Interworking method, device, equipment, medium and product between federal learning platforms

Abstract

The present disclosure provides an intercommunication method and device between federal learning platforms, equipment, medium and product, relates to the technical field of artificial intelligence, in particular to the technical field of machine learning and federal learning. The specific implementation scheme is: in response to the opening instruction of the cross-platform job task in the local initiating node in the local federal learning platform, the opposite federal learning platform matched with the cross-platform job task is determined; in each local candidate node, the local collaborative node matched with the opposite federal learning platform is determined; the global dependency graph matched with the cross-platform job task sent by the local initiating node is obtained, and the target dependency graph is obtained by translating the global dependency graph when the global dependency graph meets the translation condition; the target dependency graph is sent to the local collaborative node. The scheme of the present disclosure solves the problem of intercommunication between federal learning platforms at the present stage, can realize the intercommunication between different federal learning platforms, and improves the processing efficiency of the job task.

Description

Technical Field

[0001] This disclosure relates to the field of artificial intelligence technology, and more particularly to the field of machine learning and federated learning technology, specifically to a method, apparatus, device, medium and product for interoperability between federated learning platforms. Background Technology

[0002] Federated learning is essentially a distributed machine learning framework that achieves data sharing and collaborative modeling while ensuring data privacy, security, and legal compliance. Its core idea is that when multiple data sources participate in model training, the original data does not need to be transferred; instead, the model is trained jointly only by exchanging intermediate parameters, and the original data does not need to leave the local machine. This approach achieves a balance between data privacy protection and data sharing analysis, namely a data application model that is "usable but not visible."

[0003] At present, different vendors have developed different federated learning platforms based on different implementation methods and underlying frameworks, which can be used to process different federated learning tasks. Summary of the Invention

[0004] This disclosure provides a method, apparatus, device, medium, and product for interoperability between federated learning platforms.

[0005] According to one aspect of this disclosure, a method for interoperability between federated learning platforms is provided, executed by an interoperability service built into the federated learning platform, comprising:

[0006] In response to the command to start a cross-platform job task in the local initiating node of the local federated learning platform, determine the counterpart federated learning platform that matches the cross-platform job task.

[0007] Among the candidate nodes of the local party, determine the local party's collaborative node that matches the other party's federated learning platform;

[0008] Obtain the global dependency graph sent by the local initiating node that matches the cross-platform job task, and when the global dependency graph meets the translation conditions, translate the global dependency graph to obtain the target dependency graph;

[0009] The target dependency graph is sent to the local collaborative node, so that the local collaborative node can jointly execute the cross-platform job task with the other party's federated learning platform based on the target dependency graph.

[0010] According to another aspect of this disclosure, an interoperability device between federated learning platforms is provided, executed by an interoperability service built into the federated learning platform, comprising:

[0011] The opposite federal learning platform determination module is configured to determine, in response to an opening instruction of a cross-platform job task in a self federal learning platform, an opposite federal learning platform matched with the cross-platform job task;

[0012] The self collaborative node determination module is configured to determine, in each self candidate node, a self collaborative node matched with the opposite federal learning platform;

[0013] The target dependency graph determination module is configured to obtain a global dependency graph sent by the self initiating node and matched with the cross-platform job task, and translate the global dependency graph into a target dependency graph when the global dependency graph meets a translation condition;

[0014] The target dependency graph sending module is configured to send the target dependency graph to the self collaborative node, so that the self collaborative node performs the cross-platform job task together with the opposite federal learning platform based on the target dependency graph.

[0015] According to another aspect of the present disclosure, an electronic device is provided, comprising:

[0016] at least one processor; and

[0017] a memory connected with the at least one processor in communication; wherein,

[0018] The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method described in any embodiment of the present disclosure.

[0019] According to another aspect of the present disclosure, a non-transitory computer readable storage medium storing computer instructions is provided, wherein the computer instructions are used to make the computer perform the method described in any embodiment of the present disclosure.

[0020] According to another aspect of the present disclosure, a computer program product is provided, comprising a computer program which, when executed by a processor, implements the method described in any embodiment of the present disclosure.

[0021] It should be understood that the contents described in this part are not intended to identify key or important features of the embodiments of the present disclosure, nor to limit the scope of the present disclosure. Other features of the present disclosure will become apparent through the following description. BRIEF DESCRIPTION OF DRAWINGS

[0022] The accompanying drawings are used to better understand the present scheme, and do not constitute a limitation on the present disclosure. Among them:

[0023] Figure 1is a schematic diagram of an interworking method between federated learning platforms according to an embodiment of the present disclosure;

[0024] Figure 2 is a schematic diagram of another interworking method between federated learning platforms according to an embodiment of the present disclosure;

[0025] Figure 3 is a schematic diagram of still another interworking method between federated learning platforms according to an embodiment of the present disclosure;

[0026] Figure 4 is a schematic diagram of yet another interworking method between federated learning platforms according to an embodiment of the present disclosure;

[0027] Figure 5a is a translation scenario applicable to the interconnection and interworking service according to an embodiment of the present disclosure;

[0028] Figure 5b is an extension scenario applicable to the interconnection and interworking service according to an embodiment of the present disclosure;

[0029] Figure 5c is a schematic diagram of a translation mode according to an embodiment of the present disclosure;

[0030] Figure 5d is a schematic diagram of a direct drive mode according to an embodiment of the present disclosure;

[0031] Figure 6 is a structural schematic diagram of an interworking device between federated learning platforms according to an embodiment of the present disclosure;

[0032] Figure 7 is a block diagram of an electronic device for implementing the interworking method between federated learning platforms according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0033] Exemplary embodiments of the present disclosure are described below with reference to the accompanying drawings, which include various details of the embodiments of the present disclosure to assist in understanding, and should be considered as merely exemplary. Thus, those of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the present disclosure. Also, for the sake of brevity and clarity, descriptions of well-known functions and constructions are omitted from the following description.

[0034] Figure 1 is a schematic diagram of an interworking method between federated learning platforms according to an embodiment of the present disclosure, which can be applicable to implement data interworking between heterogeneous federated learning platforms, and further enable heterogeneous federated learning platforms to cooperatively complete cross-platform job tasks. The method can be executed by an interconnection and interworking service built in the federated learning platform of the present embodiment; specifically, referenceFigure 1 The method specifically comprises the following steps.

