Interoperability methods, devices, architectures, equipment, and media for federated learning algorithms

Abstract

This application discloses an interconnection method, apparatus, architecture, device, and storage medium based on a federated learning algorithm. The method is applied to a cross-platform architecture comprising at least two nodes; each node is equipped with an algorithm component; and includes: acquiring task information for a first node to be federated; the first node being any one of the at least two nodes; invoking the algorithm component based on the task information to load a target algorithm onto a second node, obtaining configuration information of the target algorithm; the second node being any one of the at least two nodes other than the first node; determining resource information in the second node that matches the task information; and adaptively adapting the first node and the second node for interconnection using the configuration information and the resource information.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to an interconnection method, apparatus, architecture, device and storage medium based on a federated learning algorithm. Background Technology

[0002] Decoupling the underlying technical implementation of the federated learning framework requires both parties to confirm, coordinate, and unify data formats, task initiation requirements, and resource scheduling strategies step by step and layer by layer during collaborative tasks. This necessitates coordinating task initiation requests and coordinating unified resource scheduling strategies after task initiation. This approach is costly and burdensome because the high frequency of interaction and confirmation undoubtedly exacerbates communication and computational overhead. Furthermore, the resulting unified specifications are limited to the two parties involved in this collaboration and cannot be migrated or promoted, resulting in significant limitations.

[0003] By agreeing on a encrypted data transmission method, data can be exchanged between nodes, achieving interconnectivity between heterogeneous federated learning platforms. The federated learning process involves complex data transmission between layers such as the business application layer, management service layer, algorithm framework layer, and infrastructure resource layer, as well as between nodes within the same layer. A unified encrypted transmission method is difficult to implement and is limited to some publicly available and recognized federated learning algorithms. Some vendor-specific algorithms contain unique logic and encryption methods, making it impossible to publicly establish a unified encrypted method. Therefore, the technology in this application has significant limitations and a limited scope of application.

[0004] Interoperability between different federated learning systems is achieved by standardizing the computational primitives for collaborative tasks between nodes. Federated learning tasks involve a wide variety of computational primitives, with significant differences between different algorithms, including encryption primitives, Multimedia Personal Computer (MPC) primitives, homomorphic primitives, summarization primitives, and transport primitives. It is difficult to find common ground between different algorithms, therefore this method is only applicable to limited scenarios with a limited number of algorithms, and the collaboration cost is high, making it difficult to generalize.

[0005] By breaking down federated learning into subtasks to form a unified underlying cryptographic protocol, cross-platform and cross-architecture interoperability can be achieved. However, this method increases the workload by splitting tasks into subtasks, and it is also inefficient in seeking a unified protocol, as many tasks, even when broken down into very fine parts, cannot effectively form a unified underlying cryptographic protocol. Summary of the Invention

[0006] To address the related technical issues, embodiments of this application provide an interconnection method, apparatus, architecture, device, and storage medium based on a federated learning algorithm.

[0007] The technical solution of this application embodiment is implemented as follows:

[0008] This application provides an interoperability method based on a federated learning algorithm, applied to a cross-platform architecture including at least two nodes; each node is equipped with an algorithm component; the method includes:

[0009] Obtain the task information of the first node to be federated learning; the first node is any one of the at least two nodes.

[0010] Based on the task information, the algorithm component is invoked to load the target algorithm onto the second node, thereby obtaining the configuration information of the target algorithm; the second node is any node other than the first node among the at least two nodes.

[0011] In the second node, determine the resource information that matches the task information;

[0012] The configuration information and resource information are used to perform adaptive adaptation for interconnection between the first node and the second node.

[0013] In the above scheme, before obtaining the task information of the first node to be federated, the method further includes:

[0014] The system stores attribute information corresponding to each node; the attribute information includes the node's identity identifier (ID), name information, affiliated organization identifier, system version information, external service address information, authentication method information, and credential data information; the attribute information is used to determine the second node.

[0015] In the above scheme, before invoking the algorithm component based on the task information to load the target algorithm on the second node, the method further includes:

[0016] Based on the task information, the algorithm component is invoked to authenticate the node, resulting in the successfully authenticated second node.

[0017] In the above scheme, the step of calling the algorithm component based on the task information to authenticate the node and obtain the successfully authenticated second node includes:

[0018] Based on the task information, the algorithm component is invoked to authenticate the identity of the node, and a first authentication result is obtained;

[0019] If the first authentication result indicates that the node's identity authentication is successful, the first data of the node is formatted to obtain the second data.

[0020] The second data is audited to obtain the third data;

[0021] The third data is then subjected to security authentication to obtain the second node that has been successfully authenticated.

[0022] In the above scheme, the target algorithm includes at least one of the following:

[0023] Federated learning algorithms;

[0024] Interactive operator algorithm;

[0025] Non-interactive operator algorithm.

[0026] In the above scheme, the configuration information includes at least one of the following:

[0027] Algorithm information;

[0028] Data input and output information;

[0029] Algorithm orchestration information;

[0030] Task scheduling information.

[0031] In the above scheme, the resource information includes at least one of the following:

[0032] Data resources;

[0033] Computing resources;

[0034] Storage resources;

[0035] Online resources.

[0036] In the above scheme, the adaptive adaptation for interconnection and interoperability between the first node and the second node using the configuration information and the resource information includes:

[0037] The adaptation protocol between the first node and the second node is determined using the configuration information and the resource information.

[0038] The first node and the second node are interconnected based on the adaptation protocol.

[0039] In the above scheme, the adaptation protocol includes at least one of the following:

[0040] The data input and / or output specification protocol between the first node and the second node;

[0041] The algorithm orchestration specification protocol for the first node and the second node;

[0042] The task scheduling specification protocol between the first node and the second node.

[0043] This application also provides an interconnection device based on a federated learning algorithm, applied to a cross-platform architecture including at least two nodes; each node is equipped with an algorithm component; including:

[0044] An acquisition unit is used to acquire task information of the first node to be federated learning; the first node is any one of the at least two nodes.

[0045] The loading unit is used to call the algorithm component to load the target algorithm on the second node based on the task information, and obtain the configuration information of the target algorithm; the second node is any node other than the first node among the at least two nodes;

[0046] A determining unit is used to determine resource information that matches the task information in the second node;

[0047] An adaptation unit is used to adaptively adapt the first node and the second node for interconnection and interoperability using the configuration information and the resource information.

[0048] This application also provides an interconnection architecture based on a federated learning algorithm, including at least two nodes, a control layer, and a resource layer, wherein each node is equipped with an algorithm component; the algorithm component is connected to the control layer and the resource layer respectively; wherein...

[0049] The control layer is used to obtain task information of the first node to be federated learning; the first node is any one of the at least two nodes.

