimation unit 20 can be configured to estimate a reliable communication link of one of the two computing clusters to connect to the other of the two computing clusters in turn via at least one trusted proxy node based on the geographical location information of the two computing clusters, the amount of data to be transmitted between the two computing clusters, and the link information between the plurality of trusted proxy nodes.

[0085] As an example, the communication link recommendation apparatus can further include a second link monitoring unit (not shown) configured to monitor, in real time, a communication quality of the reliable communication link connecting each two computing clusters to each other. Accordingly, the link estimation unit 20 can be further configured to, when it is monitored that the communication quality of the reliable communication link connecting any two computing clusters to each other satisfies the damaged condition, estimate, according to the geographical location information of the two computing clusters and the updated link information between the plurality of trusted proxy nodes, a reliable communication link of one of the two computing clusters to the other of the two computing clusters via at least one trusted proxy node in turn. The link recommendation unit 30 can be further configured to recommend the two computing clusters to use the re-estimated reliable communication link to connect to each other, so that the two computing clusters use the re-recommended reliable communication link to interact with each other.

[0086] As an example, the communication link recommendation apparatus can further include a state data acquisition unit (not shown) and a link information updating unit (not shown). The state data acquisition unit is configured to acquire the link state data reported by the plurality of trusted proxy nodes in real time. The link information updating unit is configured to update the link information between the plurality of trusted proxy nodes according to the acquired link state data of the plurality of trusted proxy nodes, so as to estimate the reliable communication link according to the updated link information.

[0087] As an example, the link state data reported by the trusted proxy node can include a transmission rate, a bandwidth, and a time delay of the link between the trusted proxy node and other trusted proxy nodes.

[0088] It should be understood that the specific implementation of the communication link recommendation apparatus for federated learning according to the exemplary embodiments of the present disclosure can be implemented with reference to the related specific implementation described in combination with Figure 1 and Figure 2 and will not be described here again.

[0089] The units included in the communication link recommendation apparatus for federated learning according to the exemplary embodiments of the present disclosure can be respectively configured as software, hardware, firmware or any combination of the above for performing specific functions. For example, these units can correspond to dedicated integrated circuits, can also correspond to pure software codes, and can also correspond to software and hardware combined modules. In addition, one or more functions implemented by these units can also be uniformly performed by components in a physical entity device (for example, a processor, a client or a server, etc.).

[0090] It should be understood that the communication link recommendation method for federated learning according to exemplary embodiments of the present disclosure can be implemented by a program recorded on a computationally readable medium. For example, according to exemplary embodiments of the present disclosure, a computer-readable medium for recommending communication links for federated learning can be provided, wherein a computer program for performing the following method steps is recorded on the computer-readable medium: acquiring geographical location information of a plurality of computing clusters and a central server; for each of the plurality of computing clusters, estimating a reliable communication link for the computing cluster to sequentially connect to the central server via at least one trusted proxy node, based on the geographical location information of the computing cluster, the geographical location information of the central server, and link information between a plurality of trusted proxy nodes in a constructed multi-hop network; recommending the estimated reliable communication link to the central server to each computing cluster, so that each computing cluster uses the recommended reliable communication link to interact with the central server; wherein the plurality of computing clusters are used for federated learning of the same model based on local data, and the central server is used to maintain the parameters and / or parameter gradients of the same model.

[0091] The computer program in the aforementioned computer-readable medium can run in an environment deployed in computer devices such as clients, hosts, agent devices, and servers. It should be noted that the computer program can also be used to perform additional steps besides those described above, or to perform more specific processing while performing the above steps. The content of these additional steps and further processing has been described in reference to... Figure 1 and Figure 2 The above has already been described, and will not be repeated here to avoid repetition.

[0092] It should be noted that the communication link recommendation apparatus for federated learning according to the exemplary embodiments of this disclosure can rely entirely on the operation of a computer program to achieve the corresponding functions. That is, each unit corresponds to each step in the functional architecture of the computer program, so that the entire system is invoked through a specialized software package (e.g., a lib library) to achieve the corresponding functions.

