Privacy computing device and privacy computing method

Abstract

The embodiments of the present specification provide a privacy computing device and a privacy computing method. The privacy computing device is arranged on a computing node to perform a privacy task in cooperation with multiple parties. The privacy computing device arranged on a single computing node can include a control module, a synchronization controller, and an execution module. The control module is used for creating, scheduling, and issuing a privacy computing task. The synchronization controller is used for listening to the control module of another computing node to synchronize resources and parameters of a privacy computing task related to the current computing node created by the control module of the other computing node to the local. The execution module is used for obtaining and executing a privacy computing task related to the current computing node by listening to the local control module or the control module of the other computing node. By arranging the above privacy computing device, the computing node can work in different modes by activating different modules to expand the privacy computing network.

Description

Technical Field

[0001] This specification relates to the field of multi-party secure computation technology, and in particular to a privacy computing device and a privacy computing method. Background Technology

[0002] The term "privacy computing" was formally introduced in the 2016 report, "Research Scope and Development Trends of Privacy Computing," which defines it as: "Computational theories and methods oriented towards the protection of privacy information throughout its entire lifecycle; a computable model and axiomatic system for privacy measurement, the cost of privacy leakage, and the complexity of privacy protection and analysis when the ownership, management, and usage rights of privacy information are separated." Essentially, privacy computing addresses data service issues such as data circulation and data application while protecting data privacy.

[0003] With the development of privacy-preserving computation technology, some business operations may require joint modeling by multiple business parties (such as multiple banks or wealth management institutions) to break down data silos. Furthermore, there is a need for various business parties to quickly and conveniently jointly execute privacy-preserving computation tasks. Therefore, it is possible that many business parties will join a large privacy-preserving computation network. In conventional multi-party privacy-preserving computation frameworks, regardless of the presence of a trusted third party, each business party acts as a computing node, and the architecture and number of computing nodes are relatively fixed. However, with the development of multi-party joint secure computation business needs, there may be a demand for more business parties to join the privacy-preserving computation framework as computing nodes, forming a privacy-preserving computation network. Therefore, how to construct a privacy-preserving computation network is a crucial technical problem that needs to be solved to meet these new requirements. Summary of the Invention

[0004] This specification describes one or more embodiments of a scalable privacy computing system and a device expansion method within the privacy computing system, to address one or more of the problems mentioned in the background art.

[0005] According to a first aspect, a privacy computing device is provided, comprising a control module, a synchronization controller, and an execution module, wherein:

[0006] The control module is used for creating, scheduling, and distributing privacy computing tasks;

[0007] The synchronization controller is used to monitor the control modules of other computing nodes in order to synchronize the resources and parameters of privacy computing tasks related to the current computing node created by the control modules of other computing nodes to the local machine.

[0008] The execution module is used to execute privacy computing tasks related to the current computing node. These privacy computing tasks are obtained by listening to the local control module or the control module of other computing nodes.

[0009] In one embodiment, the operating modes of the device include at least one of the following: master node mode, slave node mode, and peer-to-peer mode. The master node mode and slave node mode are suitable for centralized federated computing architectures. The device operates in master node mode when a single computing node is configured as a central node, and operates in slave node mode when other computing nodes are configured as central nodes. The peer-to-peer mode is suitable for decentralized federated computing architectures.

[0010] In one embodiment, in master node mode, the device creates, schedules, and distributes privacy computing tasks through a control module.

[0011] In one embodiment, in slave node mode, the device monitors the master node's control module through the execution module so that if the master node creates a privacy computing task related to the current computing node through the control module, the device can execute the privacy computing task via the execution module.

[0012] In one embodiment, in peer-to-peer mode, the device monitors the control modules of other nodes through the synchronization controller. If it detects that other nodes have created privacy computing tasks related to the current computing node, it synchronizes the resource and parameter information of the privacy computing task to the current computing node and executes the privacy computing task through the execution module.

[0013] According to a second aspect, a privacy computing method executed by a first node is provided. The first node is equipped with the privacy computing device as described in the first aspect and forms a decentralized privacy computing architecture with a second node. Simultaneously, it also forms a centralized privacy computing architecture as a master node and at least one other computing node as a slave node. The privacy computing method includes: monitoring the second node through a synchronization controller to locally synchronize first task parameters corresponding to the first privacy computing task when the second node creates a first privacy computing task related to the first node; and creating multiple subtasks for the first privacy computing task through a control module, so that each slave node, based on the processing of the corresponding subtasks, jointly completes the first privacy computing task with the second node.

[0014] In one embodiment, the first task parameters include at least one of the following: the start command for the first privacy computing task, the input file address or path, and the output file path.

[0015] In one embodiment, each slave node communicates with the second node while processing its respective subtask.