[0050] The algorithm component is used to load the target algorithm onto the second node based on the task information to obtain the configuration information of the target algorithm; the second node is any node other than the first node among the at least two nodes.

[0051] The resource layer is used to determine resource information that matches the task information in the second node;

[0052] The algorithm component is also used to adaptively adapt the first node and the second node for interconnection using the configuration information and the resource information.

[0053] This application also provides an electronic device, including:

[0054] Memory, used to store executable instructions;

[0055] A processor, when executing executable instructions stored in the memory, implements any step of the method described above.

[0056] This application also provides a computer-readable storage medium storing executable instructions for implementing any step of the method described above when executed by a processor.

[0057] This application provides an interoperability method, apparatus, architecture, device, and storage medium based on a federated learning algorithm. The method is applied to a cross-platform architecture including at least two nodes; each node is equipped with an algorithm component; the method includes: acquiring task information of a first node to be federated; the first node being any one of the at least two nodes; invoking the algorithm component based on the task information to load a target algorithm onto a second node, obtaining configuration information of the target algorithm; the second node being any one of the at least two nodes other than the first node; determining resource information matching the task information in the second node; and utilizing the configuration information and the... The resource information described in this application is used to adaptively adapt the first node and the second node for interconnection and interoperability. The solution of this application is to load the target algorithm on the second node by calling the algorithm component based on the task information to be federated learning on the first node, and obtain the configuration information of the target algorithm; determine the resource information matching the task information in the second node; and use the configuration information and resource information to adaptively adapt the first node and the second node for interconnection and interoperability. That is, the interconnection and interoperability between nodes is realized at the algorithm component level, which gets rid of the disadvantage of the heavy configuration of the engine and management service system; the adaptive adaptation of interconnection and interoperability realizes the adaptation of algorithm information, data input and output format, algorithm orchestration and task scheduling. Attached Figure Description

[0058] Figure 1 This is a schematic diagram illustrating the process of an interconnection method based on a federated learning algorithm, as provided in an embodiment of this application.

[0059] Figure 2 This is a schematic diagram of the interconnection architecture based on the federated learning algorithm in the embodiments of this application;

[0060] Figure 3 This is a schematic diagram of the logical structure of the self-adaptive algorithm component in the embodiments of this application;

[0061] Figure 4 This is a schematic diagram of the protocol adaptation layer logic structure in the embodiments of this application;

[0062] Figure 5 This is a schematic diagram of the adaptive matching and collaboration process in an embodiment of this application;

[0063] Figure 6 This is a schematic diagram of an interconnection device based on a federated learning algorithm, according to an embodiment of this application.

[0064] Figure 7This application provides a schematic diagram of an interconnection architecture based on a federated learning algorithm, which is an embodiment of the present application.

[0065] Figure 8 This is a schematic diagram of the hardware entity structure of an electronic device in an embodiment of this application. Detailed Implementation

[0066] The present application will now be described in further detail with reference to the accompanying drawings and embodiments.

[0067] Federated learning cross-platform and cross-framework interoperability typically refers to federated learning technology platforms based on different system architectures and algorithm principles, which, through agreed-upon interoperability protocols and interfaces, enable the interaction and collaboration of computing and data resources across platforms to support users on different platforms in collaboratively completing the same federated learning task.

[0068] The main concepts or methods for achieving interconnectivity currently include:

[0069] By decoupling the underlying technical implementation of the federated learning framework, and by standardizing the data format requirements, task initiation requirements, and resource scheduling strategies of both parties, cross-platform and cross-architecture interconnection is achieved.

[0070] By agreeing on a method for transmitting encrypted data, data can be exchanged between nodes, enabling interconnection and interoperability of heterogeneous federated learning platforms. This agreed-upon method includes the segmentation, packaging, sending, receiving, and execution of the original encrypted data.

[0071] Interoperability between different federated learning systems can be achieved by standardizing computational primitives for collaborative tasks between nodes. For example, a data transformation method based on SPDZ and ABY3 systems, using secret sharing factors and secret transformation factors, can realize a secure multi-party computation protocol.

[0072] By breaking down federated learning subtasks into a unified underlying cryptographic protocol, cross-platform and cross-architecture interoperability can be achieved.

[0073] Regarding the above concepts or methods for achieving interconnectivity,

[0074] Decoupling the underlying technical implementation of the federated learning framework requires both parties to confirm, coordinate, and unify data formats, task initiation requirements, and resource scheduling strategies step by step and layer by layer during collaborative tasks. This necessitates coordinating task initiation requests and coordinating unified resource scheduling strategies after task initiation. This approach is costly and burdensome because the high frequency of interaction and confirmation undoubtedly exacerbates communication and computational overhead. Furthermore, the resulting unified specifications are limited to the two parties involved in this collaboration and cannot be migrated or promoted, resulting in significant limitations.

[0075] By agreeing on a encrypted data transmission method, data can be exchanged between nodes, achieving interconnectivity between heterogeneous federated learning platforms. The federated learning process involves complex data transmission between layers such as the business application layer, management service layer, algorithm framework layer, and infrastructure resource layer, as well as between nodes within the same layer. A unified encrypted transmission method is difficult to implement and is limited to some publicly available and recognized federated learning algorithms. Some vendor-specific algorithms contain unique logic and encryption methods, making it impossible to publicly establish a unified encrypted method. Therefore, the technology in this application has significant limitations and a limited scope of application.

[0076] Interoperability between different federated learning systems is achieved by standardizing the computational primitives for collaborative tasks between nodes. Federated learning tasks involve a wide variety of computational primitives, with significant differences between different algorithms, including encryption primitives, MPC primitives, homomorphic primitives, digest primitives, and transport primitives. It is difficult to find common ground between different algorithms, so this method is only applicable to limited scenarios with a limited number of algorithms, and its high collaboration cost makes it difficult to promote.

[0077] By breaking down federated learning into subtasks to form a unified underlying cryptographic protocol, cross-platform and cross-architecture interoperability can be achieved. However, this method increases the workload by splitting tasks into subtasks, and it is also inefficient in seeking a unified protocol, as many tasks, even when broken down into very fine parts, cannot effectively form a unified underlying cryptographic protocol.

[0078] Therefore, this application proposes a federated learning algorithm component-based, containerizable, and adaptively matched interoperability method.

[0079] Based on this, this application provides an interconnection method based on a federated learning algorithm, applied to interconnection devices based on the federated learning algorithm. The functions implemented by this method can be achieved by the processor in the device calling program code. Of course, the program code can be stored in a computer storage medium. Therefore, the electronic device includes at least a processor and a storage medium. As an example, the device can be a mobile phone, computer, terminal, information transceiver, tablet device, personal digital assistant, etc.

