Multi-party computing arrangement method and system
By building a multi-party computing system with graph structure, the problem of excessive development investment in diversified scenarios is solved, flexible configuration and efficient development are achieved, diverse needs are met, and the scalability of the system is improved.
Patent Information
- Application Number
- CN202411119694.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-14
- Publication Date
- 2024-11-22
AI Technical Summary
In the scenario of diversified multi-party computing needs, the different user computing needs have led to excessive human resources investing in development, making it difficult to flexibly configure and expand.
By setting up nodes, building a graph structure and orchestrating multi-party computing logic, allowing flexibly configuring nodes according to requirements to realize multi-party computing systems with strong scalability and high development efficiency.
It realizes flexible configuration and efficient development in a variety of scenarios of multi-party computing needs, meets diverse needs, and improves development efficiency and system scalability.
Smart Images

Figure CN119004532A_ABST
Abstract
Description
Technical Field
[0001] The embodiments of this specification relate to the field of computer technology, and more particularly to a multi-party computing orchestration method and system. Background Art
[0002] Multi-party computing (MPC) is a secure computing method that allows multiple parties to perform computations and data analysis without exposing private data. The goal of MPC is to ensure that each participant can only obtain the results of the computation, but cannot obtain the private data of other participants. At present, in order to achieve a specific multi-party computing requirement, development and orchestration are usually carried out directly for the specific multi-party computing requirement, and a computing flow for the characteristic multi-party computing requirement is constructed. However, for some scenarios with diverse multi-party computing requirements, there will be problems of different user computing requirements and excessive human resources invested in development. Summary of the invention
[0003] The embodiments of this specification describe a multi-party computing orchestration method and system. The computing party can build a graph structure that meets the multi-party computing needs by setting nodes. The computing logic of the multi-party computing can be orchestrated by means of the graph structure. The nodes can be flexibly configured according to the multi-party computing needs. It has strong scalability and high development efficiency. Therefore, it can meet the diverse scenarios of multi-party computing needs.
[0004] According to a first aspect, a multi-party computing orchestration method is provided, wherein the multi-party includes a first computing party and a second computing party, and the method includes: the first computing party sets a plurality of first nodes of a first program, a connection relationship between the plurality of first nodes, and a connection relationship between the first program and a second program in the second computing party, wherein each first node includes a first processing module in the multi-party computing; the second computing party sets a plurality of second nodes of a second program, a connection relationship between the plurality of second nodes, and a connection relationship between the second program and the first program, wherein each second node includes a second processing module in the plurality of computing; the first computing party generates a first program corresponding to the first computing party based on the plurality of first nodes, the connection relationship between the plurality of first nodes, and the connection relationship between the first program and the second program; the second computing party generates a second program corresponding to the second computing party based on the plurality of second nodes, the connection relationship between the plurality of second nodes, and the connection relationship between the second program and the first program.
[0005] According to a second aspect, a multi-party computing orchestration system is provided, the system comprising: a first computing party, for setting a plurality of first nodes of a first program, a connection relationship between the plurality of first nodes, and a connection relationship between the first program and a second program in the second computing party, wherein each first node comprises a first processing module in the multi-party computing; a second computing party, for setting a plurality of second nodes of a second program, a connection relationship between the plurality of second nodes, and a connection relationship between the second program and the first program, wherein each second node comprises a second processing module in the plurality of computing; the first computing party is further used to generate a first program corresponding to the first computing party based on the plurality of first nodes, the connection relationship between the plurality of first nodes, and the connection relationship between the first program and the second program; the second computing party is further used to generate a second program corresponding to the second computing party based on the plurality of second nodes, the connection relationship between the plurality of second nodes, and the connection relationship between the second program and the first program.