[0035] S110, in response to a starting instruction of a cross-platform job task in a self-side initiating node in a self-side federated learning platform, determining a counter-party federated learning platform matched with the cross-platform job task.

[0036] The self-side federated learning platform can be any version of federated learning platform developed by any manufacturer, for example, FATE federated learning platform, FedLearner federated learning platform, PaddleFL federated learning platform, or Angel powerFL federated learning platform, etc., which is not limited in the embodiment.

[0037] In the embodiment, the self-side initiating node in the self-side federated learning platform can be a participant in the self-side federated learning platform, which can receive different job tasks (for example, data processing tasks, model training tasks, or model prediction tasks, etc., which are not limited in the embodiment), and translate each task into a global dependency graph corresponding to it. Further, the global dependency graph or part of the global dependency graph can be sent to the interconnection service built in the self-side federated learning platform for subsequent processing.

[0038] In the embodiment, the cross-platform job task can also be any task, for example, a model training task, a natural language processing task, or a face recognition task that requires the self-side federated learning platform and the second federated learning platform to jointly process.

[0039] In the embodiment, the counter-party federated learning platform is a platform that jointly processes the cross-platform job task with the self-side federated learning platform, which can be determined by the task information of the cross-platform job task, or can be determined by the developer, which is not limited in the embodiment.

[0040] It should be noted that the self-side federated learning platform involved in the embodiment has an interconnection service built in, and the interconnection service supports a standard interconnection protocol. It can be understood that the standard interconnection protocol is an independent running process, which can use Remote Procedure Call (RPC) as the interconnection protocol. In the embodiment, the counter-party federated learning platform that cooperates with the self-side federated learning platform to process the federated learning task can be configured with an interconnection service, or can not be configured with an interconnection service, which is not limited in the embodiment.

[0041] In an optional implementation manner of the embodiment, the interconnection service built in the self-side federated learning platform can further determine the counter-party federated learning platform matched with the cross-platform job task when the cross-platform job task in the self-side initiating node in the self-side federated learning platform starts to execute.

[0042] Exemplarily, in the present embodiment, when the interconnection and interworking service receives the start of execution of the cross-platform job task in the self-side initiating node, the opposite-side federated learning platform can be determined according to the task information of the cross-platform job task; exemplarily, the task information of the cross-platform job task can include: the type and version information of the federated learning platform of the job task, the data required for executing the cross-platform job task, etc., which are not limited in the present embodiment.

[0043] S120, in each self-side candidate node, a self-side cooperative node matched with the opposite-side federated learning platform is determined.

[0044] In an optional implementation manner of the present embodiment, after the interconnection and interworking service determines the opposite-side federated learning platform matched with the cross-platform job task, the interconnection and interworking service can further determine the self-side cooperative node matched with the opposite-side federated learning platform in each self-side candidate node of the interconnection and interworking service; it can be understood that in the present embodiment, the number of self-side candidate nodes included in the interconnection and interworking service is not fixed, which can include 5 self-side candidate nodes, 10 self-side candidate nodes, or 50 self-side candidate nodes, which are not limited in the present embodiment.

[0045] Optionally, in the present embodiment, after the opposite-side federated learning platform is determined, the version information of the opposite-side federated learning platform can be further queried, and then the self-side cooperative node matched with the opposite-side federated learning platform is determined from each self-side candidate node according to the queried version information; it can be understood that in the present embodiment, the self-side cooperative node can cooperatively process the cross-platform job task involved in the present embodiment with the node device in the opposite-side federated learning platform.

[0046] S130, a global dependency graph matched with the cross-platform job task sent by the self-side initiating node is acquired, and when the global dependency graph meets the translation condition, a target dependency graph is obtained by translating the global dependency graph.

[0047] In the embodiment, the global dependency graph matched with the cross-platform job task can be a Directed Acyclic Graph (DAG) matched with the cross-platform job task, which is a finite directed graph without a directed cycle. Specifically, it is composed of a finite number of vertices and directed edges, each directed edge points from one vertex to another vertex; it is impossible to return to the original vertex through the directed edges from any vertex V of the DAG graph. It can be understood that it is impossible to finally loop back to V along the ordered edges starting from any vertex V of the DAG graph. In the embodiment, the global dependency graph matched with the cross-platform job task can also be a Domain-Specific Language (DSL) matched with the cross-platform job task, that is, the cross-platform job task is described through a DSL statement; in the embodiment, the global dependency graph matched with the cross-platform job task can also be a UI canvas, that is, the cross-platform job task is described directly through a DAG description file.

[0048] In an optional implementation manner of the embodiment, after determining the self-side collaborative node, the global dependency graph matched with the cross-platform job task sent by the self-side initiating node can be acquired, and it is determined whether the global dependency graph meets the translation condition; when the global dependency graph meets the translation condition, the global dependency graph can be translated into a target dependency graph; it can be understood that the target dependency graph is a dependency graph that can be understood by the self-side collaborative node.

[0049] Optionally, in the embodiment, after starting the cross-platform job task, the self-side initiating node can generate a global dependency graph matched with the cross-platform task, and directly send the global dependency graph to the interconnection service, so as to subsequently call the self-side collaborative node through the interconnection service to translate the cross-platform job task; the self-side initiating node can also send part of branches of the global dependency graph matched with the cross-platform task to the interconnection service, so as to subsequently call the self-side collaborative node through the interconnection service to process the cross-platform job task, thereby bundling the self-side initiating node and the self-side collaborative node and enhancing the capability of the self-side initiating node.

[0050] Further, after receiving the global dependency graph sent by the target node, the interconnection service can determine whether the global dependency graph meets the translation condition; in the embodiment, whether the global dependency graph meets the translation condition can be determined by a developer, or whether the global dependency graph meets the translation condition can be determined according to the complexity of the global dependency graph; for example, before processing the cross-platform job task, the developer can determine whether the global dependency graph matched with the cross-platform task meets the translation condition according to the characteristics of the self-side federated learning platform; if the developer determines that the global dependency graph meets the translation condition, the interconnection service can translate the received global dependency graph, thereby obtaining a target dependency graph.

[0051] In another optional implementation of the embodiment, if the developer determines that the global dependency graph does not meet the translation condition, the interconnection service can parse the received global dependency graph to obtain the local task and the opposite task corresponding to each task node, and sequentially forward the local task and the opposite task to realize effective processing of each task. It can be understood that this process is a static process.