[0080] Figure 1 This application provides a schematic diagram of an interconnection method based on a federated learning algorithm, as illustrated in the embodiments of this application. Figure 1 As shown, the method is applied to a cross-platform architecture comprising at least two nodes; each node is equipped with an algorithm component; the method includes:

[0081] Step 101: Obtain the task information of the first node to be federated; the first node is any one of the at least two nodes;

[0082] Step 102: Based on the task information, call the algorithm component to load the target algorithm on the second node to obtain the configuration information of the target algorithm; the second node is any node other than the first node among the at least two nodes;

[0083] Step 103: Determine the resource information that matches the task information in the second node;

[0084] Step 104: Adaptively connect the first node and the second node using the configuration information and the resource information.

[0085] It should be noted that the cross-platform architecture includes at least two nodes; the specific number of nodes in the at least two nodes can be determined according to the actual situation and is not limited here. As an example, the at least two nodes may include a first node and a second node; for example, the first node may be federated learning node A; and the second node may be federated learning node B. The cross-platform architecture can be determined according to the actual situation and is not limited here. As an example, the cross-platform architecture can be understood as different heterogeneous platforms being able to collaborate to complete the same federated learning task.

[0086] Each node is equipped with an algorithm component; the specific algorithm component can be determined based on actual conditions and is not limited here. As an example, the algorithm component is divided into two main structural layers according to its logical function: management and computation. The logical functions of the management layer include node authentication, data management, trusted evidence storage, and security authentication. Node Authentication: After the algorithm component is deployed on the platform, this module is responsible for authenticating node information. Following the unified node discovery protocol agreed upon by the protocol adaptation layer, it confirms the node's identity through an inter-node authentication mechanism. Data Management: Following the unified agreement on data input and output made by the protocol adaptation layer, it standardizes and formats input data and sends it to the computation module, while simultaneously standardizing the output of computation results. Trusted Evidence Storage: The node evidence storage and auditing function has the ability to log key data and key behaviors, meeting the needs of internal and external regulatory audits and the traceability of task responsibility. Security Authentication: Two-way identity authentication is used to ensure the trustworthiness of connected nodes. It is recommended to use cryptographic technologies such as Public Key Infrastructure (PKI), Certificate Authority (CA), and Direct Inward Dialing (DID). The computational layer structure includes federated learning algorithms, interactive operators, and non-interactive operators. The federated learning algorithm layer includes basic computation algorithms, joint modeling algorithms, feature engineering algorithms, model prediction algorithms, joint statistical algorithms, secure intersection algorithms, and model evaluation algorithms. Interactive operators are those that participate in inter-node interactions, including homomorphic encryption operators, secret-sharing operators, differential privacy operators, and unintentional transmission operators. Non-interactive operators are basic operators that do not participate in inter-node interactions, including cryptographic operators such as the Rivest-Shamir-Adleman cryptosystem (RSA) and the Secure Hash Algorithm 256 (SHA256).

[0087] The algorithm components can be independent algorithm modules, such as Public Service Identifier (PSI) and Link Register (LR). In practical applications, federated learning modeling tasks execute tasks and generate models by scheduling the algorithm resources corresponding to the underlying components. Each component is an independent algorithm module, such as PSI or LR. Components can be pre-configured according to algorithm orchestration and task scheduling specifications, including scheduling order and input-output relationships, to form a complete computational logic flow. Jobs and tasks schedule and run corresponding resources according to the algorithm orchestration and task scheduling specifications to generate corresponding results.

[0088] In step 101, the task information can be determined based on actual circumstances and is not limited here. The task information may include operational information such as job scheduling control, task scheduling control, algorithm registration and discovery, and component registration and management. The first node is any one of the at least two nodes; the first node can be determined based on actual circumstances and is not limited here. As an example, the first node can be federated learning node A.

[0089] In step 102, based on the task information, the algorithm component is invoked to load the target algorithm onto the second node, obtaining the configuration information of the target algorithm. The configuration information can be determined according to actual circumstances and is not limited here. As an example, the configuration information may include at least one of the following: algorithm information; data input / output information; algorithm orchestration information; task scheduling information. The second node is any node other than the first node among the at least two nodes; the second node can be determined according to actual circumstances and is not limited here. As an example, the second node may be federated learning node B. The target algorithm can be determined according to actual circumstances and is not limited here. As an example, the target algorithm may include at least one of the following: federated learning algorithm; interactive operator algorithm; non-interactive operator algorithm.

[0090] In practical applications, the target algorithm loading can be understood as target algorithm container loading; the target algorithm container loading can be simply referred to as algorithm container loading. This algorithm container loading uses standardized interfaces or methods to manage heterogeneous algorithms, and defines a standardized image loading mechanism and process through unified algorithm image construction standards and interfaces to achieve the security, efficiency and high availability of the algorithm container loading process.

[0091] In step 103, resource information matching the task information is determined in the second node; wherein, the resource information can be determined according to the actual situation and is not limited here. As an example, the resource information may include at least one of the following: data resources; computing power resources; storage resources; network resources.

[0092] In step 104, the specific adaptive adaptation for interconnecting the first node and the second node using the configuration information and resource information can be determined according to actual circumstances and is not limited here. As an example, the adaptive adaptation for interconnecting the first node and the second node using the configuration information and resource information may include determining an adaptation protocol for the first node and the second node using the configuration information and resource information; and interconnecting the first node and the second node based on the adaptation protocol. The adaptation protocol can be determined according to actual circumstances and is not limited here. As an example, the adaptation protocol may include at least one of the following: a data input and / or output specification protocol for the first node and the second node; an algorithm orchestration specification protocol for the first node and the second node; and a task scheduling specification protocol for the first node and the second node. The adaptation protocol may originate from a protocol adaptation layer, which coordinates and agrees on the protocols, data formats, algorithm orchestration specifications, task scheduling specifications, etc., used for node interaction.

[0093] The solution in this application embodiment loads the target algorithm onto the second node by calling the algorithm component based on the task information to be federated learning of the first node, thereby obtaining the configuration information of the target algorithm; resource information matching the task information is determined in the second node; and the configuration information and resource information are used to perform adaptive adaptation for interconnection between the first node and the second node, that is, to achieve interconnection between nodes at the algorithm component level, thus getting rid of the drawback of excessive configuration of the engine and management service system; the adaptive adaptation for interconnection achieves adaptation including algorithm information, data input and output formats, algorithm orchestration and task scheduling.

[0094] In one embodiment, before obtaining the task information of the first node to be federated, the method further includes:

[0095] The system stores attribute information corresponding to each node; the attribute information includes the node's identity ID, name information, affiliated organization ID, system version information, external service address information, authentication method information, and credential data information; the attribute information is used to determine the second node.