[0006] According to the multi-party computing orchestration method and system provided by the embodiment of this specification, the first computing party can set multiple first nodes of the first program, the connection relationship between the multiple first nodes, and the connection relationship between the first program and the second program in the second computing party. The second computing party can set multiple second nodes of the second program, the connection relationship between the multiple second nodes, and the connection relationship between the second program and the first program. After that, the first computing party can generate a first program corresponding to the first computing party based on multiple first nodes, the connection relationship between the multiple first nodes, and the connection relationship between the first program and the second program. The second computing party can generate a second program corresponding to the second computing party based on multiple second nodes, the connection relationship between the multiple second nodes, and the connection relationship between the second program and the first program. The first program in the first computing party and the second program in the second computing party can be used to implement multi-party computing. Thus, the computing party can construct a graph structure that meets the multi-party computing requirements by setting nodes, and arrange the computing logic of multi-party computing by means of a graph structure. The nodes can be flexibly configured according to the multi-party computing requirements, with strong scalability and high development efficiency. Therefore, it can meet the diverse scenarios of multi-party computing requirements. BRIEF DESCRIPTION OF THE DRAWINGS
[0007] Figure 1 A schematic diagram showing an application scenario in which the embodiments of this specification can be applied;
[0008] Figure 2 A schematic diagram of a process flow of a multi-party computing orchestration method according to an embodiment is shown;
[0009] Figure 3 A schematic diagram showing an example of a tree model;
[0010] Figure 4 A schematic diagram showing an example of a graph structure;
[0011] Figure 5 Shows the use of Figure 4 The interactive diagram of the multi-party computing processing logic arranged by the graph structure shown in the figure performs multi-party joint prediction. DETAILED DESCRIPTION
[0012] The technical solution provided in this specification is further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the relevant inventions, rather than to limit the inventions. It should also be noted that, for ease of description, only the parts related to the relevant inventions are shown in the accompanying drawings. It should be noted that, in the absence of conflict, the embodiments of this specification and the features in the embodiments can be combined with each other.
[0013] As mentioned earlier, for some scenarios with diverse multi-party computing needs, there will be problems such as different user computing needs and excessive human resources invested in development.
[0014] To this end, the embodiments of this specification provide a multi-party computing orchestration method, which arranges the computing logic of multi-party computing in a graph structure manner, can flexibly configure nodes according to multi-party computing needs, has strong scalability, high development efficiency, and can meet diverse scenarios of multi-party computing needs. Figure 1 A schematic diagram showing an application scenario in which the embodiments of this specification can be applied is shown. Figure 1 In the example shown, it is assumed that two computing parties, computing party A and computing party B, are included in the multi-party computing. Computing party A can set multiple first nodes of the first program (for example, node A01, node A02, node A03 ... node A0N, etc.), the connection relationship between multiple first nodes, and the connection relationship between the first program and the second program in the second computing party. Here, each first node can include the first processing module in the multi-party computing. Computing party B can set multiple second nodes of the second program (for example, node B01, node B02, node B03 ... node B0M, etc.), the connection relationship between multiple second nodes, and the connection relationship between the second program and the first program. Here, each second node can include the second processing module in the multi-party computing. Afterwards, computing party A can generate the first program corresponding to computing party A based on multiple first nodes, the connection relationship between multiple first nodes, and the connection relationship between the first program and the second program. Computing party B can generate the second program corresponding to computing party B based on multiple second nodes, the connection relationship between multiple second nodes, and the connection relationship between the second program and the first program. Thus, the generated first program and second program can be used to implement multi-party computing. Understandably, Figure 1The multi-party computation in the example shown only involves two parties, computing party A and computing party B. In practice, multi-party computation may also involve more than two parties, and the specific arrangement of multi-party computation involving more than two parties can refer to the arrangement of two-party computation, which will not be described in detail here.
[0015] Continue to see Figure 2 , Figure 2 The flowchart of the multi-party computing arrangement method according to one embodiment is shown. It can be understood that each computing party of the method can be implemented by any device, equipment, platform, or device cluster with computing and processing capabilities. Figure 2 In the example shown, the multiple parties may include a first computing party and a second computing party. Figure 2 As shown, the multi-party computing orchestration method may include the following steps 201 to 204, specifically:
[0016] Step 201: The first computing party sets a plurality of first nodes of a first program, a connection relationship between the plurality of first nodes, and a connection relationship between the first program and a second program in a second computing party.
[0017] In this embodiment, the administrator of the first computing party can set multiple first nodes of the first program, the connection relationship between the multiple first nodes, and the connection relationship between the first program and the second program in the second computing party according to the actual multi-party computing needs. Here, each first node can include a first processing module in the multi-party computing, and each first processing module can correspond to a data processing operation in the multi-party computing process. For example, each node may be a function, method, class instance or other executable unit.
[0018] Here, the structure and process of the first program and the second program can be represented by a graph structure, for example, a directed graph, DAG (Directed Acyclic Graph), and the like. A directed graph is a graph whose edges have directions. In data analysis and graph theory, DAG (Directed Acyclic Graph) is a special graph structure in which the edges have directions and there is no path that can return to the starting point (i.e., there is no loop). The nodes and connections (edges) in DAG are generally used to represent and model dependencies in various practical scenarios. Here, the connection relationship between nodes can be used to represent the dependencies between data. For example, a node may need to rely on the output of the previous node.