[0052] S140, send the target dependency graph to the local collaborative node, so that the local collaborative node performs the cross-platform job task with the opposite federated learning platform based on the target dependency graph.

[0053] In an optional implementation of the embodiment, when the global dependency graph received by the interconnection service matches the cross-platform job task and meets the translation condition, after the global dependency graph is translated into the target dependency graph, the target dependency graph can be sent to the local collaborative node. For example, the target dependency graph can be sent to the plug-in of the local collaborative node, so that the plug-in in the local collaborative node processes the target dependency graph subsequently. It can be understood that in the embodiment, the interconnection service can translate the global dependency graph into a form that can be understood by the local alternative node. In this way, after the local collaborative node receives the target dependency graph, it does not need to convert the target dependency graph and can directly understand and analyze the target dependency graph.

[0054] Optionally, in the embodiment, after the local collaborative node receives the target dependency graph sent by the interconnection service, it can further understand and analyze the target dependency graph to obtain the local task and the opposite task corresponding to each node task in the target dependency graph, and sequentially send each opposite task to the node device corresponding to the local collaborative node in the opposite federated learning platform according to the order of each node task, so that the node device corresponding to the local collaborative node in the opposite federated learning platform processes each opposite task in turn. Further, the data corresponding to each local task can be sequentially obtained and processed. In this way, the local federated learning platform and the opposite federated learning platform cooperatively process the cross-platform job task.

[0055] The scheme of the embodiment provides a basis for intercommunication between the local federal learning platform and the opposite federal learning platform by determining the opposite federal learning platform matched with the cross-platform job task and determining the local collaborative node matched with the opposite federal learning platform in each local candidate node in response to the starting instruction of the cross-platform job task in the local initiating node of the local federal learning platform. Further, the global dependency graph matched with the cross-platform job task sent by the local initiating node can be acquired, and the target dependency graph is obtained by translating the global dependency graph when the global dependency graph meets the translation condition. The target dependency graph is sent to the local collaborative node, and the cross-platform job task can be executed by the local collaborative node and the opposite federal learning platform based on the target dependency graph, thereby solving the complex intercommunication problem between the federal learning platforms at the present stage, realizing the intercommunication between different federal learning platforms, and improving the processing efficiency of the job task.

[0056] Figure 2 FIG. 2 is a schematic diagram of another method for intercommunication between federal learning platforms according to an embodiment of the present disclosure. The embodiment is a further refinement of the above technical solution, and the technical solution in the embodiment can be combined with each optional solution in one or more of the above embodiments. As shown in FIG. 2, the method for intercommunication between federal learning platforms includes the following steps. Figure 2

[0057] S210, in response to a starting instruction of a cross-platform job task in a local initiating node of a local federal learning platform, querying task information of the cross-platform job task, and determining an opposite federal learning platform matched with the cross-platform job task according to the task information.

[0058] In the embodiment, the task information of the cross-platform job task can include the type of the cross-platform job task, the task processing target, the information of two or more federal learning platforms participating in the cross-platform job task, and the data or data labeling method required for processing the cross-platform job task, etc., which are not limited in the embodiment.

[0059] In an optional implementation manner of the embodiment, the interconnection and intercommunication service built in the local federal learning platform can further query the task information of the cross-platform job task when receiving the starting execution of the cross-platform job task in the local initiating node of the local federal learning platform, and determine the opposite federal learning platform matched with the cross-platform job task according to the queried task information.

[0060] For example, in the embodiment, by querying the task information of the cross-platform job task, it is determined that two federal learning platforms are required to process the task, which are the local federal learning platform developed by manufacturer A and the second federal learning platform developed by manufacturer B. Therefore, the opposite federal learning platform matched with the cross-platform job task is determined to be the second federal learning platform developed by manufacturer B.​

[0061] The advantage of this setup is that it provides a basis for quickly identifying our own collaborative nodes and adapting communication with the other party's federated learning platform.

[0062] S220. Obtain the version information of the other party's federated learning platform and determine the local collaborative node that matches the version information; determine the local collaborative node from the local candidate nodes and select the local collaborative node.

[0063] The version information of the other party's federated learning platform may include the version information of the node devices in the other party's federated learning platform. For example, the version information of the node devices in the other party's federated learning platform may be FATE 1.7 or FATE 1.8, etc., and this embodiment does not limit it.

[0064] In an optional implementation of this embodiment, after determining the counterpart federated learning platform that matches the cross-platform job task, the version information of the counterpart federated learning platform can be further obtained, and the information of the local collaborative node can be determined based on the version information of the counterpart federated learning platform; furthermore, the local collaborative node can be located among the local candidate nodes of the local federated learning platform, and the located local collaborative node can be selected.

[0065] Optionally, in this embodiment, after determining the counterpart federated learning platform that matches the cross-platform task, the version information of the counterpart federated learning platform can be further obtained, and then the information of the local collaborative node can be determined based on the version information of the counterpart federated learning platform; further, the node device corresponding to the information of the local collaborative node can be determined from all local candidate nodes in the local federated learning platform, and the node device is the local collaborative node involved in this embodiment; further, the local collaborative node can be determined.

[0066] The advantage of this setup is that by identifying the collaborating nodes, a basis can be provided for the subsequent joint completion of cross-platform tasks by both the local and external federated learning platforms, thus establishing a bridge between the local and external federated learning platforms.

[0067] S230. Obtain the global dependency graph sent by the initiating node that matches the cross-platform job task.

[0068] S240. Determine whether the global dependency graph meets the translation conditions based on its complexity.

[0069] In an optional implementation of this embodiment, after receiving the global dependency graph sent by the initiating node, the interconnection service can further determine whether the global dependency graph meets the translation conditions.

[0070] Optionally, in the embodiment, whether the global dependency graph meets the translation condition can be determined according to the complexity of the global dependency graph; when the complexity meets a preset requirement, it is determined that the global dependency graph meets the translation condition; the preset requirement can be whether the complexity of the global dependency graph is greater than a preset complexity threshold; the preset complexity threshold can be 0.8, 0.9, or 0.95, etc., which is not limited in the embodiment; for example, when the complexity of the global dependency graph is greater than the preset complexity threshold, it can be determined that the global dependency graph does not meet the translation condition; when the complexity of the global dependency graph is less than or equal to the preset complexity threshold, it is determined that the global dependency graph meets the translation condition.