[0016] According to a third aspect, a privacy computing method executed by a third node is provided. The third node is equipped with the privacy computing device as described in the first aspect and forms a centralized privacy computing architecture with a fourth node as the master node, while also forming a decentralized privacy computing architecture with a fifth node. The privacy computing method includes: monitoring the fourth node and the fifth node through a synchronization controller; upon detecting a second privacy computing task related to the third node, synchronizing the second task parameters corresponding to the second privacy computing task to a local control module; and, based on the execution module's monitoring of the local control module, performing relevant privacy computing in the second privacy computing task using the second task parameters.

[0017] In one embodiment, when the second privacy computation task originates from the fifth node, the step of performing relevant privacy computations in the second privacy computation task based on the execution module's monitoring of the local control module and utilizing the second task parameters includes: executing the second privacy computation task through the local execution module; during the execution of the second privacy computation task, the third node and the fifth node communicate data under the premise of privacy protection.

[0018] In one embodiment, when the second privacy computation task originates from a fourth node, the method further includes: creating a subtask of the second privacy computation task through a local control module, wherein a single subtask is associated with a third node or a fifth node; the step of performing relevant privacy computations in the second privacy computation task using the second task parameters based on the execution module's monitoring of the local control module includes: monitoring and processing the subtask associated with the third node through the execution module.

[0019] In a further embodiment, the subtask associated with the fifth node is synchronized locally by the fifth node's monitoring of the control module and executed; wherein the third node communicates with the fifth node and the fourth node at the data layer, respectively.

[0020] According to a fourth aspect, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed in a computer, causes the computer to perform the method of the second or third aspect.

[0021] The apparatus and method provided in the embodiments of this specification enable the setup of privacy computing devices on computing nodes to perform multi-party collaborative privacy tasks. The privacy computing device on a single computing node may include a control module, a synchronization controller, and an execution module. The control module is used for creating, scheduling, and distributing privacy computing tasks. The synchronization controller monitors the control modules of other computing nodes to synchronize the resources and parameters of privacy computing tasks created by the control modules of other computing nodes related to the current computing node to the local system. The execution module obtains and executes privacy computing tasks related to the current computing node by monitoring the local control module or the control modules of other computing nodes.

[0022] By configuring the above privacy computing devices, computing nodes can activate different modules to operate in at least one of the following modes: master node mode, slave node mode, and peer-to-peer mode. When a single computing node operates in both modes simultaneously, it corresponds to two different multi-party secure computing architectures. Thus, by ensuring compatibility between the two modes, the privacy computing network can be expanded, and the privacy computing system can be scaled up. Attached Figure Description

[0023] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1a A schematic diagram of an implementation architecture for decentralized multi-party secure computation is shown.

[0025] Figure 1b A schematic diagram of a centralized multi-party secure computation implementation architecture is shown.

[0026] Figure 2 A schematic block diagram of a privacy computing device according to embodiments of this specification is shown;

[0027] Figure 3 This diagram illustrates a privacy computing interaction where a node in a centralized privacy computing implementation architecture publishes privacy computing tasks to other nodes.

[0028] Figure 4 This diagram illustrates the privacy computing interactions between nodes in a decentralized privacy computing implementation architecture.

[0029] Figure 5 A schematic diagram of a compatible implementation architecture for centralized and decentralized privacy computing is shown in a specific example.

[0030] Figure 6This diagram illustrates a compatible implementation architecture for centralized and decentralized multi-party secure computation in another specific example.

[0031] Figure 7 A schematic diagram of a privacy computing process according to an embodiment of this specification is shown;

[0032] Figure 8 A schematic diagram of a privacy computation process according to another embodiment of this specification is shown. Detailed Implementation

[0033] The solution provided in this specification will now be described with reference to the accompanying drawings.

[0034] Protecting privacy-preserving data is one of the core principles of privacy-preserving computation. Tasks performed through privacy-preserving computation can be called privacy-preserving computation tasks, which involve multiple parties collaboratively performing computations and analyses on the data while ensuring the security and privacy of the original data. Examples of privacy-preserving computation tasks include privacy-preserving data analysis, collaborative machine learning, and so on.

[0035] A common application of privacy-preserving computation is Secure Multi-Party Computation (SMPC). SMPC solves the problem of privacy-preserving collaborative computation among a group of untrusted parties, ensuring the independence of inputs, the correctness of computation, and preventing the disclosure of input values ​​to other members participating in the computation.

[0036] Generally, secure multi-party computation can be divided into centralized and decentralized architectures. Figure 1a A decentralized multi-party secure computation architecture is shown, in which each participant (such as member 1, member 2, member 3) directly interacts with encrypted data and completes relevant privacy computation tasks in a encrypted state through encrypted computation methods such as homomorphic encryption and secret sharing. Figure 1b This paper illustrates a decentralized multi-party secure computation architecture. In a centralized architecture, each participant can transmit its local computation results to a trusted third party in plaintext or ciphertext. The trusted third party then performs the fusion computation of the data from all parties and provides feedback on the results. Compared to a decentralized architecture, a centralized architecture can reduce redundant communication caused by privacy protection requirements, but this requires a trusted third party. In a decentralized multi-party secure computation architecture, establishing a trusted communication link between two nodes is costly.