[0096] In this embodiment, the attribute information includes the node's ID, name, affiliated organization identifier, system version information, external service address information, authentication method information, and credential data information. The node's ID, name, affiliated organization identifier, system version information, external service address information, authentication method information, and credential data information can all be determined according to actual circumstances and are not limited here. As an example, the node's ID can be a node id; the name information can be the node name; the system version information can be the node's system version; the external service address information can be the external service address; the authentication method information can be the node's authentication method; and the credential data information can be credential data.

[0097] In practical applications, node creation and registration involve deploying the corresponding federated learning platform. Node creation and registration require storing and managing corresponding node information, including node ID, node name, affiliated institution identifier, node system version, external service address, node authentication method, and credential data.

[0098] In one embodiment, before invoking the algorithm component to load the target algorithm on the second node based on the task information, the method further includes:

[0099] Based on the task information, the algorithm component is invoked to authenticate the node, resulting in the successfully authenticated second node.

[0100] In this embodiment, the algorithm component is invoked based on the task information to authenticate the node. The specific authentication process of the successfully authenticated second node can be determined according to the actual situation and is not limited here. As an example, the invocation of the algorithm component based on the task information to authenticate the node and obtain the successfully authenticated second node may include authenticating the node's identity based on the task information to obtain a first authentication result; if the first authentication result indicates that the node's identity authentication is successful, formatting the node's first data to obtain second data; auditing the second data to obtain third data; and performing security authentication on the third data to obtain the successfully authenticated second node. The second node can be determined according to the actual situation and is not limited here. As an example, the second node can be a federated learning node B.

[0101] In practical applications, after the algorithm component completes platform deployment, this module is responsible for authenticating node information. According to the unified node discovery protocol agreed upon by the protocol adaptation layer, the node identity is confirmed through the authentication mechanism between nodes.

[0102] In one embodiment, the step of invoking the algorithm component to authenticate the node based on the task information to obtain the successfully authenticated second node includes:

[0103] Based on the task information, the algorithm component is invoked to authenticate the identity of the node, and a first authentication result is obtained;

[0104] If the first authentication result indicates that the node's identity authentication is successful, the first data of the node is formatted to obtain the second data.

[0105] The second data is audited to obtain the third data;

[0106] The third data is then subjected to security authentication to obtain the second node that has been successfully authenticated.

[0107] In this embodiment, the algorithm component is invoked based on the task information to authenticate the identity of the node, and the first authentication result can be either successful or unsuccessful.

[0108] If the first authentication result indicates that the node's identity authentication is successful, the node's first data is formatted to obtain second data. The formatting process can be determined based on actual circumstances and is not limited here. As an example, the formatting process can be a standard formatting process, which can be understood as a unified convention for data input and output. Standard formatted input data is sent to the calculation module, and the calculation result is output in a standardized manner.

[0109] The second data is audited to obtain the third data; the audit process can be determined according to the actual situation and is not limited here. As an example, the audit process can be understood as trusted evidence storage, node evidence storage and auditing function, which has the ability to log key data and key behaviors, and can meet the needs of internal and external supervision and auditing, as well as the traceability of task responsibility.

[0110] The third data is then subjected to security authentication to obtain the successfully authenticated second node. The security authentication process can be determined based on actual circumstances and is not limited here. As an example, the security authentication can be understood as employing two-way authentication to ensure the trustworthiness of the connected node's identity. For instance, this two-way authentication can be implemented using cryptographic technologies such as PKI / CA or DID.

[0111] In one embodiment, the target algorithm includes at least one of the following:

[0112] Federated learning algorithms;

[0113] Interactive operator algorithm;

[0114] Non-interactive operator algorithm.

[0115] The federated learning algorithm can be determined according to the actual situation and is not limited here. As an example, the federated learning algorithm may include basic operation algorithms, joint modeling algorithms, feature engineering algorithms, model prediction algorithms, joint statistical algorithms, secure intersection algorithms, model evaluation algorithms, etc.

[0116] The interactive operator algorithm can be determined according to the actual situation and is not limited here. As an example, the interactive operator algorithm can be understood as an operator that participates in the interaction between nodes, including homomorphic encryption operators, secret sharing operators, differential privacy operators, and unintentional transmission operators.

[0117] The non-interactive operator algorithm can be determined according to the actual situation and is not limited here. As an example, the non-interactive operator algorithm can be understood as a non-interactive operator, a basic operator that does not participate in the interaction between nodes, including encryption operators such as RSA and SHA256.

[0118] In practical applications, the computational layer structure includes federated learning algorithms, interactive operators, and non-interactive operators. The federated learning algorithm layer includes basic computation algorithms, joint modeling algorithms, feature engineering algorithms, model prediction algorithms, joint statistical algorithms, secure intersection algorithms, and model evaluation algorithms. Interactive operators are those that participate in inter-node interactions, including homomorphic encryption operators, secret-sharing operators, differential privacy operators, and unintentional transmission operators. Non-interactive operators are basic operators that do not participate in inter-node interactions, including encryption operators such as RSA and SHA256.

[0119] In one embodiment, the configuration information includes at least one of the following:

[0120] Algorithm information;

[0121] Data input and output information;

[0122] Algorithm orchestration information;

[0123] Task scheduling information.

[0124] The algorithm information can be determined based on the actual situation and is not limited here. As an example, the algorithm information may include PSI, LR, etc.

[0125] The data input and output information can be determined according to the actual situation and is not limited here. As an example, the data input and output information may include data input and output specifications. With this specification, different platform architectures no longer need to define the design of the algorithm itself. They only need to agree on the basic information of the algorithm and the input and output protocol specifications before the two or more parties collaborate. In subsequent processes, the algorithm components will use this version of the agreement to adapt the data input and output, thus enabling the implementation of algorithm components that are compatible with various platform architectures.

[0126] The algorithm orchestration information can be determined based on actual circumstances and is not limited here. As an example, the algorithm orchestration information may include algorithm orchestration specifications. Since the algorithm orchestration specifications of federated learning platforms with different architectures vary greatly, the algorithm orchestration specifications of this application can effectively adapt to the algorithm orchestration of various platforms, ultimately evolving into the orchestration content that the algorithm components of this application can execute.

[0127] The task scheduling information can be determined based on actual circumstances and is not limited here. As an example, the task scheduling information may include task scheduling specifications. For tasks implemented by a single component, the task scheduling mechanism may differ on different nodes. Therefore, task scheduling specifications are implemented in the adaptation layer, thus avoiding execution conflicts caused by differences in task scheduling when deployed on nodes with different architectures.

