Decentralized federated learning method based on identity privacy protection and economic incentive

Abstract

The invention relates to a decentralized federated learning method based on identity privacy protection and economic incentive, and the method comprises the steps: publishing a target federated learning task based on a preset decentralized network and a target task publisher, and determining an economic incentive mechanism and privacy protection requirements of the target federated learning task, determining a data owner receiving the target federal learning task; the data owner downloads the global model, the economic incentive mechanism and the privacy protection requirement, and pays registration fees to obtain a training voucher; and training a global model by using a local training data set, evaluating the trained model to update the global model, iteratively executing evaluation and updating operations to obtain a final federated learning model, and meanwhile, based on a mixed currency server, enabling a target task publisher to pay award fees to a data owner. Therefore, the problems that in the prior art, data privacy leakage is easily caused, effective economic incentive is lacked, and data owners are severely hindered from participating in federated learning are solved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a decentralized federated learning method based on identity privacy protection and economic incentives. Background Art

[0002] In order to solve problems such as data privacy and transmission delay, the emerging federated learning method has emerged. Federated learning allows multiple data owners to collaborate to train a shared machine learning model without having to store or exchange their respective data in a centralized manner, thereby protecting data privacy, reducing data transmission delays, and saving communication resources. However, in practice, data owners are often reluctant to share their data due to privacy concerns and lack of incentives.

[0003] First, in terms of privacy, data owners are generally concerned about the leakage of personal privacy information. Although federated learning is designed to protect the privacy of original data, there may still be a risk of sensitive identity information leakage during the transmission of model parameters. Attackers may identify specific individuals by associating model parameters with real identities by intercepting information in the communication channel. In order to alleviate this identity privacy issue, some research works on federated learning have adopted anonymity protection mechanisms, such as destroying the connection between updates and their generators through anonymous communication to ensure that even if the original data is reconstructed, attackers cannot infer the privacy of specific users; when the privately held data of the client in the learning task does not contain personal identity information (such as pictures, names, or passport names), breaking the link between the client and their updates is sufficient to protect the privacy of the client. However, anonymity in federated learning also exacerbates identity authentication and may lead to Sybil attacks, that is, attackers manipulate the system by creating a large number of fake identities.

[0004] Secondly, since the training process consumes a lot of resources, such as storage and communication bandwidth, users are usually unwilling to participate for free. In the absence of any rewards, users may be unwilling to sacrifice their mobile device resources to assist in federated learning. Therefore, it is particularly important to design an effective incentive mechanism to encourage data owners to participate in data sharing. Existing incentive mechanisms include the use of reputation systems, but they are not enough to motivate users, especially when the specific benefits of high reputation are unclear. Traditional economic incentives use micropayment systems, but it is difficult to deploy trusted third parties in decentralized networks.

[0005] As a distributed method, blockchain technology provides an effective means of economic incentives, but using blockchain for transactions may reveal the identity of the data owner. Privacy-preserving federated learning based on blockchain cannot fully protect the identity privacy of the data owner, especially when the real account address is required to receive the reward. Even when using a decentralized anonymous payment solution such as Zcash, the transaction data and address will expose both parties of the transaction, namely the task publisher and the data owner. In addition, the introduction of incentive mechanisms, especially when using public chains such as Bitcoin and Ethereum, greatly increases communication and transaction costs.

[0006] In summary, existing technologies are prone to data privacy leakage and lack effective economic incentives, which seriously hinders data owners from participating in federated learning and needs to be urgently resolved. Summary of the invention

[0007] This application provides a decentralized federated learning method based on identity privacy protection and economic incentives to solve the problems that existing technologies are prone to data privacy leakage and lack effective economic incentives, which seriously hinders data owners from participating in federated learning.

[0008] The first aspect of the present application provides a decentralized federated learning method based on identity privacy protection and economic incentives, comprising the following steps: based on a preset decentralized network and a target task publisher, publishing a target federated learning task, and determining the economic incentive mechanism and privacy protection requirements of the target federated learning task, so as to determine the data owner who receives the target federated learning task according to the economic incentive mechanism and the privacy protection requirements; downloading the global model, the economic incentive mechanism and the privacy protection requirements corresponding to the target federated learning task through the data owner using a preset anonymous communication channel, and paying a preset registration fee to obtain the training certificate corresponding to the global model; training the global model using a preset local training data set to obtain an initial federated learning model, and evaluating the initial federated learning model to update the global model through the evaluation results and the local training data set, and iteratively performing evaluation and update operations according to the privacy protection requirements to obtain a final federated learning model, and at the same time, based on the economic incentive mechanism and the preset coin mixing server, the target task publisher pays a preset reward fee to the data owner.