[0037] In the process of completing a privacy-preserving computation task, each participating party can be regarded as a computing node. Typically, in a centralized or decentralized multi-party secure computation architecture, the number of computing nodes is relatively fixed, the relationships between computing nodes are relatively fixed, and it is difficult to expand the scale of computing nodes.

[0038] In view of this, this specification provides a technical concept that, by setting up privacy computing devices on each computing node, enables each computing node to be compatible with both centralized and decentralized implementation architectures. This allows for the formation of a more complex privacy computing network through the connection of some nodes, and achieves scalability of the privacy computing system.

[0039] Figure 2 A schematic block diagram of a privacy device is shown. Figure 2 As shown, the privacy device 200 may include a synchronization controller (e.g., sync controller) 201, a control module (e.g., k3s) 202, and an execution module (e.g., agent) 203. This privacy computing device can be located at any computing node in the privacy computing process, such as... Figure 1a , Figure 1b Member 1, Member 2, Member 3, etc.

[0040] Specifically, the control module 202 can be used to create, schedule, and distribute privacy computing tasks; the synchronization controller 201 can be used to monitor the control modules of other computing nodes to synchronize the resources and parameters of privacy computing tasks related to the current computing node created by the control modules of other computing nodes to the local machine; and the execution module 203 can be used to execute privacy computing tasks related to the current computing node. The privacy computing tasks can be obtained by the execution module 203 monitoring the local control module or the control modules of other computing nodes.

[0041] After the computing nodes are equipped with the above privacy computing devices, they can activate some or all of the modules to work in different modes, thereby realizing a centralized multi-party secure computing architecture or a decentralized multi-party secure computing architecture.

[0042] Figure 3 This paper illustrates a centralized multi-party secure computation architecture with two computing nodes, Alice and Bob, using node 3 as the master node. In this architecture, when Alice and Bob are performing joint data processing, they can configure node 3 as a trusted third party or service provider. Node 3 then acts as the master node, scheduling and distributing privacy-preserving computation tasks to Alice and Bob. Alice and Bob, as slave nodes, execute the privacy-preserving computation tasks created by node 3 that are relevant to them.

[0043] Specifically, for a computational task, Node 3, acting as the master node (also known as the control node), can break it down into multiple subtasks using a control module (such as k3s) in a local privacy computing device, and distribute these subtasks through containers or processes (such as... Figure 3The computing engine secretflow (the entity that executes tasks) distributes tasks to computing nodes Alice and / or Bob. Alice or Bob can then retrieve the tasks distributed locally through a long-connection monitoring mechanism (such as list-watch) of the execution module (e.g., agent) of their local privacy computing device. They then start and run the relevant processes or containers to execute the included privacy computing tasks, such as feature encoding tasks in a longitudinal federated learning process.

[0044] As an example, the master node can distribute privacy-preserving computation tasks to compute nodes using container technology. Containers can contain image files of the relevant tasks. Those skilled in the art will understand that container management can be performed using pods, etc., with a pod serving as the smallest unit of container management; one pod can hold one or more containers. Each pod has its own virtual IP address, and one node can correspond to multiple pods. The master node will automatically schedule pods to run on which node based on load. Worker nodes, upon receiving the task-distributed pods, can launch and run them as containers using pre-configured privacy-preserving computation devices.

[0045] During the execution of privacy-preserving computation tasks across computing nodes, data, intermediate results, or final task results can be exchanged. For example, in vertical federated learning, a single computing node, acting as a feature data provider, can pass intermediate processing results of its local model on to a computing node acting as a label data provider. Understandably, each computing node can also feed back its local privacy-preserving computation task execution results to the master node, which will then aggregate them and distribute subsequent data and tasks based on the aggregated results. For instance, in a horizontal federated learning architecture, each computing node can feed back its local model gradients, undetermined parameter update values, and other data to the master node, which will then aggregate these into global model gradients, undetermined parameter update values, etc., and create a local model parameter update task to distribute to each computing node to complete the synchronization of local model parameters.

[0046] exist Figure 3 In the implementation architecture shown, the control plane of each node is located on the master node, and the computing nodes can communicate at the task data plane (see reference). Figure 3 (Solid lines indicate the control link, and dashed lines indicate the data link), thereby enabling the construction of a corresponding clustered control platform and reducing the communication volume between computing nodes. Figure 3 The number of slave nodes shown is for illustrative purposes only; in practice, there can be one or more (e.g., 10), and no limitation is made here.