[0019] In some implementations, the first computing party may set the plurality of first nodes of the first program and the connection relationship between the plurality of first nodes in the following manner:
[0020] The first computing party inputs node setting information, wherein the node setting information may include a plurality of first nodes and connection relationships between the plurality of first nodes. Here, the plurality of first nodes may include nodes selected from a preset node set, and / or data processing nodes developed by the first computing party. In this example, a node set may be pre-set, and the nodes in the node set may be displayed on the first computing party. In this way, when inputting the node setting information, the administrator of the first computing party may drag and drop nodes from the node set, and then connect the edges between the nodes. In addition, in order to meet different processing requirements, in addition to setting the node set, the administrator of the first computing party may also develop new data processing nodes according to actual needs. For example, it may be written in a programming language such as Java.
[0021] In some implementations, the first program may also include a network connection with the second program. When performing multi-party computing, a network connection may be established between the first program in the first computing party and the second program in the second computing party to send information. For example, the first program in the first computing party and the second program in the second computing party may be connected to the network via a network communication protocol (e.g., HTTP / HTTPS protocol, WebSocket protocol, MQTT protocol, etc.).
[0022] Step 202: The second computing party sets a plurality of second nodes of the second program, connection relationships between the plurality of second nodes, and a connection relationship between the second program and the first program.
[0023] In this embodiment, the administrator of the second computing party can set multiple second nodes of the second program, the connection relationship between the multiple second nodes, and the connection relationship between the second program and the first program on the second computing party according to the actual multi-party computing requirements. Here, each second node can include a second processing module in the multi-party computing, and each second processing module can correspond to a data processing operation in the multi-party computing process.
[0024] In some implementations, the first program may include multiple first nodes and call relationships between the multiple first nodes, and the second program may include multiple second nodes and call relationships between the multiple second nodes. Taking the graph structure of the first program and the second program as DAG as an example, in DAG, each node in DAG can correspond to a data processing operation in the multi-party computing process, and the directed edges in DAG can represent the dependency relationship between the data processing operations. The nodes in DAG can be called through the dependency relationship represented by the directed edges. For example, assume that there are four nodes N1, N2, N3 and N4 in DAG, and their dependency relationships are as follows: N1→N2, N1→N3, N3→N4. In this example, node N1 is the starting node. After node N1 is executed, since both N2 and N3 depend on node N1, at this time, N2 and N3 can be called for execution at the same time. When N3 is executed, node N4 can be called for execution because node N4 depends on node N3.
[0025] Step 203: The first computing party generates a first program corresponding to the first computing party based on the plurality of first nodes, the connection relationship between the plurality of first nodes, and the connection relationship between the first program and the second program.
[0026] In this embodiment, the first computing party may generate a first program corresponding to the first computing party based on the multiple first nodes set in step 201, the connection relationship between the multiple first nodes, and the connection relationship between the first program and the second program.
[0027] Step 204: The second computing party generates a second program corresponding to the second computing party based on the plurality of second nodes, the connection relationship between the plurality of second nodes, and the connection relationship between the second program and the first program.
[0028] In this embodiment, the second computing party may generate a second program corresponding to the second computing party based on the plurality of second nodes, the connection relationship between the plurality of second nodes, and the connection relationship between the second program and the first program set in step 202. The first program and the second program may interact with each other and together constitute a program for performing multi-party computing.
[0029] In some implementations, the multiple first nodes set by the first computing party may include a first access node, a first data acquisition node, a first model node, and a first connection node. Specifically:
[0030] The first access node may be used to receive a prediction request from a user device and send the prediction request to the first data acquisition node. Here, the prediction request may include a user ID. As an example, when the first computing party participates in multiple prediction services, in order to select the prediction service to be used, the prediction request may also include a service ID of the prediction service corresponding to the multi-party computing, etc.
[0031] The first data acquisition node can be used to obtain the first data of the first computing party according to the user identifier in the prediction request, and send the first data to the first model node. Here, the first computing party can be configured with a data storage unit to store the privacy data of the first computing party. Thus, the first data acquisition node can obtain the first data of the first computing party according to the user identifier.
[0032] The first model node can be used to obtain a first intermediate result by processing the first data, and send the first intermediate result to the first connection node.
[0033] The first connection node can be used to send the first intermediate result to the second program.