[0071] Optionally, in the embodiment, whether the global dependency graph meets the translation condition can also be determined according to the indication information of the developer; for example, the developer can actively determine whether the global dependency graph matched with the cross-platform job task meets the translation condition before the cross-platform job task is started, and send the determination result to the interconnection service; the interconnection service can determine whether the global dependency graph needs to be translated after being received according to the information sent by the developer.

[0072] The advantage of such setting is that the cross-platform job task can be effectively and optimally processed, and the processing accuracy of the cross-platform job task is improved.

[0073] S250, when the global dependency graph meets the translation condition, the global dependency graph is translated to obtain a target dependency graph, and the target dependency graph is sent to the self-side collaborative node.

[0074] In an optional implementation manner of the embodiment, when the interconnection service determines that the global dependency graph sent by the self-side initiating node meets the translation condition, the interconnection service can convert the global dependency graph matched with the cross-platform job task into a target dependency graph matched with the adapter of the self-side collaborative node according to the semantic through the scheduling service inside the interconnection service, that is, the global dependency graph is converted into a job representation form matched with the collaborative federated learning, so that the self-side collaborative node and the opposite federated learning platform can be cooperated to process the cross-platform job task subsequently.

[0075] In an optional implementation manner of the embodiment, when the global dependency graph meets the translation condition, the global dependency graph is translated to obtain a target dependency graph, which can include: determining a dependency graph form matched with the adapter of the self-side collaborative node; and translating the global dependency graph according to the dependency graph form to obtain the target dependency graph.

[0076] Optionally, in the embodiment, when the interconnection and interworking service determines that the global dependency graph sent by the local initiating node meets the translation condition, the interconnection and interworking service can determine the dependency graph form matched by the adapter of the local collaborative node, that is, determine the representation form of the local collaborative node for the job task, and then convert the global dependency graph according to the representation form of the local collaborative node for the job task, and obtain the target dependency graph corresponding to the representation form of the local collaborative node for the job task.

[0077] The advantage of such arrangement is that by converting the global dependency graph matched with the cross-platform job task through the interconnection and interworking service, the job task representation form adapted to the back-end federated learning platform (that is, the opposite federated learning platform) can be obtained, so that the subsequent local collaborative node and the opposite federated learning platform can cooperate to process the cross-platform job task.

[0078] S260, when it is determined that the global dependency graph does not meet the translation condition, the global dependency graph is parsed to obtain each local task and each opposite task; each local task is sent to the local collaborative node, and each opposite task is sent to the interconnection and interworking service in the opposite federated learning platform.

[0079] In an optional implementation manner of the embodiment, when the interconnection and interworking service determines that the global dependency graph sent by the local initiating node does not meet the translation condition, the interconnection and interworking service can directly parse the global dependency graph matched with the cross-platform job task into specific job node tasks through the scheduling service inside the interconnection and interworking service, for example, the local task and the opposite task corresponding to each job node task can be obtained; further, each local task can be sent to the local collaborative node for processing, and each opposite task can be sent to the interconnection and interworking service in the opposite federated learning platform, and then each opposite task can be forwarded to the corresponding node device for processing through the interconnection and interworking service in the opposite federated learning platform.

[0080] For example, in the embodiment, when the developer determines that the face recognition model training cross-platform job task does not meet the translation condition, the interconnection and interworking service can further parse the global dependency graph matched with the face recognition model training cross-platform job task through the internal scheduling service after receiving the global dependency graph, for example, ten job node tasks and the local task and the opposite task matched with each node task are parsed; further, ten local tasks can be sent to the local collaborative node for processing, and ten opposite tasks can be sent to the interconnection and interworking service in the opposite federated learning platform for processing, which realizes the cooperative processing of the local federated learning platform and the federated learning platform for the face recognition model training cross-platform job task.

[0081] The scheme of the embodiment can further analyze the global dependency graph to obtain each self task and each opposite task when it is determined that the global dependency graph does not satisfy the translation condition, and send each self task to the self collaborative node and send each opposite task to the interconnection and interworking service in the opposite federated learning platform, thereby realizing the joint execution of the job task through different connected learning platforms when the job task is complex, realizing the interconnection of different federated learning platforms, and also compensating for the performance deficiency of the algorithm of the self federated learning platform, and enriching the algorithm function of the self federated learning platform.

[0082] Figure 3 It is a schematic diagram of another method for interconnection between federated learning platforms according to an embodiment of the present disclosure. The embodiment is a further refinement of the above technical solution, and the technical solution in the embodiment can be combined with each optional solution in one or more of the above embodiments. As shown in Figure 3 The method for interconnection between federated learning platforms includes the following.

[0083] S310, in response to an opening instruction of a cross-platform job task in a self-initiating node in a self federated learning platform, determining an opposite federated learning platform matched with the cross-platform job task.

[0084] S320, in each self candidate node, determining a self collaborative node matched with the opposite federated learning platform.

[0085] S330, obtaining a global dependency graph matched with the cross-platform job task sent by the self-initiating node, and translating the global dependency graph to obtain a target dependency graph when the global dependency graph satisfies a translation condition.

[0086] S340, sending the target dependency graph to the self collaborative node.

[0087] In the embodiment, after the interconnection and interworking service sends the translated target dependency graph to the self collaborative node, it can also perform any of the following operations.

[0088] S351, in response to a processing instruction of each self task in the target dependency graph by the self collaborative node, sequentially obtaining data matched with each self task, and sending each data to the self collaborative node respectively.

[0089] In one optional implementation manner of the embodiment, after the interconnection and interworking service sends the translated target dependency graph to the self collaborative node, a scheduler in the self collaborative node can analyze the received target dependency graph, analyze to obtain self tasks and opposite tasks corresponding to each task job node, and send each opposite task to a corresponding node device in the opposite federated learning platform for processing by the corresponding node device in the opposite federated learning platform.

[0090] Further, the home-side coordination node can process each home-side task in turn, and in the process of processing each home-side task, the home-side coordination node can obtain data matched with each home-side task from the interconnection service; after receiving the data obtaining instruction of the home-side coordination node, the interconnection service can obtain each data from the database in the home-side federated learning platform and feed back each data to the home-side coordination node; further, the home-side coordination node can process each task in turn according to the data sent by the interconnection service; it should be noted that the home-side coordination node can obtain data for only one home-side task from the interconnection service at a time, or can obtain data for all home-side tasks from the interconnection service at the same time, which is not limited in this embodiment.