[0047] Figure 4This illustration shows two computing nodes, Alice and Bob, performing privacy-preserving computation tasks without a trusted third party or service provider (i.e., without a central authority). It can be understood that in this decentralized architecture, all nodes perform privacy-preserving computation tasks as equals; each computing node can be referred to as an autonomous node.

[0048] like Figure 4 As shown, in this implementation architecture, each computing node can create privacy-preserving computing tasks through a local control module (such as k3s). Since each computing node initiates joint tasks locally, other nodes in the same architecture need to be aware of these privacy-preserving computing tasks, and these nodes can cooperate to complete the execution of these tasks. Therefore, in this decentralized implementation architecture where each node is an autonomous node, a synchronization controller is introduced to monitor each other's privacy-preserving computing tasks. This controller is used to detect privacy-preserving computing tasks created by other autonomous nodes that are related to itself, as well as to synchronize task parameters and resources. Figure 4 As illustrated, for example, when the Alice node creates a joint privacy computation task with the Bob node through its local control module (k3s), the synccontroller on the Bob node's side monitors the Alice control module (k3s) via a long-connection listening mechanism (list-watch). It can then perceive the task and related resources or parameters, synchronizing these resources or parameters to its local control module. This allows the local execution module (agent) to execute the joint privacy computation task with the Alice node using the task-related resources or parameters. Conversely, when the Bob node creates a joint task with the Alice node, the Alice node can similarly perceive the task-related resources or parameters and execute the joint task with the Bob node. During the execution of privacy computation tasks, the Alice and Bob nodes can exchange task data. For example, intermediate processing results of feature data by the local model in longitudinal federated learning.

[0049] pass Figure 4 It is understood that in a decentralized multi-party secure computation architecture, all computing nodes are peers. A single computing node can create privacy-preserving computation tasks through a control module, and also perceive privacy-preserving computation tasks created by other nodes related to itself through a synchronization controller, executing these tasks via an execution module. In this scenario, communication between computing nodes can exist at both the task control layer and the data layer, such as... Figure 4 The solid lines represent the control links, and the dashed lines represent the data links.

[0050] In summary, through Figure 3 , Figure 4As shown, a single compute node, when configured as follows Figure 2 After the privacy computing device shown is activated, it can operate in three different modes by activating different modules:

[0051] The master node mode exists in a centralized privacy computing architecture. It can be configured as a master node by other computing nodes and can create, schedule and distribute tasks through the control module.

[0052] The slave node mode exists in a centralized privacy computing architecture. Other computing nodes can be configured as master nodes, thus acting as slave nodes of the configured master nodes. Through the execution module, it can perceive the privacy computing tasks issued by the master node and communicate data with other slave nodes of the master node to jointly complete the privacy computing tasks.

[0053] In peer-to-peer mode, which exists in a decentralized privacy computing architecture, the peer-to-peer mode has an equal identity with other computing nodes in the privacy computing architecture. It uses a synchronization controller to perceive the privacy computing tasks related to itself built by other computing nodes and synchronizes the relevant task parameters to the local control module so that the execution module can process the relevant tasks.

[0054] As can be seen, a single computing node can be configured as a master node by other computing nodes, or configured as a master node by other computing nodes, thus operating in a centralized privacy computing architecture, or it can operate in a peer-to-peer mode with other nodes in a decentralized privacy computing architecture.

[0055] In a centralized privacy-preserving computing architecture, the master node corresponds to the master node mode, with activated modules including the control module, and the slave node corresponds to the slave node mode, with activated modules including the execution module. When a single computing node (e.g., node 3) configured as the master node by other computing nodes simultaneously processes privacy-preserving computing tasks with other nodes in a peer-to-peer mode, the master node can interact with these other nodes at the control level of privacy-preserving computing tasks (e.g., task creation, resource / parameter synchronization) through the control module and synchronization controller. During joint privacy-preserving computing task processing, the master node can distribute relevant privacy-preserving computing tasks to other computing nodes configured as the master node, which then process them using their execution modules (agents). These other computing nodes can interact with each other through computing engines (e.g., secretflow).

[0056] In a decentralized privacy-preserving computation architecture, each computing node operates in a peer-to-peer (or autonomous) mode, with activated modules including a synchronization controller, a control module, and an execution module. In this mode, a single computing node (an autonomous node in autonomous mode) can synchronize resources of privacy-preserving computation tasks created by the control modules of other nodes in the architecture through the synchronization controller. Then, the synchronized privacy-preserving computation tasks, resources, and parameters are passed to the local control module, which creates a local privacy-preserving computation task in the form of a container or process. Alternatively, the execution module can listen for privacy-preserving computation tasks from the control module via a long-connection listening mechanism. Upon detecting a privacy-preserving computation task, the execution module can launch the container or process to complete the privacy-preserving computation task. During the completion of the privacy-preserving computation task, secure data interaction with another computing node is also possible.