[0034] In some implementations, the plurality of second nodes set by the second computing party may include a second data acquisition node, a second model node, and a second connection node. Specifically:
[0035] The second connection node may be used to receive the user identifier and the first intermediate result from the first connection node. As an example, when the second computing party participates in multiple prediction services, in order to select the prediction service to be used, the first connection node may also send the service ID (identifier) of the prediction service corresponding to the multi-party computing to the second connection node, etc.
[0036] The second data acquisition node can be used to acquire the second data of the second computing party according to the user identifier. Here, the second computing party can be configured with a data storage unit to store the privacy data of the second computing party. Thus, the second data acquisition node can acquire the second data of the second computing party according to the user identifier.
[0037] The second model node can be used to process the second data, wherein the first model node corresponds to the second model node. Here, the first model node and the second model node can form a complete model, and the model can be various machine learning models, for example, LR (Logistic Regression) model, tree model, deep learning model, etc.
[0038] In some implementations, the first computing party may further include a first privacy protection node among the multiple first nodes set up, and the second computing party may further include a second privacy protection node among the multiple second nodes set up. The first privacy protection node may be used to perform privacy processing on the output result of the first model node, for example, by using homomorphic encryption, obfuscated circuits, etc. The second privacy protection node may be used to perform privacy processing on the output result of the second model node. Corresponding to the first privacy protection node, the second privacy protection node may use homomorphic encryption, obfuscated circuits, etc. for privacy processing.
[0039] As an example, the first model node and its corresponding second model node can form a logistic regression model, wherein the first model node can hold part of the weight of the logistic regression model, and the second model node can hold another part of the weight of the logistic regression model. For example, assuming that the LR model is: y = w_1*x_1+w_2*x_2+b, wherein the first model node can hold the model weight w_1, the first computing party can hold the feature x_1, the second model node can hold the model weight w_2 and the intercept b, and the second computing party can hold the feature x_2. Based on this, the first computing party can calculate the value of w_1*x_1, and use the homomorphic public key to encrypt Enc(w_1*x_1) to obtain y1, send y1 and the encrypted public key to the second computing party, and the second computing party calculates the value of w_2*x_2+b and encrypts it using the public key to obtain y2. Afterwards, the second computing party uses homomorphic encryption addition y1+y2 to obtain Enc(y) and sends it to the first computing party, and the first computing party obtains Enc(y) and decrypts it to obtain y. In this way, the multi-party joint prediction of the first computing party and the second computing party is completed.
[0040] As another example, the first model node and its corresponding second model node may form a tree model, wherein the first model node may hold some nodes of the tree model, and the second model node may hold another part of the nodes of the tree model. Figure 3 As shown, Figure 3 A schematic diagram showing an example of a tree model is shown in Figure 3Take the tree model shown as an example. In this example, there are two participants in the multi-party computing, computing party A and computing party B. The data of both parties AB will be used in the computing process. In the end, A will obtain the prediction result, and A is required not to obtain B's original data. In this example, the purpose of the tree model is to find a path based on the values of the features {a, b, c, d, e, f, g}. The value of the end point of this path (i.e., the leaf node) is the output of the tree model. In this example, it is assumed that the white nodes are the data owned by computing party A, i.e., {c, d, e, g}. The black nodes are the data owned by computing party B, i.e., {a, b, f}. Assume that the feature values of a user are {a: 20, b: 5, c: 10, d: 3, e: 8, f: 12, g: 9}. According to the above assumptions, computing party B has {a, b, f}. Computing party B first traverses the tree model to obtain the decision branches of the nodes related to itself. In this example, computing party B transmits the {1,4,11} decision branch (indicated by the dotted arrow in the figure) to computing party A. For example, computing party A can provide an interface for calling, for example, it can be an HTTP interface. In this way, computing party B can send the {1,4,11} decision branch and user ID to computing party A by calling the interface. Here, the HTTP interface is based on the HTTP protocol and can be used for data transmission between two computing devices. HTTP supports multiple data transmission methods, the most common of which are GET and POST methods. Among them, the GET method is usually used to obtain data, while the POST method is used to send data. In this example, computing party B can use the POST method to send data (including the {1,4,11} decision branch and user ID) to computing party A, and the data can be placed in the message body of the POST request for sending.
[0041] After computing party A obtains the decision branch {1,4,11} and the user ID, it can make predictions. Starting from the root node, the first is a, and the corresponding branch is {1,2}. Since 1 is already in the branch set, it is directly selected and enters the second layer. At this time, it is faced with identifying b. Similarly, since the subordinate branches of b are 3 or 4, and 4 is in the existing branch set, it is directly selected and enters the third layer. The current node is e. Since e is a local feature with a value of 8, it takes branch 10. Next, the leaf node is obtained, so the result of this tree is 6. In this way, the multi-party joint prediction of the first computing party and the second computing party is completed.