[0091] S352, before the home-side coordination node executes each home-side task, data matched with each home-side task is obtained and each data is sent to the home-side coordination node.

[0092] In an optional implementation of this embodiment, after the interconnection service sends the translated target dependency graph to the home-side coordination node, the scheduler in the home-side coordination node can parse the received target dependency graph, obtain the home-side task and the opposite task corresponding to each task job node, and send each opposite task to the corresponding node device in the opposite federated learning platform for processing by the corresponding node device in the opposite federated learning platform.

[0093] Further, before the home-side coordination node processes each home-side task in turn, the interconnection service can actively obtain data matched with each home-side data, and obtain each data from the database of the home-side federated learning platform in turn, and send each obtained data to the home-side coordination node, so that the home-side coordination node processes each home-side task according to each data.

[0094] It should be noted that in this embodiment, the home-side coordination node can actively obtain data required for processing each task or passively obtain data required for processing each task in the process of processing each home-side task, which is not limited in this embodiment; wherein S351 is a data passive obtaining process, and S352 is a data passive obtaining process.

[0095] S360, in response to the obtaining instruction of the home-side initiation node for the processing result of the cross-platform job task, downloading the processing result from the home-side coordination node and returning the processing result to the home-side initiation node.

[0096] In an optional implementation of the embodiment, after the home-side collaborative node completes processing of each home-side task, if the interconnection service receives a result acquisition instruction for the cross-platform job task from the home-side initiating node, the interconnection service can download the task processing result from the home-side collaborative node and send the processing result to the home-side initiating node, so that the home-side initiating node acquires the task result matched with the cross-platform job task.

[0097] The scheme of the embodiment can achieve on-demand data acquisition and improve the efficiency of data transmission in the following aspects. On the one hand, the home-side collaborative node acquires the data required for processing the task from the interconnection service, that is, acquires the data in a passive manner, which can achieve on-demand data acquisition and improve the efficiency of data transmission. On the other hand, the interconnection service can actively send the data required for processing each task to the home-side collaborative node before the home-side collaborative node processes each home-side task, that is, acquires the data in an active manner, which saves the step of acquiring data from the interconnection service by the home-side collaborative node, saves the processing time of the algorithm, and improves the execution efficiency of the algorithm.

[0098] Figure 4 FIG. 6 is a schematic diagram of another method for interconnection between federated learning platforms according to an embodiment of the present disclosure. The embodiment is a further refinement of the above technical solution, and the technical solution in the embodiment can be combined with one or more optional schemes in the above embodiments. As shown in FIG. 6, the method for interconnection between federated learning platforms includes the following steps. Figure 4

[0099] S410, in response to an opening instruction of a cross-platform job task in a home-side initiating node in a home-side federated learning platform, determining a counterpart federated learning platform matched with the cross-platform job task.

[0100] S420, in each home-side candidate node, determining a home-side collaborative node matched with the counterpart federated learning platform.

[0101] S430, acquiring a partial dependency graph matched with the cross-platform job task sent by the home-side initiating node and sending the partial dependency graph to the home-side collaborative node.

[0102] It can be understood that the global dependency graph matched with the cross-platform job task can have multiple branches or nodes, for example, 2, 5, or 10, etc. In the embodiment, the developer can determine to split one or more branches to obtain a partial dependency graph, or combine one or more nodes to obtain a partial dependency graph, and send the partial dependency graph to the interconnection service, so as to call the home-side collaborative node through the interconnection service and execute the cross-platform job task at the same time as the home-side initiating node, which realizes the use of algorithms from different software in one job for federated learning.

[0103] ​For example, the global dependency graph corresponding to the cross-platform job task contains two branches. The developer can decide to send one of the branches to the local collaborative node, so that the local collaborative node and the local initiating node can simultaneously complete the cross-platform job task with the other party's federated learning platform.

[0104] In this embodiment, after determining the local collaborative node that matches the other party's federated learning platform, the solution can also obtain a partial dependency graph sent by the local initiating node that matches the cross-platform task, and send the partial dependency graph to the local collaborative node. This can expand the node devices that process cross-platform tasks, thereby improving the efficiency of cross-platform tasks and expanding the federated learning platform software.

[0105] To enable those skilled in the art to better understand the interoperability methods between federated learning platforms involved in this disclosure, Figure 5a This refers to the translation scenarios applicable to the interoperability services provided according to the embodiments of this disclosure. Figure 5a In this diagram, 510 is the local initiating node, 520 is the local collaborating node, and 530 is the interoperability service. The local initiating node 510 contains the global dependency graph corresponding to the cross-platform task. In this embodiment, the interoperability service 520 can parse or translate the global dependency graph and send the parsed task nodes (not shown in the diagram) or the translated target dependency graph (not shown in the diagram) to the local collaborating node 520 for further processing. By using the local collaborating node 530, which is compatible with the other party's federated learning platform protocol, to collaborate with the other party's federated learning platform, the problem of the local initiating node and the other party's federated learning platform not being interoperable can be solved. It is understood that... Figure 5a The proxy involved is the plugin corresponding to the interconnection service in this embodiment, which enables the sending of the global dependency graph.

[0106] Figure 5b This refers to an extended scenario applicable to the interconnection services provided according to the embodiments of this disclosure. Figure 5b In this diagram, 510 is the local initiating node, 520 is the local collaborating node, and 530 is the interoperability service. The local initiating node 510 represents the global dependency graph corresponding to the cross-platform task. In this embodiment, a branch of the cross-platform task can be sent to the local collaborating node via the interoperability service 520. Then, both the local initiating node 510 and the local collaborating node process one branch of the cross-platform task. This binds and enhances the capabilities of the two federated learning software programs, allowing a single task to utilize the algorithms of both programs simultaneously.

[0107] Understandable,Figure 5b The agent involved in the present embodiment, that is, the plug-in corresponding to the interconnection service in the present embodiment, realizes sending of subgraph placement of the global dependency graph through the plug-in.

[0108] It should be noted that the interconnection service plug-in involved in the present embodiment is a client that realizes a standard protocol, realizes a federated learning operator interface and a wrapping form, and can be deployed in a federated learning operator environment. The home initiator node can send all or part of the subgraph corresponding to the cross-platform job task to the home collaborative node through the plug-in for translation or expansion of the function.