[0057] In addition, computing nodes in various modes can include a gateway, which is responsible for basic privacy computing functions such as routing authentication and network authentication between nodes, which will not be elaborated here.

[0058] It is worth noting that, for a single computing node, the above three working modes can be used simultaneously if they do not conflict.

[0059] In one embodiment, a single computing node can operate simultaneously in peer-to-peer and slave modes. In this case, joint privacy-preserving computation tasks created by the corresponding master node can be jointly executed with other computing nodes in peer-to-peer mode via the control module and execution module. Related joint privacy-preserving computation tasks created by other computing nodes in peer-to-peer mode can be detected by the synchronization controller and executed by the control module and execution module.

[0060] As an example, see reference Figure 5 As shown, node 3, Bob node, and Joke node constitute a centralized privacy computing architecture. Bob node and Joke node can configure node 3 as a trusted third party or service provider, thus the control layer for Bob node and Joke node to complete privacy computing tasks is implemented by node 3. Specifically, node 3 can schedule and distribute privacy computing tasks to Bob node and Joke node through the control module (k3s). Bob node and Joke node can obtain privacy computing tasks from node 3 through a long-connection listening mechanism and complete the relevant privacy computing tasks through their local execution module (agent). Bob node and Joke node can interact at the data level, such as... Figure 5 The data link is shown by the dashed line.

[0061] On the other hand, Alice nodes and Bob nodes constitute a decentralized privacy computing architecture. For example... Figure 5As shown, Alice and Bob nodes each have a synchronization controller and a control module (k3s). A single compute node's synchronization controller can monitor other compute nodes' control modules (k3s) through a long-connection listener mechanism (list-watch), thereby becoming aware of task-related resources or parameters, such as the task's start command, input files, output file paths, etc., and synchronizing these resources or parameters to the local control module. This allows the local privacy computing agent to be aware of task-related resources or parameters and collaborate with other compute nodes to complete the privacy computing task. Alice and Bob nodes can interact at both the control plane and the data plane, such as... Figure 5 The solid lines represent the control links, and the dashed lines represent the data links.

[0062] It can be seen that, Figure 5 In the illustrated architecture, Bob can operate as a slave node in a centralized architecture and as a peer-to-peer node in a decentralized architecture. Furthermore, for the joint privacy computing tasks created by the corresponding master node (node ​​3), the Bob node can jointly execute them with other computing nodes in peer-to-peer mode through the control module and execution module. For related joint privacy computing tasks created by other computing nodes (Alice nodes) in peer-to-peer mode, they can be detected by the synchronization controller and executed by the control module and execution module. The Bob node can interact with both the Joke node and the Alice node at the data level, such as... Figure 5 The data link is shown by the dashed line.

[0063] In another embodiment, a single computing node can operate simultaneously in peer-to-peer mode and master-node mode. In this case, upon detecting a privacy-related computing task, the control module can distribute the task to a slave node configured as the master node for execution. In this scenario, the single computing node can be configured with a synchronization controller and control module, but does not require an execution module.

[0064] As a concrete example, see reference Figure 6 As shown, with Alice and Bob nodes configuring Node 3 as the master node, Alice, Bob, and Node 3 can form a centralized privacy computing architecture. Node 3 can handle the creation, scheduling, and distribution of control-level tasks in the joint privacy computing of Alice and Bob nodes. Alice and Bob nodes process data according to the tasks distributed by Node 3 and can perform data-level communication.

[0065] On the other hand, node 3 and the Joke node constitute a decentralized privacy computing architecture. For example... Figure 6 As shown, node 3 and the Joke node each have a synchronization controller and a k3s control module. They are aware of each other's resources and parameters and collaborate to complete privacy-related computation tasks created by the other node. When completing a privacy-related computation task, node 3 can distribute the relevant tasks to the Alice and Bob nodes, without needing to complete it itself. Therefore, node 3 and the Joke node can communicate at the control plane level, such as... Figure 6 The control link is shown by the solid line in the middle. Alice and Bob nodes, as the data processing system of node 3, can communicate with Joke node at the data layer, such as... Figure 6 The data link is shown by the dashed line.

[0066] in this way, Figure 6 The node 3 shown can operate in both peer-to-peer and master-node modes simultaneously. In this mode, in addition to creating privacy computing tasks related to the Joke, Alice, and Bob nodes, after detecting privacy computing tasks related to itself created by the Joke node, it can use the control module to distribute the relevant tasks to the slave nodes (i.e., the Alice and Bob nodes) configured as master nodes for execution. At this time, node 3 can be configured with a synchronization controller and control module without needing to configure an execution module.