[0128] When the algorithm components of this application are deployed on any architecture node platform in a containerized manner, they can adaptively adapt to the corresponding data input and output, algorithm orchestration and task scheduling according to the execution requirements of the other party, thereby ensuring the executability of any platform and the interoperability with the algorithm components of any node. It has a wide range of adaptability and strong collaboration capabilities.

[0129] In one embodiment, the resource information includes at least one of the following:

[0130] Data resources;

[0131] Computing resources;

[0132] Storage resources;

[0133] Online resources.

[0134] The data resources mentioned herein can be determined based on actual circumstances and are not limited here. As an example, the data resources may include resources such as datasets required for federated learning modeling assignments.

[0135] The computing resources mentioned can be determined based on actual circumstances and are not limited here. As an example, the computing resources may include the computing resources required for federated learning modeling tasks, etc.

[0136] The storage resources can be determined based on actual circumstances and are not limited here. As an example, the storage resources may include data storage resources required for federated learning modeling tasks, etc.

[0137] The network resources mentioned can be determined based on actual circumstances and are not limited here. As an example, the network resources may include network resources required for federated learning modeling assignments, etc.

[0138] In one embodiment, the adaptive adaptation for interconnection and interoperability between the first node and the second node using the configuration information and the resource information includes:

[0139] The adaptation protocol between the first node and the second node is determined using the configuration information and the resource information.

[0140] The first node and the second node are interconnected based on the adaptation protocol.

[0141] The adaptation protocol can be determined based on actual circumstances and is not limited here. As an example, the adaptation protocol may include at least one of the following: a data input and / or output specification protocol between the first node and the second node; an algorithm orchestration specification protocol between the first node and the second node; and a task scheduling specification protocol between the first node and the second node. The adaptation protocol may originate from a protocol adaptation layer, which is used to coordinate and agree on the protocols, data formats, algorithm orchestration specifications, task scheduling specifications, etc., used for node interaction.

[0142] In practical applications, firstly, the algorithm component is deployed on the target platform using open-source application container engines such as Docker and Kubernetes. The node control layer is invoked to load the algorithm container, and algorithm registration is completed via the registry center. Users configure the protocol adaptation layer according to interoperability requirements, mainly including algorithm information, data input / output, algorithm orchestration, and task scheduling information. Through the node control layer, node resource information is obtained, mainly including data resources, computing power resources, storage resources, and network resources. The algorithm component, through the configuration of the protocol adaptation layer, completes adaptive interoperability adaptation, mainly including unifying algorithm information, converting data input / output formats, and translating and adapting algorithm orchestration and task scheduling. Data resources are obtained according to the standardized data input format, job resources are requested, and relevant jobs are executed according to the translated and adapted algorithm orchestration. Simultaneously, tasks and job scheduling from the control layer are responded to according to the translated and adapted job scheduling rules, thereby enabling interaction and coordination with the target node to achieve interoperability and collaborative job requirements. Finally, data is output according to the standardized data output format, and the results are stored.

[0143] In one embodiment, the adaptation protocol includes at least one of the following:

[0144] The data input and / or output specification protocol between the first node and the second node;

[0145] The algorithm orchestration specification protocol for the first node and the second node;

[0146] The task scheduling specification protocol between the first node and the second node.

[0147] This embodiment primarily addresses the current issues in implementing interconnectivity in federated learning. This application pioneered the addition of a protocol matching layer on top of the algorithm components. The data input and / or output specification protocols, the algorithm orchestration specification protocols, and the task scheduling specification protocols are all located in the protocol matching layer. This protocol matching layer is used to coordinate the protocols, data formats, algorithm orchestration specifications, and task scheduling specifications used for the interaction of agreed-upon nodes.

[0148] In practical applications, the data input and / or output specification protocol can be referred to as the data input / output specification; the algorithm orchestration specification protocol can be referred to as the algorithm orchestration specification; and the task scheduling specification protocol can be referred to as the task scheduling specification. Data Input / Output Specification: With this specification layer, different platform architectures no longer need to define the algorithm design itself. Only the basic information of the algorithm and the input / output protocol specifications need to be agreed upon before collaboration between two or more parties. Subsequent process algorithm components will use this version of the agreement for data input and output adaptation, thus achieving compatibility with algorithm components developed for various platform architectures. Algorithm Orchestration Specification: Since the algorithm orchestration specifications of federated learning platforms using different architectures vary greatly, the algorithm orchestration specification of this application can effectively adapt to the algorithm orchestration of various platforms, ultimately evolving into the orchestration content that the algorithm components of this application can execute. Task Scheduling Specification: For tasks implemented by a single component, the task scheduling mechanism also differs on different nodes. Therefore, a task scheduling specification is implemented in the adaptation layer, thus avoiding execution conflicts caused by differences in task scheduling when deployed on nodes of different architectures.

[0149] To better understand, the interconnection method based on federated learning algorithms is specifically an interconnection method with interconnectable algorithmic components that have adaptive matching.

[0150] 1. Interoperability solution.

[0151] This application proposes a standardized job scheduling and algorithm container loading process to enable job scheduling control, task scheduling control, algorithm registration and discovery, component registration and management, and other operations during the federated learning modeling job operation process. This achieves secure and controllable interoperable job operation, enabling different heterogeneous platforms to collaborate in completing the same federated learning task.

[0152] 1) Overall architecture design.

[0153] The entire architecture can be broken down into a control layer, a computation layer, and a resource layer. The control layer primarily handles node creation and registration, job and task scheduling and management, algorithm registration, algorithm container loading, job resource application, and node resource directory acquisition. The computation layer handles the installation and adaptation of algorithm components, while the collaborative control layer handles algorithm registration and algorithm container loading. The computation layer is the core layer for unfolding federated learning modeling jobs and is also the key layer for achieving interconnectivity and self-adaptation. On one hand, this layer is responsible for the specific modeling and related model applications in federated learning; on the other hand, it ensures interconnectivity at the platform level using the same protocol specifications. The resource layer is an abstraction of specific physical resources, mainly including data resources, computing power resources, storage resources, and network resources. This content can be combined with... Figure 2 To understand, Figure 2 This is a schematic diagram of the interconnection architecture based on the federated learning algorithm in the embodiments of this application.

[0154] Control layer:

[0155] Node creation and registration: Creating a node means deploying the corresponding federated learning platform. Node creation and registration require storing and managing the corresponding node information, including node ID, node name, affiliated institution identifier, node system version, external service address, node authentication method, credential data, etc.

[0156] The scheduling and management of jobs and tasks includes managing and scheduling related federated learning jobs and tasks, storing and managing corresponding job and task information, including job ID, task ID, process ID, job status, task status, related time information, and parameter configuration information.

[0157] Algorithm registration is the core of federated learning. It enables the unified description of collaborative task algorithms and standardizes the algorithm process.