[0109] In the present embodiment, according to different DAG scheduling implementation manners, the present embodiment can be divided into a translation mode and a direct drive mode. The translation mode is to convert the DAG job dispatched by the home initiator node into a JOB representation manner of the opposite federated learning platform according to semantics, which can be a DSL language or a DAG graph, and then send the DAG graph to the scheduler of the home collaborative node. Figure 5c is a schematic diagram of the translation mode, which mainly includes the following: a home database 510 in the home federated learning platform, a home initiator node 520, an interconnection service 530, a home collaborative node 540, an opposite participating node 550 and an opposite database 560 in the opposite federated learning platform;

[0110] S510, sending a DAG graph matched with a cross-platform job task.

[0111] S520, translating the DAG graph into a target DAG graph matched with the home collaborative node.

[0112] S521, sending the target DAG graph to the scheduler of the home collaborative node.

[0113] S530, analyzing the target DAG graph, sending the home task to the home plug-in, and sending the opposite task to the plug-in in the opposite participating node in the opposite federated learning platform.

[0114] S531, obtaining data corresponding to the first home node task.

[0115] S522, obtaining data corresponding to the first home node task from the home database, and sending the data to the home plug-in of the home collaborative node; until all node tasks are processed.

[0116] S532, processing the first home node task according to the data.

[0117] S511, obtaining a processing result matched with the cross-platform job task.

[0118] S523, download the processing result from the self-party collaborative node, and send the processing result to the self-party initiating node.

[0119] In the embodiment, a direct driving mode is used, in which an internal scheduling service of a standard protocol service is directly used, a specific job node task is directly parsed, and sequential execution is performed. The task is directly sent to a federated learning software at the back end. Figure 5d is a schematic diagram of the direct driving mode, which mainly includes the following: a self-party database 510, a self-party initiating node 520, an interconnection service 530, and a self-party collaborative node 540 in a self-party federated learning platform; an interconnection service 550, a counterpart participating node 560, and a counterpart database 570 in a counterpart federated learning platform;

[0120] S510, send a DAG graph matched with a cross-platform job task.

[0121] S520, parse the DAG graph to obtain each self-party task and each counterpart task; send the self-party task to the interconnection service in the self-party collaborative node, and directly start the corresponding plug-in; and send the counterpart task to the interconnection service in the counterpart federated learning platform.

[0122] The interconnection service in the counterpart federated learning platform sends the counterpart task to the plug-in in the counterpart participating node for processing.

[0123] S530, obtain data corresponding to the first self-party node task.

[0124] S521, obtain data corresponding to the first self-party node task from the database, and send the data to the self-party plug-in in the self-party collaborative node; until all node tasks are processed.

[0125] S531, process the first self-party node task according to the data.

[0126] S511, obtain a processing result matched with the cross-platform job task.

[0127] S522, download the processing result from the self-party collaborative node, and send the processing result to the self-party initiating node.

[0128] It should be noted that, Figure 5c and Figure 5d only take the processing process of one node task as an example for description (the processing processes of other node tasks are not shown in the figure), and the processing processes of other node tasks are exactly the same as the processing process of the first node task, which is not repeated in the embodiment and is not a limitation on the embodiment.

[0129] In this embodiment, the interconnection service supports two modes, active mode and passive mode, for balancing compatibility and performance in data access mode. In the active mode, the interconnection service actively sends the required data to the backend node device through uploading and other means. The process may involve data format conversion, storage format conversion and other operations for subsequent calculation process. The active mode mainly uses the interface of the data set exposed by the node device for operation, so the compatibility is good. However, due to the need for data transmission and copying operations, it will cause large storage space and low efficiency. In the passive mode, the interconnection service passively waits for the algorithm to read the corresponding data during the operator calculation process. In the passive mode, the interconnection service is passive in waiting for data requests and does not actively push, so the operator in the node device is pulled as needed and discarded after use, so it does not cause space waste and has high reading efficiency according to the demand. However, due to the need for modification of the operator, the application range will be limited.

[0130] At present, the overall federated learning platform or engine is regarded as a whole through the top-level interconnection, that is, the black box interconnection, and the internal functions are exposed through the unified interface of the top layer. However, due to the differences in the implementation of software by different parties, the interface exposed by the upper layer does not have a unified interface, and different methods need to be used to interface different manufacturers, which is a very large workload. Moreover, the iteration structure of the software will also change, which will cause the downstream interconnection manufacturers to have to adjust accordingly.

[0131] The scheme of the present disclosure greatly reduces the traditional one-to-one mode to adapt to the software interfaces of different manufacturers. Through the interconnection service, all existing federated learning software that supports the interconnection service can be connected, thereby realizing interconnection.

[0132] Figure 6 is a structural schematic diagram of an interconnection device between federated learning platforms according to an embodiment of the present disclosure; the device can perform the interconnection method between federated learning platforms involved in any embodiment of the present disclosure; refer to Figure 6 The interconnection device 600 between federated learning platforms includes an opposite federated learning platform determination module 610, a self-side collaborative node determination module 620, a target dependency graph determination module 630 and a target dependency graph sending module 640.

[0133] The opposite federated learning platform determination module 610 is configured to determine an opposite federated learning platform matched with the cross-platform job task in response to an opening instruction of the cross-platform job task in the self-side initiating node in the self-side federated learning platform.

[0134] The self-side collaborative node determination module 620 is configured to determine a self-side collaborative node matched with the opposite federated learning platform in each self-side candidate node.

[0135] The target dependency graph determination module 630 is configured to acquire a global dependency graph sent by the local initiating node and match the cross-platform job task, and when the global dependency graph meets a translation condition, translate the global dependency graph to obtain a target dependency graph.

[0136] The target dependency graph sending module 640 is configured to send the target dependency graph to the local collaborative node, so that the local collaborative node jointly performs the cross-platform job task with the opposite federated learning platform based on the target dependency graph.

[0137] The scheme of the embodiment determines the opposite federated learning platform matched with the cross-platform job task in response to the opening instruction of the cross-platform job task in the local initiating node of the local federated learning platform through the opposite federated learning platform determination module, determines the local collaborative node matched with the opposite federated learning platform in each local candidate node through the local collaborative node determination module, acquires the global dependency graph sent by the local initiating node and matched with the cross-platform job task through the target dependency graph determination module, and when the global dependency graph meets the translation condition, translates the global dependency graph to obtain the target dependency graph, and sends the target dependency graph to the local collaborative node through the target dependency graph sending module, so that the local collaborative node jointly performs the cross-platform job task with the opposite federated learning platform based on the target dependency graph, solves the problem of complex intercommunication between federated learning platforms at the present stage, can realize the intercommunication between different federated learning platforms, and improves the processing efficiency of the job task.