[0067] In another embodiment, a computing node can also operate in both master node mode and slave node mode (not shown). In this case, privacy-preserving computing tasks obtained from other master nodes based on slave node mode can be split and distributed to other slave nodes that are master nodes by the control module, which will not be elaborated further here.

[0068] In other embodiments, it is possible that a single computing node can be configured to work in three working modes simultaneously, which will not be elaborated here.

[0069] The privacy computing device provided under the technical concept of this specification can be located on any computing node participating in privacy-preserving computation. Thus, since a single computing node can simultaneously operate in two or more modes, including master node mode, slave node mode, and peer-to-peer mode, the number of computing nodes or the connection relationships in the privacy computing system can be expanded based on the expansion of operating modes, forming a multi-level scalable operating mode, facilitating the construction of complex privacy computing networks.

[0070] On the other hand, this specification also provides a privacy-preserving computation method that can be applied to any computation node in privacy-preserving computation, for example, denoted as the first node. This first node can be any computer, device, or server with a certain computing capability, such as... Figure 6 Node 3 in the middle. The first node can be set as follows: Figure 2 The privacy computing device shown can be used with a second node (such as...) Figure 6 The Joke nodes in the dataset constitute a decentralized privacy computing architecture, performing collaborative privacy computing in a peer-to-peer mode. Simultaneously, the first node can also be configured as the master node by other computing nodes, thus forming a centralized privacy computing architecture. In this case, the other computing nodes act as slave nodes of the first node. Figure 6 Taking the architecture shown as an example, other computing nodes are such as Alice nodes or Bob nodes.

[0071] Figure 7 This is a schematic diagram of a privacy computation process according to one embodiment. Figure 7 As shown, the privacy computing process applicable to the first node above may include: Step 702, monitoring the second node through the synchronization controller, so as to synchronize the first task parameters corresponding to the first privacy computing task locally when the second node creates a first privacy computing task related to the first node; Step 704, creating multiple subtasks for the first privacy computing task through the control module, so that each slave node can work with the second node to complete the first privacy computing task based on the processing of the corresponding subtasks.

[0072] First, in step 702, when the control module of the second node is detected by the synchronization controller to create a first privacy computing task related to the first node, the first task parameters corresponding to the first privacy computing task are synchronized locally.

[0073] It's worth noting that when the first and second nodes operate as autonomous nodes in peer-to-peer mode, they need to perform joint business processing based on secure computation, while protecting their respective privacy data, such as jointly training a business model. In this case, the first and second nodes can each create corresponding privacy-preserving computation tasks according to the business processing flow. A privacy-preserving computation task can be a computation task that must be completed while ensuring that privacy data is not leaked to the other party. In the case of jointly training a model, this privacy-preserving computation task could be, for example, the joint computation of the sigmoid activation function.

[0074] Privacy-preserving computation tasks can be created by either the first node or the second node. Therefore, the first and second nodes can monitor each other's control modules in real time through a synchronization controller. Upon detecting a privacy-preserving computation task relevant to themselves, they can synchronize the task's resources and parameters through the synchronization controller. These resources and parameters are those related to the created privacy-preserving computation task, such as the task's start command, input file, and output file path. In the case of jointly trained models, resources and parameters related to the privacy-preserving computation task may include publicly available hyperparameters, such as gradient update step size and the number of training epochs.

[0075] The first node can synchronize the resources and parameters of the privacy computing task to its local control module via the synchronization controller. When the privacy computing task created by the second node and associated with the first node is designated as the first privacy computing task, the corresponding resources and parameters can be uniformly recorded as the first task parameters.

[0076] Then, in step 704, based on the first task parameters, the control module creates multiple subtasks for the first privacy computing task to be distributed downwards.

[0077] Since the first node is also configured as a master node by other computing nodes, it operates in master node mode. It may hold some data locally or none at all, while the privacy computing tasks can be completed by the other computing nodes configured as master nodes. Specifically, the first node can create new privacy computing tasks based on the first task parameters; these newly created privacy computing tasks can be considered subtasks of the first privacy computing task.

[0078] During the creation of subtasks, the first node can perform unified scheduling based on the data held by each slave node and its current load. For example, more subtasks can be created related to slave nodes with lower current loads. When creating subtasks, the first node can associate them with the corresponding slave nodes through identifiers such as their addresses and registered names.

[0079] As mentioned earlier, slave nodes can also be equipped with privacy-preserving computing devices. Their execution modules can monitor the master node's control module in real time. Upon detecting that the master node has created subtasks related to itself, the slave node can acquire these subtasks and complete their business processing through its execution module. Furthermore, if data from other slave nodes is needed during subtask processing, data can be exchanged through data communication with them, thus protecting privacy. This includes methods such as encrypted data exchange during homomorphic computing.

[0080] The slave nodes of the first node can interact with the second node by executing relevant sub-tasks, thereby jointly completing the first privacy computing task with the second node.