[0009] Optionally, in one embodiment of the present application, based on the preset decentralized network and the target task publisher, the target federated learning task is published, and the economic incentive mechanism and privacy protection requirements of the target federated learning task are determined, so as to determine the data owner who receives the target federated learning task according to the economic incentive mechanism and the privacy protection requirements, including: determining the privacy protection requirements and task payment budget of the target federated learning task through the target task publisher; constructing a first transaction channel between the target task publisher and the preset coin mixing server, and allowing the target task publisher to pledge a first one-way pledge fund in the first transaction channel; publishing the target federated learning task on the preset decentralized network through the target task publisher, and broadcasting a first one-way pledge transaction corresponding to the first one-way pledge fund, so that the data owner who receives the first one-way pledge transaction analyzes the target federated learning task according to the first one-way pledge fund and the task payment budget, and generates a corresponding analysis result, so as to accept the target federated learning task if the analysis result meets the preset reception requirements.

[0010] Optionally, in one embodiment of the present application, the data owner downloads the global model, the economic incentive mechanism and the privacy protection requirements corresponding to the target federated learning task through a preset anonymous communication channel, and pays a preset registration fee to obtain the training certificate corresponding to the global model, including: establishing a second one-way pledge transaction from the data owner to the preset coin mixing server, and establishing a third one-way pledge transaction from the preset coin mixing server to the target task publisher through the preset coin mixing server according to a preset registration pledge anonymous authentication strategy, so as to construct a second transaction channel according to the second one-way pledge transaction and the third one-way pledge transaction; converting the transaction state of the third one-way pledge transaction into a first hidden state commitment, so that the target task publisher verifies whether the first hidden state commitment satisfies a preset correctness condition; if the first hidden state commitment satisfies the preset correctness condition, verifying whether the first hidden state commitment satisfies a preset legality condition through the data owner; if the first hidden state commitment satisfies the preset legality condition, randomizing the first hidden state commitment, A randomized first hidden state commitment is obtained, and a randomized second hidden state commitment and the preset registration fee are generated by using the preset coin mixing server according to a preset verification and update strategy; a randomized third hidden state commitment is generated based on the randomized second hidden state commitment and the preset randomization processing strategy, and at the same time, the target task publisher verifies whether the randomized third hidden state commitment meets the preset correctness condition according to the preset training token, and generates a first updated hidden state of the second transaction channel corresponding to the randomized third hidden state commitment when the randomized third hidden state commitment meets the preset correctness condition; the randomized third hidden state commitment is blindly signed by the target task publisher, and the global model and the blindly signed randomized third hidden state commitment are sent to the data owner, so that the data owner generates a training voucher corresponding to the training token through the preset de-blinding processing strategy; the first updated hidden state is submitted through the preset coin mixing server and the target task publisher, and a first target payment process is initiated, so that the data owner pays the preset registration fee to the preset coin mixing server.

[0011] Optionally, in one embodiment of the present application, the economic incentive mechanism and the preset coin mixing server are used to enable the target task publisher to pay the preset reward fee to the data owner, including: establishing a fourth one-way pledge transaction between the preset coin mixing server and the data publisher according to the preset reward pledge anonymous authentication mechanism; based on the preset coin mixing server, converting the transaction state corresponding to the fourth one-way pledge transaction into a fourth hidden state commitment, and verifying and randomizing the fourth hidden state commitment by the data owner to obtain a randomized fourth hidden state commitment corresponding to the fourth hidden state commitment; sending the randomized fourth hidden state commitment to the preset coin mixing server so that the preset coin mixing server The device generates a second updated hidden state and the preset reward fee corresponding to the randomized fourth hidden state commitment according to the randomized fourth hidden state commitment and the preset verification update strategy; verifies whether the second updated hidden state satisfies the preset legitimacy condition through the target task publisher, and if the second updated hidden state satisfies the preset legitimacy condition, randomizes the second updated hidden state through the data owner to generate a third updated hidden state corresponding to the second updated hidden state; submits the third updated hidden state through the preset coin mixing server and the data owner, and starts a second target payment process, so that the target task publisher pays the preset reward fee to the data owner.

[0012] The second aspect of the present application provides a decentralized federated learning device based on identity privacy protection and economic incentives, including: a task publishing module, which is used to publish a target federated learning task based on a preset decentralized network and a target task publisher, and determine the economic incentive mechanism and privacy protection requirements of the target federated learning task, so as to determine the data owner who receives the target federated learning task according to the economic incentive mechanism and the privacy protection requirements; a download module, which is used to download the global model, the economic incentive mechanism and the privacy protection requirements corresponding to the target federated learning task through the data owner using a preset anonymous communication channel, and pay a preset registration fee to obtain a training certificate corresponding to the global model; a reward module, which is used to train the global model using a preset local training data set to obtain an initial federated learning model, and evaluate the initial federated learning model to update the global model through the evaluation results and the local training data set, and iteratively perform evaluation and update operations according to the privacy protection requirements to obtain a final federated learning model, and at the same time, based on the economic incentive mechanism and the preset coin mixing server, the target task publisher pays a preset reward fee to the data owner.