[0158] Algorithm container loading uses standardized interfaces or methods to manage heterogeneous algorithms. It defines standardized image loading mechanisms and processes through unified algorithm image construction standards and interfaces, and achieves secure, efficient and highly available algorithm container loading processes.

[0159] Application for assignment resources enables the instantiation and execution of the federated learning modeling process.

[0160] Obtain the node resource directory, which contains information about collaborating nodes, including node ID, node name, affiliated organization identifier, node system version, external service address, node authentication method, and credential data.

[0161] Computation layer:

[0162] The protocol adaptation layer is used to coordinate the protocols, data formats, algorithm orchestration specifications, and task scheduling specifications used for the interaction of agreed nodes.

[0163] At the algorithm component layer, federated learning modeling tasks execute tasks and generate models by scheduling the algorithm resources corresponding to the underlying components. Each component is an independent algorithm module, such as PSI and LR. Components can be pre-configured according to algorithm orchestration and task scheduling specifications, including scheduling order and input-output relationships, thus orchestrating a process with complete computational logic. Jobs and tasks schedule and run their corresponding resources according to the algorithm orchestration and task scheduling specifications to generate corresponding results.

[0164] Resource layer:

[0165] Data resources, such as datasets required for federated learning modeling assignments.

[0166] Computing resources, computing resources required for federated learning modeling tasks, etc.

[0167] Storage resources, including data storage resources required for federated learning modeling tasks, etc.

[0168] Network resources, including network resources required for federated learning modeling assignments, etc.

[0169] 2) Algorithm component-based.

[0170] An algorithm component is a series of usable modular units that encapsulate the implementation of the algorithm's functions for federated learning algorithms.

[0171] This application constructs a pluggable platform-supported algorithm component, which is deployed via containerization and is plug-and-play. This component achieves interoperability at the platform level using the same set of protocol specifications through a protocol adaptation layer. This content can be combined with... Figure 3 To understand, Figure 3 This is a schematic diagram of the logical structure of the self-adaptive algorithm component in the embodiments of this application.

[0172] The algorithm components are divided into two main structural layers based on their logical functions: management and computation. The logical functions of the management layer include node authentication, data management, trusted storage, and security authentication.

[0173] Node authentication: After the algorithm components are deployed on the platform, this module is responsible for authenticating node information. It confirms the identity of nodes through the authentication mechanism between nodes, according to the unified node discovery protocol agreed upon by the protocol adaptation layer.

[0174] Data Management: In accordance with the unified conventions made by the protocol adaptation layer for data input and output, the input data is formatted in a standard format and sent to the calculation module, while the calculation results are output in a standardized manner.

[0175] Trusted Evidence Storage: The node evidence storage and auditing function has the ability to log key data and key behaviors, which can meet the needs of internal and external supervision and auditing, as well as the traceability of task responsibility.

[0176] Security Authentication: Two-way authentication is used to ensure the trustworthiness of the connected node. It is recommended to use cryptographic technologies such as PKI / CA and DID.

[0177] The computational layer architecture includes federated learning algorithms, interactive operators, and non-interactive operators.

[0178] Federated learning algorithms include basic computational algorithms, joint modeling algorithms, feature engineering algorithms, model prediction algorithms, joint statistical algorithms, secure intersection algorithms, and model evaluation algorithms.

[0179] Interactive operators are operators that participate in interactions between nodes, including homomorphic encryption operators, secret sharing operators, differential privacy operators, and unintentional transmission operators.

[0180] Non-interactive operators are basic operators that do not participate in inter-node interactions, including cryptographic operators such as RSA and SHA256.

[0181] 3) Containerizable deployment.

[0182] Container deployment facilitates the modularization and service-orientation of algorithm components. Service-orientation allows for natural integration with the cloud and supports open-source application container engines such as Docker and Kubernetes.

[0183] 4) Adaptive matching.

[0184] To address the current challenges in achieving interoperability in federated learning, this application pioneered the addition of a protocol matching layer on top of the algorithm components. This layer coordinates the protocols, data formats, algorithm orchestration specifications, and task scheduling specifications used for node interactions. This content can be combined with... Figure 4 To understand, Figure 4 This is a schematic diagram of the protocol adaptation layer logic structure in the embodiments of this application.

[0185] Data input / output specifications: With this specification, different architecture platforms no longer need to define the design of the algorithm itself. They only need to agree on the basic information of the algorithm and the input / output protocol specifications before two or more parties collaborate. In the subsequent process, the algorithm components will use this version of the agreement to adapt the data input and output. Therefore, it has the feasibility of being compatible with algorithm components developed on various architecture platforms.

[0186] Algorithm orchestration specifications: Since the algorithm orchestration specifications of federated learning platforms with different architectures vary greatly, the algorithm orchestration specifications of this application can effectively adapt to the algorithm orchestration of various platforms, and ultimately evolve into the orchestration content that the algorithm components of this application can execute.

[0187] Algorithm orchestration specification JSON;

[0188]

[0189]

[0190] Task scheduling specifications: For tasks implemented by a single component, the task scheduling mechanism may differ on different nodes. Therefore, task scheduling specifications were implemented in the adaptation layer. This avoids execution conflicts caused by differences in task scheduling when deployed on nodes with different architectures.

[0191] Task scheduling specification JSON,

[0192]

[0193]

[0194]

[0195] In summary, when the algorithm components of this application are deployed on any architecture node platform in a containerized manner, they can adaptively adapt to the corresponding data input and output, algorithm orchestration and task scheduling according to the execution requirements of the other party, thereby ensuring the executability of any platform and the interoperability with the algorithm components of any node. It has a wide range of adaptability and strong collaboration capabilities.

[0196] 2. Interoperability process.

[0197] First, the algorithm component is deployed on the target platform using open-source application container engines such as Docker and K8s. The node control layer is called to load the algorithm container and complete the algorithm registration through the registry center.

[0198] Users configure the protocol adaptation layer according to the needs of interconnection and interoperability, which mainly includes algorithm information, data input and output, algorithm orchestration and task scheduling information;

[0199] The node control layer obtains node resource information, which mainly includes data resources, computing power resources, storage resources, and network resources.

[0200] Algorithm components achieve adaptive adaptation for interconnection through the configuration content of the protocol adaptation layer. This mainly includes unifying algorithm information, converting data input and output formats, translating and adapting algorithm orchestration and task scheduling.

[0201] Data resources are acquired according to a standardized data input format, job resources are requested, and relevant jobs are executed according to the translated and adapted algorithm. Simultaneously, tasks and job scheduling from the control layer are responded to according to the translated and adapted job scheduling rules, thereby enabling interaction and coordination with target nodes to achieve interoperable job collaboration requirements. Finally, data is output according to a standardized data output format, and the results are stored. This content can be combined with... Figure 5 To understand, Figure 5 This is a schematic diagram of the adaptive matching and collaboration process in an embodiment of this application.