[0042] In some implementations, the multiple first nodes set by the first computing party may include multiple first model nodes, and the multiple second nodes set by the second computing party may include multiple second model nodes corresponding to the multiple first model nodes. Based on this, the multiple first nodes set by the first computing party may also include a result processing node, which can be used to process the output results of the multiple second model nodes to obtain the prediction results for the prediction request.
[0043] For example, it is assumed that the plurality of first nodes may include the first model node M01 and the first model node M02, and the plurality of second nodes may include the second model node M03 and the second model node M04 corresponding to the first model node M01 and the first model node M02. Here, the first model node M01 may correspond to the second model node M03, that is, the first model node M01 and the second model node M03 may form a complete model, which is referred to as model 1 in this example, and the second model node M03 may output the model prediction result of the model 1. The first model node M02 may correspond to the second model node M04, that is, the first model node M02 and the second model node M04 may form a complete model, which is referred to as model 2 in this example, and the second model node M04 may output the model prediction result of the model 2. Based on this, the result processing node in the first calculation party may process the model prediction results output by the second model node M03 and the second model node M04 (for example, weighting, statistical analysis, etc.) to obtain the prediction result for the prediction request.
[0044] In order to more clearly describe the diagram structure corresponding to the program in the embodiments of this specification, Figure 4 A schematic diagram showing an example of a graph structure. Figure 4 The example shown may include two computing parties, namely computing party G and computing party H.
[0045] The computing party G may include a first access node, a first front-end system node, a first data acquisition node, a first data preprocessing node, a first connection node, two first model nodes, a result processing node, etc. The connection relationship between the nodes in the computing party G is as follows: Figure 4 shown.
[0046] Here, the first access node may be used to receive a prediction request from the user equipment and send the prediction request to the first front-end system node, wherein the prediction request may include a user ID.
[0047] The first front-end system node may be a node set according to actual computing requirements, and may be a general node or a customized node. As an example, the first front-end system node may be used to process logic such as authentication, billing, and load balancing.
[0048] The first data acquisition node can be used to acquire the first data on the computing party G side according to the user ID. The first data may be the characteristic data of the user held by the computing party G side, and the characteristic data is private data and cannot be leaked to other parties during the multi-party computing process. As an example, the first data acquisition node may acquire data through PUSH, PULL, and the like. If the PUSH method is used, it means that the characteristic data is acquired by the first front-end system node according to the user ID and pushed to the first data acquisition node. If the PULL method is used, it is necessary to configure the characteristic data acquisition method at the first data acquisition node, for example, by acquiring it from a pre-set DB (Database), or by acquiring it through a URL (Uniform Resource Locator), and so on.
[0049] The first data preprocessing node can be used to transform the feature data acquired by the first data acquisition node, for example, outlier processing, type conversion, etc. It can be understood that the data processing logic corresponding to the first data preprocessing node can be configured with custom rules using scripting languages such as JS (JavaScript) according to actual needs.
[0050] The two first model nodes may include a first model node corresponding to the LR model and a first model node corresponding to the tree model.
[0051] The first connection node can be used to send the processing results of the first model node corresponding to the LR model and the processing results of the first model node corresponding to the tree model to the processing program corresponding to the computing party H. It can be understood that in addition to the processing results of the first model node, the first connection node can also send other information, such as user ID, service ID, etc., according to the needs of multi-party computing.
[0052] The result processing node can be used to process the model prediction results corresponding to the LR model and the tree model to obtain the prediction results for the prediction request.
[0053] The computing side H may include a second connection node, a second front-end system node, a second data acquisition node, a second data preprocessing node, and two second model nodes, etc. The connection relationship between the nodes in the computing side H is as follows: Figure 4 shown.
[0054] The second connection node can be used to receive information from the first connection node of the computing party G, for example, including the processing results of the first model node corresponding to the LR model, the processing results of the first model node corresponding to the tree model, and the user ID, etc.
[0055] The second front-end system node may be a node set according to actual computing requirements, and may be a general node or a customized node. As an example, the second front-end system node may be used to process logic such as authentication, billing, and load balancing.