[0093] On the other hand, the various units included in the communication link recommendation apparatus for federated learning according to the exemplary embodiments of this disclosure can also be implemented by hardware, software, firmware, middleware, microcode, or any combination thereof. When implemented in software, firmware, middleware, or microcode, the program code or code segment for performing the corresponding operation can be stored in a computer-readable medium such as a storage medium, so that a processor can perform the corresponding operation by reading and running the corresponding program code or code segment.

[0094] For example, the exemplary embodiments of the present disclosure can also be implemented as a computing device including a storage component and a processor, the storage component having stored therein a set of computer-executable instructions which, when executed by the processor, perform a communication link recommendation method for federated learning.

[0095] In particular, the computing device can be deployed in a server or a client, or on a node device in a distributed network environment. In addition, the computing device can be a PC computer, a tablet device, a personal digital assistant, a smart phone, a web application, or other device capable of executing the above-mentioned set of instructions.

[0096] Here, the computing device is not necessarily a single computing device, but can also be a collection of any device or circuit capable of executing the above-mentioned instructions (or set of instructions) alone or jointly. The computing device can also be part of an integrated control system or system manager, or can be configured as a portable electronic device that interfaces with local or remote (e.g., via wireless transmission) devices.

[0097] In the computing device, the processor can include a central processing unit (CPU), a graphics processing unit (GPU), a programmable logic device, a special-purpose processor system, a microcontroller, or a microprocessor. By way of example and not limitation, the processor can also include an analog processor, a digital processor, a microprocessor, a multi-core processor, a processor array, a network processor, etc.

[0098] Some of the operations described in the communication link recommendation method for federated learning according to the exemplary embodiments of the present disclosure can be implemented in software, some of the operations can be implemented in hardware, and in addition, these operations can also be implemented in a combination of software and hardware.

[0099] The processor can execute instructions or codes stored in one of the storage components, wherein the storage components can also store data. Instructions and data can also be sent and received via a network through a network interface device, wherein the network interface device can use any known transmission protocol.

[0100] The storage component can be integrated with the processor, for example, RAM or flash memory is arranged within an integrated circuit microprocessor, etc. In addition, the storage component can include a separate device, such as an external disk drive, a storage array, or other storage device that can be used by any database system. The storage component and the processor can be operatively coupled or can communicate with each other, for example, through an I / O port, a network connection, etc., so that the processor can read files stored in the storage component.

[0101] In addition, the computing device can also include a video display (such as a liquid crystal display) and a user interaction interface (such as a keyboard, a mouse, a touch input device, etc.). All components of the computing device can be connected to each other via a bus and / or a network.

[0102] The operations involved in the communication link recommendation method for federated learning according to the exemplary embodiments of the present disclosure can be described as various interconnected or coupled functional blocks or functional diagrams. However, these functional blocks or functional diagrams can be equally integrated as a single logical device or operate in non-exact boundaries.

[0103] For example, as described above, the computing device for communication link recommendation for federated learning according to the exemplary embodiments of the present disclosure can include a storage component and a processor, wherein a set of computer executable instructions are stored in the storage component, when the set of computer executable instructions are executed by the processor, the following steps are performed: obtaining geographical location information of a plurality of computing clusters and a center server; for each computing cluster in the plurality of computing clusters, estimating a reliable communication link of the computing cluster connecting to the center server via at least one trusted proxy node in sequence based on the geographical location information of the computing cluster, the geographical location information of the center server, and link information between a plurality of trusted proxy nodes in a multi-hop network that has been constructed; recommending the estimated reliable communication link of each computing cluster connecting to the center server to each computing cluster, so that each computing cluster uses the recommended reliable communication link to interact with the center server for data; wherein the plurality of computing clusters are used to perform federated learning on a same model based on local data, and the center server is used to maintain parameters and / or parameter gradients of the same model.

[0104] The above describes various exemplary embodiments of the present disclosure, and it should be understood that the above description is only exemplary and is not exhaustive, and the present disclosure is not limited to the disclosed exemplary embodiments. Many modifications and changes are obvious to those skilled in the art without departing from the scope and spirit of the present disclosure. Therefore, the protection scope of the present disclosure should be subject to the scope of claims.