[0138] In an optional implementation of the embodiment, the opposite federated learning platform determination module 610 is specifically configured to query task information of the cross-platform job task, and determine the opposite federated learning platform matched with the cross-platform job task according to the task information.

[0139] In an optional implementation of the embodiment, the local collaborative node determination module 620 is specifically configured to acquire version information of the opposite federated learning platform, and determine the local collaborative node matched with the version information, determine the local collaborative node in each local candidate node, and select the local collaborative node.

[0140] In an optional implementation of the embodiment, the target dependency graph determination module 630 further includes a target dependency graph translation determination sub-module configured to determine whether the global dependency graph meets the translation condition according to a complexity of the global dependency graph, and when the complexity meets a preset requirement, determine that the global dependency graph meets the translation condition.

[0141] In an optional implementation of the embodiment, the target dependency graph determination module 630 is specifically configured to determine a dependency graph form matched by the adapter of the local collaborative node; and translate the global dependency graph according to the dependency graph form to obtain the target dependency graph.

[0142] In an optional implementation of the embodiment, the target dependency graph determination module 640 further includes a global dependency graph analysis submodule configured to analyze the global dependency graph to obtain each local task and each opposite task when it is determined that the global dependency graph does not satisfy the translation condition; and send each local task to the local collaborative node and send each opposite task to an interconnection service in the opposite federated learning platform.

[0143] In an optional implementation of the embodiment, the interconnection device between federated learning platforms further includes a data transmission module configured to, in response to a processing instruction of the local collaborative node for each local task in the target dependency graph, sequentially obtain data matched with each local task and send each data to the local collaborative node; or, before the local collaborative node executes each local task, obtain data matched with each local task and send each data to the local collaborative node.

[0144] In an optional implementation of the embodiment, the interconnection device between federated learning platforms further includes a processing result transmission module configured to, in response to an acquisition instruction of the local initiating node for a processing result of the cross-platform job task, download the processing result from the local collaborative node and return the processing result to the local initiating node.

[0145] In an optional implementation of the embodiment, the interconnection device between federated learning platforms further includes a partial dependency graph acquisition module configured to obtain a partial dependency graph matched with the cross-platform job task sent by the local initiating node and send the partial dependency graph to the local collaborative node.

[0146] The interconnection device between federated learning platforms described above can execute the interconnection method between federated learning platforms provided by any embodiment of the present disclosure, and has the function modules and beneficial effects corresponding to the execution method. Technical details not described in detail in the embodiment can be referred to the generation method of the labeled data provided by any embodiment of the present disclosure.

[0147] In the technical solution of the present disclosure, the acquisition, storage and application of user personal information involved comply with relevant laws and regulations and do not violate public order and good customs.

[0148] According to the embodiments of the present disclosure, the present disclosure further provides an electronic device, a readable storage medium and a computer program product.

[0149] Figure 7 A schematic block diagram of an example electronic device 800 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0150] like Figure 7 As shown, device 700 includes a computing unit 701, which can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) 702 or a computer program loaded from storage unit 708 into random access memory (RAM) 703. RAM 703 may also store various programs and data required for the operation of device 700. The computing unit 701, ROM 702, and RAM 703 are interconnected via bus 704. Input / output (I / O) interface 705 is also connected to bus 704.

[0151] Multiple components in device 700 are connected to I / O interface 705, including: input unit 706, such as keyboard, mouse, etc.; output unit 707, such as various types of monitors, speakers, etc.; storage unit 708, such as disk, optical disk, etc.; and communication unit 709, such as network card, modem, wireless transceiver, etc. Communication unit 709 allows device 700 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0152] The computing unit 701 can be various general and / or special purpose processing components with processing and computing capabilities. Some examples of the computing unit 701 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various specialized artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 701 performs various methods and processes described above, such as the interworking method between federated learning platforms. For example, in some embodiments, the interworking method between federated learning platforms can be implemented as a computer software program tangibly embodied in a machine-readable medium, such as the storage unit 708. In some embodiments, part or all of the computer program can be loaded and / or installed onto the device 700 via the ROM 702 and / or the communication unit 709. When the computer program is loaded onto the RAM 703 and executed by the computing unit 701, one or more steps of the interworking method between federated learning platforms described above can be performed. Alternatively, in other embodiments, the computing unit 701 can be configured to perform the interworking method between federated learning platforms by any other suitable means, such as by means of firmware.

[0153] Various implementations of the systems and techniques described above can be realized in digital electronic circuitry, integrated circuitry, a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), a system on a chip (SOC), a complex programmable logic device (CPLD), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include implementation in one or more computer programs that are executable and / or interpretable on a programmable system including at least one programmable processor, which can be special or general purpose, coupled to receive data and instructions from, and to transmit data and instructions to, a storage system, at least one input device, and at least one output device.

[0154] Program code for carrying out methods of the present disclosure can be written in any combination of one or more programming languages. The program code can be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the program code, when executed by the processor or controller, produces a means for implementing the functions / acts specified in the flowcharts and / or block diagrams. The program code can be executed entirely on a machine, partially on a machine, partially on a machine and partially on a remote machine or entirely on a remote machine or server.

[0155] In the context of this disclosure, a machine-readable medium can be a tangible medium that contains or stores a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include but is not limited to an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of the machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0156] To provide for interaction with a user, the systems and techniques described here can be implemented on a computer having a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user and a keyboard and a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the computer. Other kinds of devices can be used to provide for interaction with a user as well; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form, including acoustic, speech, or tactile input.

[0157] The systems and techniques described here can be implemented in a computing system that includes a back end component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front end component (e.g., a user computer having a graphical user interface or a Web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such back end, middleware, or front end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network (LAN), a wide area network (WAN), and the Internet.

[0158] The computer system can include clients and servers. A client and server are generally remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other. The server can be a cloud server, a server of a distributed system, or a server combined with a blockchain.

[0159] It should be understood that the various forms of flow shown above can be used to reorder, add, or remove steps. For example, the steps described in the present disclosure can be performed in parallel, in series, or in a different order, as long as the desired results of the technology disclosed in the present disclosure are achieved, which is not limited herein.