[0056] The second data acquisition node can be used to acquire the second data on the computing party H side according to the user ID. The second data can be the characteristic data of the user held by the computing party H side, and the characteristic data is private data. Similar to the first data acquisition node, the second data acquisition node can also acquire data by PUSH (push), PULL (pull), etc.
[0057] The second data preprocessing node can be used to transform the feature data acquired by the second data acquisition node, for example, outlier processing, type conversion, etc. It can be understood that the data processing logic corresponding to the second data preprocessing node can be configured with custom rules using scripting languages such as JS (JavaScript) according to actual needs.
[0058] The two second model nodes may include a second model node corresponding to the LR model and a second model node corresponding to the tree model. The second model node corresponding to the LR model corresponds to the first model node corresponding to the LR model. The second model node corresponding to the tree model corresponds to the first model node corresponding to the tree model. After the two second model nodes process the data, the processing results of the second model node corresponding to the LR model and the processing results of the second model node corresponding to the tree model can be fed back to the processing program corresponding to the calculation party G.
[0059] Understandably, Figure 4 The example only includes two parties, the computing party G and the computing party H. In practice, according to actual needs, one computing party G and multiple computing parties H may be set. Figure 4 The nodes configured in the computing square G and computing square H, the connection mode between the nodes, and the direction of the edges between the nodes are also only for illustration, and are not intended to limit the configured nodes, the connection mode between the nodes, and the direction of the edges between the nodes. For example, Figure 4 In the example shown, the two second model nodes directly feed back the processing results to the processing program corresponding to the computing party G. In practice, according to actual needs, the two second model nodes can also send the processing results to the second connection node, which is fed back to the first connection node by the second connection node, and then fed back to the processing program corresponding to the computing party G. In practice, various nodes can be set in each computing party according to actual needs.
[0060] use Figure 4 The multi-party computing processing logic arranged in the graph structure shown can be used to perform multi-party joint prediction. Figure 5 As shown, Figure 5 Shows the use of Figure 4 The interactive diagram of the multi-party computing processing logic arranged by the graph structure shown in the figure performs multi-party joint prediction. Figure 5 The interactive diagram shown includes the following steps 501 to 508, specifically:
[0061] Step 501: The computing party G receives a prediction request sent by a user equipment.
[0062] In this example, the computing party G may use the first access node to receive the prediction request, wherein the prediction request may include a user ID and a service ID. The computing party G may also use the first front-end system node to process the prediction request. The computing party G may also query the first program corresponding to the service ID according to the service ID.
[0063] Step 502: The computing party G obtains the first data according to the user ID.
[0064] In this example, the computing party G can use the first data acquisition node to acquire the first data according to the user ID.
[0065] Step 503: the calculator G processes the first data to obtain an intermediate result 1.
[0066] In this example, the calculation party G can use the first data preprocessing node, two first model nodes, etc. to process the first data to obtain the intermediate result 1.
[0067] Step 504, computing party G sends the user ID, service ID and intermediate result 1 to computing party H.
[0068] In this example, the computing party G can use the first connection node to send information such as user ID, service ID and intermediate result 1 to the computing party H.
[0069] Step 504: The computing party H obtains the second data according to the user ID.
[0070] In this example, the computing party H can use the second connection node to receive information such as the user ID, service ID, and intermediate result 1 sent by the computing party G. The computing party H can also use the second front-end system node to process the received information. The computing party H can also query the second program corresponding to the service ID according to the service ID. The computing party H can use the second data acquisition node to obtain the second data according to the user ID.
[0071] Step 505 , the calculator H processes the second data to obtain an intermediate result 2 .
[0072] In this example, the computing party H can use the second data preprocessing node and two second model nodes to process the second data and the intermediate result 1 to obtain the intermediate result 2.
[0073] Step 506, computing party H sends the intermediate result 2 to computing party G.
[0074] In this example, the computing party H can feed back the intermediate result 2 to the computing party G.
[0075] Step 507: the calculator G processes the intermediate result 2 to obtain a predicted scoring result.
[0076] In this example, computing party G can receive intermediate result 2 from computing party H, and can use the result processing node to process intermediate result 2 to obtain a predicted scoring result.
[0077] Step 508: The computing party G sends the predicted scoring result to the user device.
[0078] In this example, the computing party G may feed back the prediction scoring result corresponding to the prediction request sent by the user device to the user device.