Claims

1. A communication link recommendation method for federated learning, comprising: obtaining geographical location information of a plurality of computing clusters and a center server; for each computing cluster in the plurality of computing clusters, estimating a reliable communication link of the computing cluster to the center server via at least one trusted proxy node in a multi-hop manner according to the geographical location information of the computing cluster, the geographical location information of the center server, and link information between a plurality of trusted proxy nodes in a multi-hop network that has been constructed; recommending the estimated reliable communication link of each computing cluster to the center server to each computing cluster, so that each computing cluster uses the recommended reliable communication link to interact with the center server for data; wherein the plurality of computing clusters perform federated learning based on local data for a same model, and the center server maintains parameters and / or parameter gradients of the same model; wherein the federated learning is collaborative training; the center server receives a request of reading the parameters of the same model sent by each computing cluster through the reliable communication link of each computing cluster, and sends the parameters of the same model to each computing cluster through the reliable communication link of each computing cluster in response to the request, receives the gradients of the parameters sent by each computing cluster through the reliable communication link of each computing cluster, and updates the parameters based on the gradients of the parameters, wherein the gradients of the parameters are obtained by each computing cluster based on the parameters requested from the center server and local training data.

2. The communication link recommendation method of claim 1, wherein, The step of estimating the reliable communication link of each computing cluster to the center server via at least one trusted proxy node in a multi-hop manner comprises: inputting the geographical location information of each computing cluster, the geographical location information of the center server, and the link information between the plurality of trusted proxy nodes into a pre-trained link recommendation model to obtain a reliable communication link of the computing cluster to the center server output by the link recommendation model.

3. The communication link recommendation method of claim 1, wherein, The step of estimating the reliable communication link of each computing cluster to the center server via at least one trusted proxy node in a multi-hop manner comprises: estimating the reliable communication link of each computing cluster to the center server via at least one trusted proxy node in a multi-hop manner according to the geographical location information of each computing cluster, the geographical location information of the center server, the amount of data to be transmitted between each computing cluster and the center server, and the link information between the plurality of trusted proxy nodes.

4. The communication link recommendation method of claim 1, wherein, The link information between the plurality of trusted proxy nodes comprises at least one of the following information of the link between each two trusted proxy nodes: whether to block, blocking degree, link carrying capacity, reliability information, packet loss condition, latency, transmission rate, link history record, and whether to allow to establish a link to transmit data based on regional rules.

5. The communication link recommendation method of claim 1, wherein, The communication link recommendation method further comprises: monitoring the communication quality of the recommended reliable communication link of each computing cluster to the center server in real time; when it is monitored that the communication quality of the recommended reliable communication link connecting to the center server of any one computing cluster meets the damaged condition, re-estimating the reliable communication link of the computing cluster connecting to the center server via at least one trusted agent node in turn according to the geographical position information of the computing cluster, the geographical position information of the center server, and the updated link information between the plurality of trusted agent nodes; recommending the re-estimated reliable communication link connecting to the center server to the computing cluster, so that the computing cluster uses the re-recommended reliable communication link to interact with the center server.

6. The communication link recommendation method of claim 1, wherein, The plurality of trusted agent nodes are constructed on a public cloud. And / or, each two trusted agent nodes in the plurality of trusted agent nodes directly establish a link or establish a link based on other routing nodes.

7. The communication link recommendation method of claim 1, wherein, The communication link recommendation method further comprises: For each two computing clusters in the plurality of computing clusters, estimating a reliable communication link of one computing cluster in the two computing clusters connecting to another computing cluster in the two computing clusters via at least one trusted agent node in turn according to the geographical position information of the two computing clusters and the link information between the plurality of trusted agent nodes; recommending the estimated reliable communication link connecting to each other to the two computing clusters, so that the two computing clusters use the recommended reliable communication link connecting to each other to interact with each other.

8. The communication link recommendation method of claim 7, wherein, The step of estimating the reliable communication link of one computing cluster in the two computing clusters connecting to another computing cluster in the two computing clusters via at least one trusted agent node in turn comprises: inputting the geographical position information of the two computing clusters and the link information between the plurality of trusted agent nodes into a pre-trained link recommendation model to obtain the reliable communication link of the two computing clusters connecting to each other output by the link recommendation model.