[0081] As can be seen from the above, the first node can work in both master node mode and peer mode at the same time, and the activated modules include at least the synchronization controller and the control module.

[0082] Furthermore, this specification also provides a privacy-preserving computation method that can be applied to any computation node in privacy-preserving computation, for example, denoted as a third node. This third node can be any computer, device, or server with a certain computing capability, such as... Figure 5 The Bob node in the code. The third node can be configured as follows: Figure 2 The privacy computing device shown can be configured with other computing nodes (such as...) Figure 5 Node 3 (referred to as the fourth node) is the trusted master node, which can then be responsible for the task control of the third node in master node mode, while the third node operates in slave node mode. Simultaneously, the third node can also perform privacy-preserving computation tasks with the fifth node in a peer-to-peer mode.

[0083] Figure 8 This is a schematic diagram of a privacy computation process according to one embodiment. Figure 8 As shown, the privacy computation process applicable to the third node above may include:

[0084] Step 802: Monitor the fourth and fifth nodes through the synchronization controller;

[0085] Step 804: If a second privacy computing task related to the third node is detected, the second task parameters corresponding to the second privacy computing task are synchronized to the local control module.

[0086] Step 806: Based on the execution module's monitoring of the local control module, perform relevant privacy calculations in the second privacy calculation task using the second task parameters.

[0087] It is worth noting that the source of the second privacy computation task is different, and the way the third node, which is working in slave mode and peer mode, handles the second privacy computation task may be different.

[0088] On the one hand, the second privacy-preserving computation task detected by the third node originates from the fifth node (such as...). Figure 5 In the case of the Alice node, the third node can execute the second privacy computation task through the local execution module, and during the execution of the second privacy computation task, the third node can communicate with the fifth node with data under the premise of privacy protection.

[0089] On the other hand, the second privacy computing task detected by the third node comes from the fourth node (such as...). Figure 5 In the case of node 3), the third node can independently perform relevant computations through its local execution module to complete the second privacy computation task, or it can collaborate with the fifth node to complete the second privacy computation task. If the second privacy computation task can be completed using only the data from the third node, the execution module of the third node can obtain relevant data for the second privacy computation task, such as relevant image files, by listening to the control module, and then execute the second privacy computation task by launching it as a container. This second privacy computation task may require collaboration with other nodes (such as the fourth node, etc.). Figure 5 The second privacy computation task is jointly completed by the Alice node (in the main node). The third node can break down the second privacy computation task into multiple subtasks through the control module. The third node listens to the control module through the execution module, which can then process subtasks related to local data. Simultaneously, since the third and fourth nodes mutually listen to each other's control modules, the fourth node, upon sensing subtasks related to itself, can complete the relevant subtask computations through its local execution module. During the execution of relevant subtasks, the third and fourth nodes can interact with intermediate computation results or the final subtask execution results based on privacy protection. In the joint execution of the second privacy computation task, the third node can collaborate with other slave nodes corresponding to the main node (such as...). Figure 5 The Joke node in the system conducts data communication under the premise of privacy protection.

[0090] Understandable. Figure 7 , Figure 8 The illustrated processes involve a single computing node acting as either a master or slave node in a centralized privacy computing environment, while simultaneously configuring a peer-to-peer mode within a decentralized privacy computing framework. The centralized and decentralized architectures work together to achieve privacy-preserving computation. Alternatively, a centralized privacy computing master node and slave node can be configured to operate in peer-to-peer mode within a decentralized joint secure computing framework, or a computing node operating in peer-to-peer mode within a decentralized privacy framework can simultaneously act as a master node in a centralized privacy computing framework (e.g., ...). Figure 5 The Alice node in the data is also... Figure 6 In cases such as node 3), various working modes are nested together to complete the processing of relatively complex multi-party joint privacy computing business.

[0091] In other words, a single computing node can be configured as a trusted third party or service provider through other computing nodes, thus acting as the master node for these other computing nodes and becoming the controller for scheduling and distributing privacy tasks. On the other hand, a single computing node can also act as a trusted third party or service provider with a master node, collaborating with other ordinary nodes to complete privacy-preserving computation tasks. Furthermore, it can also collaborate with other nodes in a decentralized, collaborative, and secure computation manner to complete corresponding privacy-preserving computation tasks. In this way, interconnection between nodes can be achieved through multiple methods, enabling multi-level expansion and forming a large privacy-preserving computation network.

[0092] It is worth noting that, Figure 7 , Figure 8 This is not necessarily an exhaustive list of the actual workflows of each computing node in this specification. In other embodiments, computing nodes configured with different modes may have other privacy-preserving computing processes, which will not be elaborated here.

[0093] According to another embodiment, a computer-readable storage medium is also provided, on which a computer program is stored, which, when executed in a computer, causes the computer to perform a combination Figure 7 , Figure 8 The methods described above.