[0079] Reviewing the above process, in the embodiment of this specification, the first computing party can set multiple first nodes of the first program, the connection relationship between the multiple first nodes, and the connection relationship between the first program and the second program in the second computing party. The second computing party can set multiple second nodes of the second program, the connection relationship between the multiple second nodes, and the connection relationship between the second program and the first program. After that, the first computing party can generate a first program corresponding to the first computing party based on multiple first nodes, the connection relationship between the multiple first nodes, and the connection relationship between the first program and the second program. The second computing party can generate a second program corresponding to the second computing party based on multiple second nodes, the connection relationship between the multiple second nodes, and the connection relationship between the second program and the first program. The first program in the first computing party and the second program in the second computing party can be used to implement multi-party computing. Thus, the computing party can construct a graph structure that meets the needs of multi-party computing by setting nodes, and arrange the computing logic of multi-party computing by means of a graph structure. The nodes can be flexibly configured according to the needs of multi-party computing, with strong scalability and high development efficiency. Therefore, it can meet the diverse scenarios of multi-party computing needs.
[0080] According to another embodiment, a multi-party computing orchestration system is provided. The multi-party computing orchestration system may include a first computing party and a second computing party, specifically:
[0081] A first computing party, used to set a plurality of first nodes of a first program, a connection relationship between the plurality of first nodes, and a connection relationship between the first program and a second program in the second computing party, wherein each first node includes a first processing module in the multi-party computing;
[0082] A second computing party, used to set a plurality of second nodes of a second program, a connection relationship between the plurality of second nodes, and a connection relationship between the second program and the first program, wherein each second node includes a second processing module in the plurality of computings;
[0083] The first computing party is further configured to generate a first program corresponding to the first computing party based on the plurality of first nodes, the connection relationship between the plurality of first nodes, and the connection relationship between the first program and the second program;
[0084] The second computing party is further used to generate a second program corresponding to the second computing party based on the plurality of second nodes, the connection relationship between the plurality of second nodes, and the connection relationship between the second program and the first program.
[0085] In some optional implementations of this embodiment, the first program includes the multiple first nodes and the calling relationships between the multiple first nodes, and the second program includes the multiple second nodes and the calling relationships between the multiple second nodes.
[0086] In some optional implementations of this embodiment, the first program further includes a network connection with the second program.
[0087] In some optional implementations of this embodiment, the above-mentioned multiple first nodes include a first access node, a first data acquisition node, a first model node, and a first connection node, wherein: the above-mentioned first access node is used to receive a prediction request from a user device, and send the above-mentioned prediction request to the above-mentioned first data acquisition node; the above-mentioned first data acquisition node is used to obtain the first data of the above-mentioned first calculation party according to the user identifier in the above-mentioned prediction request, and send the above-mentioned first data to the above-mentioned first model node; the above-mentioned first model node is used to obtain a first intermediate result by processing the above-mentioned first data, and send the above-mentioned first intermediate result to the above-mentioned first connection node; the above-mentioned first connection node is used to send the above-mentioned first intermediate result to the above-mentioned second program.
[0088] In some optional implementations of this embodiment, the above-mentioned multiple second nodes include a second data acquisition node, a second model node, and a second connection node, wherein: the above-mentioned second connection node is used to receive the above-mentioned user identifier and the first intermediate result from the above-mentioned first connection node; the above-mentioned second data acquisition node is used to obtain the second data of the above-mentioned second calculation party according to the above-mentioned user identifier; the above-mentioned second model node is used to process the above-mentioned second data, and the above-mentioned first model node corresponds to the above-mentioned second model node.
[0089] In some optional implementations of the present embodiment, the multiple first nodes include multiple first model nodes, the multiple second nodes include multiple second model nodes corresponding to the multiple first model nodes, and the multiple first nodes also include result processing nodes, which are used to process output results of the multiple second model nodes to obtain prediction results for the prediction requests.
[0090] In some optional implementations of this embodiment, the above-mentioned multiple first nodes include a first privacy protection node, the above-mentioned multiple second nodes include a second privacy protection node, the above-mentioned first privacy protection node is used to perform privacy processing on the output result of the first model node, and the above-mentioned second privacy protection node is used to perform privacy processing on the output result of the second model node.
[0091] In some optional implementations of the present embodiment, the first computing party sets multiple first nodes of the first program and the connection relationship between the multiple first nodes, including: the first computing party inputs node setting information, wherein the node setting information includes the multiple first nodes and the connection relationship between the multiple first nodes, the multiple first nodes include nodes selected from a preset node set, and / or data processing nodes developed by the first computing party.
[0092] In some optional implementations of this embodiment, the first model node and its corresponding second model node form a logistic regression model, the first model node holds part of the weight of the above logistic regression model, and the second model node holds another part of the weight of the above logistic regression model.