[0202] Compared with existing technologies, the advantages of this application are mainly reflected in the following aspects:

[0203] First: We were the first to add a protocol matching layer on top of the algorithm components, which enabled the adaptation of algorithm information, data input and output formats, algorithm orchestration and task scheduling.

[0204] Second: It achieves interconnection and interoperability between nodes at the algorithm component level, thus getting rid of the drawbacks of excessive configuration of the engine and management service system.

[0205] Third: Containerized deployment enables the modularization and service-orientation of algorithm components, which is beneficial for cloud deployment.

[0206] To implement the method of the embodiments of this application, the embodiments of this application also provide an interconnection device 600 based on a federated learning algorithm, applied to a cross-platform architecture including at least two nodes; each node is provided with an algorithm component; Figure 6 This is a schematic diagram of an interconnection device based on a federated learning algorithm, as described in an embodiment of this application; Figure 6 As shown, it includes:

[0207] The acquisition unit 601 is used to acquire task information of the first node to be federated learning; the first node is any one of the at least two nodes.

[0208] The loading unit 602 is used to call the algorithm component to load the target algorithm on the second node based on the task information, and obtain the configuration information of the target algorithm; the second node is any node other than the first node among the at least two nodes;

[0209] The determining unit 603 is used to determine resource information that matches the task information in the second node;

[0210] The adaptation unit 604 is used to perform adaptive adaptation for interconnection between the first node and the second node using the configuration information and the resource information.

[0211] Here, in one embodiment, before obtaining the task information to be federated learning for the first node, the device 600 further includes a storage unit for storing attribute information corresponding to each node; the attribute information includes the node's identity ID, name information, affiliated organization ID, system version information, external service address information, authentication method information, and credential data information; the attribute information is used to determine the second node.

[0212] Here, in one embodiment, before the algorithm component is invoked based on the task information to load the target algorithm on the second node, the device 600 further includes an authentication unit, which is used to invoke the algorithm component based on the task information to authenticate the node and obtain the successfully authenticated second node.

[0213] Here, in one embodiment, the authentication unit is further configured to call the algorithm component based on the task information to authenticate the identity of the node and obtain a first authentication result; if the first authentication result indicates that the identity authentication of the node is successful, format the first data of the node to obtain second data; perform auditing processing on the second data to obtain third data; and perform security authentication on the third data to obtain the second node that has been successfully authenticated.

[0214] Here, in one embodiment, the target algorithm includes at least one of the following:

[0215] Federated learning algorithms;

[0216] Interactive operator algorithm;

[0217] Non-interactive operator algorithm.

[0218] In one embodiment, the configuration information includes at least one of the following:

[0219] Algorithm information;

[0220] Data input and output information;

[0221] Algorithm orchestration information;

[0222] Task scheduling information.

[0223] Here, in one embodiment, the resource information includes at least one of the following:

[0224] Data resources;

[0225] Computing resources;

[0226] Storage resources;

[0227] Online resources.

[0228] In one embodiment, the adaptation unit is further configured to determine the adaptation protocol of the first node and the second node using the configuration information and the resource information; and to interconnect the first node and the second node based on the adaptation protocol.

[0229] Here, in one embodiment, the adaptation protocol includes at least one of the following:

[0230] The data input and / or output specification protocol between the first node and the second node;

[0231] The algorithm orchestration specification protocol for the first node and the second node;

[0232] The task scheduling specification protocol between the first node and the second node.

[0233] It should be noted that the interconnection device based on the federated learning algorithm provided in the above embodiments is only illustrated by the division of the above program modules when performing interconnection based on the federated learning algorithm. In practical applications, the above processing can be assigned to different program modules as needed, that is, the internal structure of the device can be divided into different program modules to complete all or part of the processing described above. In addition, the interconnection device based on the federated learning algorithm provided in the above embodiments and the interconnection method embodiments based on the federated learning algorithm belong to the same concept, and the specific implementation process is detailed in the method embodiments, which will not be repeated here.

[0234] This application provides an interconnection architecture based on a federated learning algorithm, such as... Figure 7 As shown, Figure 7 This application provides a schematic diagram of an interconnection architecture based on a federated learning algorithm. The interconnection architecture includes at least two nodes, a control layer, and a resource layer. Each node is equipped with an algorithm component. The algorithm component is connected to both the control layer and the resource layer.

[0235] The control layer is used to obtain task information of the first node to be federated learning; the first node is any one of the at least two nodes.

[0236] The algorithm component is used to load the target algorithm onto the second node based on the task information to obtain the configuration information of the target algorithm; the second node is any node other than the first node among the at least two nodes.

[0237] The resource layer is used to determine resource information that matches the task information in the second node;

[0238] The algorithm component is also used to adaptively adapt the first node and the second node for interconnection using the configuration information and the resource information.

[0239] In this embodiment, the interconnection architecture includes at least two nodes; the specific number of nodes can be determined according to actual conditions and is not limited here. Figure 7 The example illustrates two nodes, designated as the first node and the second node. Both the first and second nodes can be determined based on actual circumstances and are not limited here. As an example, the first node can be federated learning node A, and the second node can be federated learning node B.

[0240] The description of the above interconnection architecture embodiments is similar to the description of the above method embodiments, and has similar beneficial effects. For technical details not disclosed in the interconnection architecture embodiments of this application, please refer to the description of the method embodiments of this application for understanding.

[0241] Based on the hardware implementation of the above program modules, this application embodiment also provides an interconnection device based on a federated learning algorithm, including a memory and a processor. The memory stores a computer program that can run on the processor. When the processor executes the program, it implements the steps in the interconnection method based on a federated learning algorithm provided in the above embodiment.

[0242] Correspondingly, embodiments of this application provide a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps in the interconnection method based on the federated learning algorithm provided in the above embodiments.

[0243] It should be noted that the descriptions of the storage medium and device embodiments above are similar to the descriptions of the method embodiments above, and have similar beneficial effects. For technical details not disclosed in the storage medium and device embodiments of this application, please refer to the descriptions of the method embodiments of this application for understanding.

[0244] It should be noted that, Figure 8 This is a schematic diagram of a hardware entity structure of an electronic device in an embodiment of this application, such as... Figure 8 As shown, the hardware entity of the electronic device 800 includes a processor 801 and a memory 803. Optionally, the electronic device 800 may also include a communication interface 802.