9. The communication link recommendation method of claim 7, wherein, The step of estimating the reliable communication link of one computing cluster in the two computing clusters connecting to another computing cluster in the two computing clusters via at least one trusted agent node in turn comprises: estimating the reliable communication link of one computing cluster in the two computing clusters connecting to another computing cluster in the two computing clusters via at least one trusted agent node in turn according to the geographical position information of the two computing clusters, the amount of data to be transmitted between the two computing clusters, and the link information between the plurality of trusted agent nodes.

10. The communication link recommendation method of claim 7, wherein, The communication link recommendation method further comprises: monitoring the communication quality of the recommended reliable communication link connecting to each other of each two computing clusters in real time; when it is monitored that the communication quality of the recommended reliable communication link connecting to each other of any two computing clusters meets the damaged condition, estimating the reliable communication link of one computing cluster in the two computing clusters connecting to another computing cluster in the two computing clusters via at least one trusted agent node in turn according to the geographical position information of the two computing clusters and the updated link information between the plurality of trusted agent nodes; re-estimating reliable communication links connecting the two computing clusters to each other, so that the two computing clusters use the re-estimated reliable communication links for data interaction.

11. The communication link recommendation method of claim 1, wherein, The communication link recommendation method further includes: obtaining link state data reported by the plurality of trusted agent nodes in real time; updating link information between the plurality of trusted agent nodes according to the obtained link state data of the plurality of trusted agent nodes, so as to estimate reliable communication links according to the updated link information.

12. The communication link recommendation method of claim 11, wherein The link state data reported by the trusted agent nodes includes: transmission rate, bandwidth, and latency of the link between the trusted agent node and other trusted agent nodes.

13. A communication link recommendation device for federated learning, comprising: a location information obtaining unit configured to obtain geographic location information of a plurality of computing clusters and a center server; a link estimation unit configured to, for each computing cluster in the plurality of computing clusters, estimate a reliable communication link connecting the computing cluster to the center server via at least one trusted agent node in a multi-hop manner according to the geographic location information of the computing cluster, the geographic location information of the center server, and link information between a plurality of trusted agent nodes in a multi-hop network that has been constructed; a link recommendation unit configured to recommend the estimated reliable communication link connecting each computing cluster to the center server to each computing cluster, so that each computing cluster uses the recommended reliable communication link to interact with the center server for data; wherein the plurality of computing clusters are used to perform federated learning on a same model based on local data, and the center server is used to maintain parameters and / or parameter gradients of the same model; wherein the federated learning is collaborative training; the center server receives a request for reading parameters of the same model sent by each computing cluster through the reliable communication link of each computing cluster, and in response to the request, sends the parameters of the same model to the corresponding computing cluster through the reliable communication link of each computing cluster, and receives the gradient of the parameters sent by each computing cluster through the corresponding reliable communication link, and updates the parameters based on the gradient of the parameters, wherein the gradient of the parameters is obtained by each computing cluster based on the parameters requested from the center server and local training data.

14. The communication link recommendation device of claim 13, wherein, The link estimation unit is configured to input the geographic location information of each computing cluster, the geographic location information of the center server, and the link information between the plurality of trusted agent nodes into a pre-trained link recommendation model, to obtain a reliable communication link connecting the computing cluster to the center server output by the link recommendation model.

15. The communication link recommendation device of claim 13, wherein, The link estimation unit is configured to estimate a reliable communication link connecting each computing cluster to the center server via at least one trusted agent node in a multi-hop manner according to the geographic location information of each computing cluster, the geographic location information of the center server, the amount of data to be transmitted between the computing cluster and the center server, and the link information between the plurality of trusted agent nodes.

16. The communication link recommendation device of claim 13, wherein, The link information between the plurality of trusted agent nodes includes at least one of the following information of a link between each two trusted agent nodes: Whether to block, blocking degree, link carrying capacity, reliability information, packet loss, latency, transmission rate, link history, whether to allow to establish link transmission data based on regional rules.