[0093] In some optional implementations of this embodiment, the first model node and its corresponding second model node form a tree model, the first model node holds some nodes of the tree model, and the second model node holds another part of the nodes of the tree model.
[0094] Those of ordinary skill in the art should further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Those of ordinary skill in the art may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.
[0095] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be implemented using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.
[0096] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.
Claims
1. A multi-party computing arrangement method, wherein the multi-party includes a first computing party and a second computing party, wherein the method include: The first computing party sets a plurality of first nodes of a first program, a connection relationship between the plurality of first nodes, and a connection relationship between the first program and a second program in the second computing party, wherein each first node includes a first processing module in the multi-party computing; The second computing party sets a plurality of second nodes of the second program, a connection relationship between the plurality of second nodes, and a connection relationship between the second program and the first program, wherein each second node includes a second processing module in the plurality of computings; The first computing party generates a first program corresponding to the first computing party based on the plurality of first nodes, the connection relationship between the plurality of first nodes, and the connection relationship between the first program and the second program; The second computing party generates a second program corresponding to the second computing party based on the plurality of second nodes, the connection relationship between the plurality of second nodes, and the connection relationship between the second program and the first program.
2. The method according to claim 1, in, The first program includes the plurality of first nodes and the calling relationships between the plurality of first nodes, and the second program includes the plurality of second nodes and the calling relationships between the plurality of second nodes.
3. The method according to claim 1, in, The first program also includes a network connection to the second program.
4. The method according to claim 1, in, The multiple first nodes include a first access node, a first data acquisition node, a first model node, and a first connection node, wherein: The first access node is used to receive a prediction request from a user equipment, and send the prediction request to the first data acquisition node; The first data acquisition node is used to acquire the first data of the first computing party according to the user identifier in the prediction request, and send the first data to the first model node; The first model node is used to obtain a first intermediate result by processing the first data, and send the first intermediate result to the first connection node; The first connection node is used to send the first intermediate result to the second program.
5. The method according to claim 4, in, The plurality of second nodes include a second data acquisition node, a second model node, and a second connection node, wherein: The second connection node is used to receive the user identifier and the first intermediate result from the first connection node; The second data acquisition node is used to acquire the second data of the second computing party according to the user identifier; The second model node is used to process the second data, and the first model node corresponds to the second model node.
6. The method according to claim 5, in, The multiple first nodes include multiple first model nodes, the multiple second nodes include multiple second model nodes corresponding to the multiple first model nodes, and the multiple first nodes also include a result processing node, which is used to process the output results of the multiple second model nodes to obtain a prediction result for the prediction request.
7. The method according to claim 5 or 6, in, The multiple first nodes include a first privacy protection node, and the multiple second nodes include a second privacy protection node. The first privacy protection node is used to perform privacy processing on the output result of the first model node, and the second privacy protection node is used to perform privacy processing on the output result of the second model node.
8. The method according to claim 1, in, The first computing party sets a plurality of first nodes of the first program and connection relationships between the plurality of first nodes, including: The first computing party inputs node setting information, wherein the node setting information includes connection relationships between the multiple first nodes and the multiple first nodes, the multiple first nodes include nodes selected from a preset node set, and / or data processing nodes developed by the first computing party.
9. The method according to claim 5, in, The first model node and its corresponding second model node form a logistic regression model, the first model node holds a part of the weight of the logistic regression model, and the second model node holds another part of the weight of the logistic regression model.
10. The method according to claim 5, in, The first model node and its corresponding second model node form a tree model, the first model node holds part of the nodes of the tree model, and the second model node holds another part of the nodes of the tree model.
11. A multi-party computing orchestration system, the system include: A first computing party, used to set a plurality of first nodes of a first program, a connection relationship between the plurality of first nodes, and a connection relationship between the first program and a second program in the second computing party, wherein each first node includes a first processing module in the multi-party computing; A second computing party, used to set a plurality of second nodes of a second program, a connection relationship between the plurality of second nodes, and a connection relationship between the second program and the first program, wherein each second node includes a second processing module in the plurality of computings; The first computing party is further configured to generate a first program corresponding to the first computing party based on the plurality of first nodes, the connection relationship between the plurality of first nodes, and the connection relationship between the first program and the second program; The second computing party is further used to generate a second program corresponding to the second computing party based on the multiple second nodes, the connection relationship between the multiple second nodes, and the connection relationship between the second program and the first program.