[0245] It is understood that memory 803 can be volatile memory or non-volatile memory, or both. Non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), ferromagnetic random access memory (FRAM), flash memory, magnetic surface memory, optical disc, or compact disc read-only memory (CD-ROM); magnetic surface memory can be disk storage or magnetic tape storage. Volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as Static Random Access Memory (SRAM), Synchronous Static Random Access Memory (SSRAM), Dynamic Random Access Memory (DRAM), Synchronous Dynamic Random Access Memory (SDRAM), Double Data Rate Synchronous Dynamic Random Access Memory (DDRSDRAM), Enhanced Synchronous Dynamic Random Access Memory (ESDRAM), SyncLink Dynamic Random Access Memory (SLDRAM), and Direct Rambus Random Access Memory (DRRAM).The memory 803 described in the embodiments of this application is intended to include, but is not limited to, these and any other suitable types of memory.

[0246] The methods disclosed in the embodiments of this application can be applied to or implemented by processor 801. Processor 801 may be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method can be completed by the integrated logic circuit of the hardware in processor 801 or by instructions in software form. The processor 801 may be a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Processor 801 can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. A general-purpose processor may be a microprocessor or any conventional processor, etc. The steps of the methods disclosed in the embodiments of this application can be directly manifested as being executed by a hardware decoding processor, or being executed by a combination of hardware and software modules in the decoding processor. The software modules may be located in a storage medium, which is located in memory 803. Processor 801 reads the information in memory 803 and combines it with its hardware to complete the steps of the aforementioned method.

[0247] In an exemplary embodiment, the device may be implemented by one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), general-purpose processors, controllers, microcontrollers (MCUs), microprocessors, or other electronic components to perform the aforementioned method.

[0248] It should be understood that the phrase "one embodiment" or "an embodiment" throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of this application. Therefore, "in one embodiment" or "in an embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. It should be understood that in the various embodiments of this application, the sequence numbers of the above-described processes do not imply a sequential order of execution; the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application. The sequence numbers of the above-described embodiments are merely descriptive and do not represent the superiority or inferiority of the embodiments.

[0249] It should be noted that, in this application, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0250] The methods disclosed in the several method embodiments provided in this application can be arbitrarily combined without conflict to obtain new method embodiments.

[0251] The features disclosed in the several product embodiments provided in this application can be arbitrarily combined without conflict to obtain new product embodiments.

[0252] The features disclosed in the several method or device embodiments provided in this application can be arbitrarily combined without conflict to obtain new method or device embodiments.

[0253] The above description is merely an embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. An interconnection and interworking method based on a federated learning algorithm, characterized in that, The application is applied to a cross-platform architecture comprising at least two nodes; Each of the nodes is provided with an algorithm component; the method comprises: Obtaining task information of a first node to be federated learning; the first node is any node in the at least two nodes; Based on the task information, calling the algorithm component to load a target algorithm on a second node to obtain configuration information of the target algorithm; the second node is any node in the at least two nodes except the first node; Determining resource information matched with the task information in the second node; Adaptively adapting the first node and the second node to interconnect and intercommunicate by using the configuration information and the resource information.

2. The method of claim 1, wherein, Before the obtaining task information of a first node to be federated learning, the method further comprises: Storing attribute information corresponding to each of the nodes; the attribute information comprises an identity ID, name information, an organization identifier, system version information, external service address information, authentication mode information and credential data information of the node; the attribute information is used to determine the second node.

3. The method of claim 1, wherein, Before the calling the algorithm component to load a target algorithm on a second node based on the task information, the method further comprises: Based on the task information, calling the algorithm component to authenticate the node to obtain the second node successfully authenticated.

4. The method of claim 3, wherein, The calling the algorithm component to authenticate the node based on the task information to obtain the second node successfully authenticated comprises: Based on the task information, calling the algorithm component to authenticate the identity of the node to obtain a first authentication result; In a case where the first authentication result represents that the identity authentication of the node is successful, formatting first data of the node to obtain second data; Auditing the second data to obtain third data; Security authenticating the third data to obtain the second node successfully authenticated.

5. The method of claim 1, wherein, The target algorithm comprises at least one of: A federated learning algorithm; An interactive operator algorithm; A non-interactive operator algorithm.

6. The method of claim 1, wherein, The configuration information comprises at least one of: Algorithm information; Data input and output information; Algorithm arrangement information; Task scheduling information.

7. The method of claim 1, wherein, The resource information comprises at least one of: Data resources; Computing power resources; Storage resources; Network resources.

8. The method of claim 1, wherein, The adaptively adapting the first node and the second node to interconnect and intercommunicate by using the configuration information and the resource information comprises: Determining an adaptation protocol of the first node and the second node by using the configuration information and the resource information; Based on the adaptation protocol, interconnecting and intercommunicating the first node and the second node.

9. The method of claim 1, wherein, The adaptation protocol comprises at least one of: A specification protocol of data input and output of the first node and the second node; An algorithm arrangement specification protocol of the first node and the second node; A task scheduling specification protocol of the first node and the second node.

10. An interconnection and interworking device based on a federated learning algorithm, characterized in that, The application is applied to a cross-platform architecture comprising at least two nodes; Each of the nodes is provided with an algorithm component; the method comprises: An acquisition unit is configured to acquire task information to be federated learned by a first node, the first node being any one of the at least two nodes; A loading unit is configured to call the algorithm component to load a target algorithm for a second node based on the task information, to obtain configuration information of the target algorithm, the second node being any one of the at least two nodes except the first node; A determination unit is configured to determine resource information matching the task information in the second node; An adaptation unit is configured to adaptively adapt the first node and the second node for interconnection and intercommunication by using the configuration information and the resource information.

11. An interconnection and interworking architecture based on a federated learning algorithm, characterized in that, The system comprises at least two nodes, a control layer and a resource layer, each of the nodes is provided with an algorithm component, the algorithm component is connected with the control layer and the resource layer respectively, and The control layer is configured to acquire task information to be federated learned by a first node, the first node being any one of the at least two nodes; The algorithm component is configured to load a target algorithm for a second node based on the task information, to obtain configuration information of the target algorithm, the second node being any one of the at least two nodes except the first node; The resource layer is configured to determine resource information matching the task information in the second node; The algorithm component is further configured to adaptively adapt the first node and the second node for interconnection and intercommunication by using the configuration information and the resource information.

12. An interconnection and interworking device based on a federated learning algorithm, characterized in that, The system comprises: A memory is configured to store executable instructions; A processor is configured to execute the executable instructions stored in the memory, to implement the federated learning algorithm-based interconnection and intercommunication method according to any one of claims 1 to 9.

13. A computer-readable storage medium, characterized in that, Executable instructions are stored in the memory, and when executed by the processor, the federated learning algorithm-based interconnection and intercommunication method according to any one of claims 1 to 9 is implemented.

Images