[0094] Those skilled in the art will recognize that the functions described in the embodiments of this specification in one or more of the above examples can be implemented using hardware, software, firmware, or any combination thereof. When implemented in software, these functions can be stored in a computer-readable medium or transmitted as one or more instructions or code on a computer-readable medium.

[0095] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the technical concept in this specification. It should be understood that the above description is only a specific embodiment of the technical concept in this specification and is not intended to limit the scope of protection of the technical concept in this specification. Any modifications, equivalent substitutions, improvements, etc., made on the basis of the technical solutions of the embodiments in this specification should be included within the scope of protection of the technical concept in this specification.

Claims

1. A privacy computing device, comprising a control module, a synchronization controller, and an execution module, wherein: The control module is used for creating, scheduling, and distributing privacy computing tasks; The synchronization controller is used to monitor the control modules of other computing nodes in order to synchronize the resources and parameters of privacy computing tasks created by the control modules of other computing nodes that are related to the current computing node setting of the device to the local machine. The execution module is used to execute privacy computing tasks related to the current computing node. The privacy computing tasks are obtained by monitoring the local control module or the control modules of other computing nodes. When different modules are activated, the current computing node can work simultaneously in two of the following modes: master node mode, slave node mode, and peer mode.

2. The privacy computing device as claimed in claim 1, wherein, Master node mode and slave node mode are suitable for centralized federated computing architectures. When a single computing node is configured as the central node, it works in master node mode, and when other computing nodes are configured as central nodes, it works in slave node mode. Peer mode is suitable for decentralized federated computing architectures.

3. In the privacy computing device as described in claim 2, in master node mode, the device creates, schedules, and distributes privacy computing tasks through a control module.

4. The privacy computing device as described in claim 2, in slave node mode, the device monitors the control module of the master node through the execution module, so as to execute the privacy computing task via the execution module when the master node creates a privacy computing task related to the current computing node through the control module.

5. The privacy computing device as described in claim 2, in peer-to-peer mode, the device monitors the control module of other nodes through the synchronization controller, so that when it detects that other nodes have created a privacy computing task related to the current computing node, it synchronizes the resource and parameter information of the privacy computing task to the current computing node, and executes the privacy computing task through the execution module.

6. A privacy computing method executed by a first node, wherein the first node is equipped with the privacy computing device as described in claim 1, and forms a decentralized privacy computing architecture with a second node, and also forms a centralized privacy computing architecture with at least one other computing node as a master node and as a slave node, the privacy computing method comprising: The synchronization controller monitors the second node so that, if the second node creates a first privacy computing task related to the first node, the first task parameters corresponding to the first privacy computing task are synchronized locally. The control module creates multiple subtasks for the first privacy computing task, which are then distributed downwards. Each slave node, based on the processing of the corresponding subtask, works in conjunction with the second node to complete the first privacy computing task.

7. The method of claim 6, wherein, The first task parameters include at least one of the following: the start command for the first privacy computing task, the input file address or path, and the output file path.

8. The method of claim 6, wherein, Each slave node communicates with the second node while processing its respective subtask.

9. A privacy computing method executed by a third node, wherein the third node is equipped with the privacy computing device as described in claim 1, and forms a centralized privacy computing architecture with a fourth node as a master node, and simultaneously forms a decentralized privacy computing architecture with a fifth node, the privacy computing method comprising: The fourth and fifth nodes are monitored through the synchronization controller; Upon detecting a second privacy computing task related to the third node, the second task parameters corresponding to the second privacy computing task are synchronized to the local control module. Based on the execution module's monitoring of the local control module, the relevant privacy calculations in the second privacy calculation task are executed using the second task parameters.

10. The method of claim 9, wherein when the second privacy computation task originates from the fifth node, the step of performing relevant privacy computations in the second privacy computation task based on the execution module's monitoring of the local control module and utilizing the second task parameters includes: The second privacy computation task is executed through a local execution module. During the execution of the second privacy computation task, the third node and the fifth node communicate data under the premise of privacy protection.

11. The method of claim 9, wherein when the second privacy computation task originates from the fourth node, the method further comprises: Subtasks of the second privacy computing task are created through the local control module, with each subtask associated with a third or fifth node; The step of performing relevant privacy calculations in the second privacy calculation task based on the execution module's monitoring of the local control module and utilizing the second task parameters includes: The module listens for and processes subtasks associated with the third node.

12. The method as described in claim 11, wherein the subtask associated with the fifth node is synchronized locally by the fifth node's monitoring of the control module and executed; in, The third node communicates with the fifth and fourth nodes at the data layer.

13. A computer-readable storage medium having a computer program stored thereon, which, when executed in a computer, causes the computer to perform the method of any one of claims 6-12.

Images