[0160] The specific implementation described above does not constitute a limitation on the protection scope of the present disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent replacements, and improvements made within the spirit and principles of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. An interworking method between federal learning platforms, executed by an interconnection service built in a local federal learning platform, comprising: in response to an opening instruction of a cross-platform job task in a local initiating node in the local federal learning platform, determining a counterpart federal learning platform matched with the cross-platform job task; in each local candidate node, determining a local collaborative node matched with the counterpart federal learning platform; obtaining a global dependency graph sent by the local initiating node and matched with the cross-platform job task, and translating the global dependency graph into a target dependency graph when the global dependency graph meets a translation condition; sending the target dependency graph to the local collaborative node, so that the local collaborative node jointly executes the cross-platform job task with the counterpart federal learning platform based on the target dependency graph; when it is determined according to complexity of the global dependency graph that the global dependency graph does not meet the translation condition, parsing the global dependency graph to obtain each local task and each counterpart task; sending each local task to the local collaborative node, and sending each counterpart task to an interconnection service in the counterpart federal learning platform.

2. The method of claim 1, wherein, The determination of the counterpart federal learning platform matched with the cross-platform job task comprises: querying task information of the cross-platform job task, and determining the counterpart federal learning platform matched with the cross-platform job task according to the task information.

3. The method of claim 1, wherein, The determination of the local collaborative node matched with the counterpart federal learning platform in each local candidate node comprises: obtaining version information of the counterpart federal learning platform, and determining a local collaborative node matched with the version information; determining the local collaborative node in each local candidate node, and selecting the local collaborative node.

4. The method of claim 1, wherein, After obtaining the global dependency graph sent by the local initiating node and matched with the cross-platform job task, the method further comprises: determining whether the global dependency graph meets the translation condition according to complexity of the global dependency graph; when the complexity meets a preset requirement, determining that the global dependency graph meets the translation condition.

5. The method of claim 4, wherein, The translation of the global dependency graph into the target dependency graph when the global dependency graph meets the translation condition comprises: determining a dependency graph form matched with an adapter of the local collaborative node; translating the global dependency graph according to the dependency graph form to obtain the target dependency graph.

6. The method of claim 1, wherein, After sending the target dependency graph to the local collaborative node, the method further comprises: in response to a processing instruction of each local task in the target dependency graph by the local collaborative node, sequentially obtaining data matched with each local task, and sending each data to the local collaborative node respectively; or, before the local collaborative node executes each local task, obtaining data matched with each local task, and sending each data to the local collaborative node respectively.

7. The method of claim 1, wherein, After sending the target dependency graph to the local collaborative node, the method further comprises: In response to an acquisition instruction of the processing result of the cross-platform job task initiated by the local node, the processing result is downloaded from the local collaborative node and returned to the local initiating node.

8. The method of claim 1, wherein, After determining the local collaborative node matched with the opposite federal learning platform, further comprising: Acquiring the partial dependency graph sent by the local initiating node matched with the cross-platform job task, and sending the partial dependency graph to the local collaborative node.

9. An interworking device between federal learning platforms, executed by an interconnection and interworking service built in a local federal learning platform, comprising: An opposite federal learning platform determination module, configured to determine an opposite federal learning platform matched with a cross-platform job task in response to an opening instruction of the cross-platform job task in a local initiating node in a local federal learning platform; A local collaborative node determination module, configured to determine a local collaborative node matched with the opposite federal learning platform in each local candidate node; A target dependency graph determination module, configured to acquire a global dependency graph sent by the local initiating node matched with the cross-platform job task, and translate the global dependency graph to obtain a target dependency graph when the global dependency graph meets a translation condition; A target dependency graph sending module, configured to send the target dependency graph to the local collaborative node, so that the local collaborative node performs the cross-platform job task with the opposite federal learning platform based on the target dependency graph; The target dependency graph determination module further comprises a global dependency graph analysis submodule, configured to analyze the global dependency graph to obtain each local task and each opposite task when it is determined according to the complexity of the global dependency graph that the global dependency graph does not meet the translation condition; each local task is sent to the local collaborative node, and each opposite task is sent to an interconnection and interworking service in an opposite federal learning platform.

10. The apparatus of claim 9, wherein, The opposite federal learning platform determination module is specifically configured to Query task information of the cross-platform job task, and determine an opposite federal learning platform matched with the cross-platform job task according to the task information.

11. The apparatus of claim 9, wherein, The local collaborative node determination module is specifically configured to Acquire version information of the opposite federal learning platform, and determine a local collaborative node matched with the version information; Determine the local collaborative node in each local candidate node, and select the local collaborative node.

12. The apparatus of claim 9, wherein, The target dependency graph determination module further comprises a target dependency graph translation determination submodule, configured to Determine whether the global dependency graph meets a translation condition according to the complexity of the global dependency graph; When the complexity meets a preset requirement, it is determined that the global dependency graph meets the translation condition.

13. The apparatus of claim 12, wherein, The target dependency graph determination module is specifically configured to Determine a dependency graph form matched with an adapter of the local collaborative node; Translate the global dependency graph according to the dependency graph form to obtain the target dependency graph.

14. The apparatus of claim 9, wherein, The interworking device between federal learning platforms further comprises a data transmission module, configured to In response to the processing instruction of each local task in the target dependency graph, the local collaborative node sequentially acquires data matched with each local task and sends each data to the local collaborative node respectively. Alternatively, before the local collaborative node executes each local task, data matched with each local task is acquired and sent to the local collaborative node respectively.

15. The apparatus of claim 9, wherein, The intercommunication device between the federated learning platforms further comprises a processing result transmission module, configured to In response to the processing result acquisition instruction of the target node device for the cross-platform job task, the processing result is downloaded from the local collaborative node and returned to the local initiating node.

16. The apparatus of claim 9, wherein, The intercommunication device between the federated learning platforms further comprises a partial dependency graph acquisition module, configured to Acquire the partial dependency graph matched with the cross-platform job task sent by the local initiating node and send the partial dependency graph to the local collaborative node.

17. An electronic device, comprising: at least one processor; and a memory connected to the at least one processor in communication; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-8.

18. A non-transitory computer readable storage medium having stored thereon computer instructions, wherein, The computer instructions are used to enable the computer to perform the method according to any one of claims 1-8.

19. A computer program product comprising a computer program which, when executed by a processor, implements the method according to any one of claims 1-8.

Images