17. The communication link recommendation device of claim 13, wherein, The communication link recommendation device further includes: A first link monitoring unit configured to monitor in real time a communication quality condition of a reliable communication link recommended to each computing cluster to connect to the center server; The link estimation unit is further configured to, when it is monitored that the communication quality condition of the reliable communication link recommended to any one computing cluster to connect to the center server meets the damaged condition, re-estimate a reliable communication link of the computing cluster to connect to the center server in turn via at least one trusted agent node according to geographical position information of the computing cluster, geographical position information of the center server and updated link information between the plurality of trusted agent nodes; The link recommendation unit is further configured to recommend the re-estimated reliable communication link of the computing cluster to connect to the center server, so that the computing cluster uses the re-recommended reliable communication link to interact with the center server.

18. The communication link recommendation apparatus of claim 13, wherein, The plurality of trusted agent nodes are constructed on a public cloud. And / or, each two trusted agent nodes in the plurality of trusted agent nodes directly establish a link or establish a link based on other routing nodes.

19. The communication link recommendation apparatus of claim 13, wherein, The link estimation unit is further configured to, for each two computing clusters in the plurality of computing clusters, estimate a reliable communication link of one computing cluster in the two computing clusters to connect to another computing cluster in the two computing clusters in turn via at least one trusted agent node according to geographical position information of the two computing clusters and link information between the plurality of trusted agent nodes. The link recommendation unit is further configured to recommend the estimated reliable communication link of the two computing clusters to connect to each other, so that the two computing clusters use the recommended reliable communication link to connect to each other to interact with each other.

20. The communication link recommendation device of claim 19, wherein, The link estimation unit is configured to input the geographical position information of the two computing clusters and the link information between the plurality of trusted agent nodes into a pre-trained link recommendation model to obtain a reliable communication link of the two computing clusters to connect to each other output by the link recommendation model.

21. The communication link recommendation device of claim 19, wherein, The link estimation unit is configured to estimate a reliable communication link of one computing cluster in the two computing clusters to connect to another computing cluster in the two computing clusters in turn via at least one trusted agent node according to geographical position information of the two computing clusters, an amount of data to be transmitted between the two computing clusters and link information between the plurality of trusted agent nodes.

22. The communication link recommendation device of claim 19, wherein, The communication link recommendation device further includes: A second link monitoring unit configured to monitor in real time a communication quality condition of a reliable communication link recommended to each two computing clusters to connect to each other; The link estimation unit is further configured to, when it is monitored that the communication quality condition of the reliable communication link recommended to any one computing cluster to connect to the center server meets the damaged condition, re-estimate a reliable communication link of the computing cluster to connect to the center server in turn via at least one trusted agent node according to geographical position information of the computing cluster, geographical position information of the center server and updated link information between the plurality of trusted agent nodes; The link estimation unit is further configured to, when it is monitored that the communication quality of the reliable communication link recommended by any two computing clusters to connect each other meets the damaged condition, estimate a reliable communication link in which one of the two computing clusters is connected to the other of the two computing clusters via at least one trusted agent node in multiple hops according to the geographical position information of the two computing clusters and the updated link information between the plurality of trusted agent nodes. The link recommendation unit is further configured to recommend the two computing clusters to use the re-estimated reliable communication link to connect each other, so that the two computing clusters use the re-recommended reliable communication link to interact data.

23. The communication link recommendation device of claim 13, wherein, The communication link recommendation device further comprises: a state data acquisition unit configured to acquire link state data reported by the plurality of trusted agent nodes in real time; a link information updating unit configured to update the link information between the plurality of trusted agent nodes according to the acquired link state data of the plurality of trusted agent nodes, so as to estimate a reliable communication link according to the updated link information.

24. The communication link recommendation device of claim 23, wherein The link state data reported by the trusted agent node includes a transmission rate, a bandwidth, and a time delay of a link between the trusted agent node and other trusted agent nodes.

25. A system comprising at least one computing device and at least one storage device storing instructions, wherein, The instructions, when executed by the at least one computing device, cause the at least one computing device to perform the communication link recommendation method for federated learning of any one of claims 1 to 12.

26. A computer-readable storage medium storing instructions, wherein, The instructions, when executed by the at least one computing device, cause the at least one computing device to perform the communication link recommendation method for federated learning of any one of claims 1 to 12.

Images