[0013] Optionally, in one embodiment of the present application, the task publishing module includes: a determination unit, used to determine the privacy protection requirements and task payment budget of the target federated learning task through the target task publisher; a construction unit, used to construct a first transaction channel between the target task publisher and the preset coin mixing server, and enable the target task publisher to pledge a first one-way pledge fund in the first transaction channel; an analysis unit, used to publish the target federated learning task through the target task publisher on the preset decentralized network, and broadcast a first one-way pledge transaction corresponding to the first one-way pledge fund, so that the data owner who receives the first one-way pledge transaction analyzes the target federated learning task according to the first one-way pledge fund and the task payment budget, and generates a corresponding analysis result, so as to accept the target federated learning task if the analysis result meets the preset reception requirements.

[0014] Optionally, in one embodiment of the present application, the download module includes: a first establishment unit, used to establish a second one-way pledge transaction from the data owner to the preset coin mixing server, and establish a third one-way pledge transaction from the preset coin mixing server to the target task publisher through the preset coin mixing server according to a preset registration pledge anonymous authentication strategy, so as to build a second transaction channel according to the second one-way pledge transaction and the third one-way pledge transaction; a first conversion unit, used to convert the transaction state of the third one-way pledge transaction into a first hidden state commitment, so that the target task publisher verifies whether the first hidden state commitment satisfies a preset correctness condition; a first verification unit, used to verify whether the first hidden state commitment satisfies a preset legality condition through the data owner when the first hidden state commitment satisfies the preset legality condition; an update unit, used to randomize the first hidden state commitment when the first hidden state commitment satisfies the preset legality condition, obtain a randomized first hidden state commitment, and use the preset coin mixing server to verify and update the first hidden state commitment according to the preset verification and update conditions. The strategy generates a randomized second hidden state commitment and the preset registration fee; a first randomization unit, which is used to generate a randomized third hidden state commitment based on the randomized second hidden state commitment and the preset randomization processing strategy, and at the same time enables the target task publisher to verify whether the randomized third hidden state commitment meets the preset correctness condition according to the preset training token, and generate the first updated hidden state of the second transaction channel corresponding to the randomized third hidden state commitment when the randomized third hidden state commitment meets the preset correctness condition; a blind signature unit, which is used to blindly sign the randomized third hidden state commitment through the target task publisher, and send the global model and the blindly signed randomized third hidden state commitment to the data owner, so that the data owner generates a training voucher corresponding to the training token through the preset de-blinding processing strategy; a first submission unit, which is used to submit the first updated hidden state through the preset mixing server and the target task publisher, and start a first target payment process, so that the data owner pays the preset registration fee to the preset mixing server.

[0015] Optionally, in one embodiment of the present application, the reward module includes: a second establishment unit, used to establish a fourth one-way pledge transaction between the preset coin mixing server and the data publisher according to a preset reward pledge anonymous authentication mechanism; a second conversion unit, used to convert the transaction state corresponding to the fourth one-way pledge transaction into a fourth hidden state commitment based on the preset coin mixing server, and verify and randomize the fourth hidden state commitment through the data owner to obtain a randomized fourth hidden state commitment corresponding to the fourth hidden state commitment; a second verification unit, used to send the randomized fourth hidden state commitment to the preset coin mixing server, so that the preset coin mixing server verifies and randomizes the fourth hidden state commitment according to the randomized fourth hidden state commitment and A preset verification update strategy generates a second updated hidden state and the preset reward fee corresponding to the randomized fourth hidden state commitment; a second randomization unit is used to verify whether the second updated hidden state meets the preset legitimacy condition through the target task publisher, and if the second updated hidden state meets the preset legitimacy condition, the second updated hidden state is randomized by the data owner to generate a third updated hidden state corresponding to the second updated hidden state; a second submission unit is used to submit the third updated hidden state through the preset coin mixing server and the data owner, and start a second target payment process so that the target task publisher pays the preset reward fee to the data owner.

[0016] The third aspect of the present application provides an electronic device, comprising: a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the program to implement the decentralized federated learning method based on identity privacy protection and economic incentives as described in the above embodiments.

[0017] The fourth aspect embodiment of the present application provides a computer-readable storage medium, which stores a computer program, which, when executed by a processor, implements the above-mentioned decentralized federated learning method based on identity privacy protection and economic incentives.

[0018] Therefore, the embodiments of the present application have the following beneficial effects:

[0019] The embodiment of the present application can publish the target federated learning task based on the preset decentralized network and the target task publisher, and determine the economic incentive mechanism and privacy protection requirements of the target federated learning task, so as to determine the data owner who receives the target federated learning task according to the economic incentive mechanism and privacy protection requirements; the data owner downloads the global model, economic incentive mechanism and privacy protection requirements corresponding to the target federated learning task through the preset anonymous communication channel, and pays the preset registration fee to obtain the training certificate corresponding to the global model; the global model is trained using the preset local training data set to obtain the initial federated learning model, and the initial federated learning model is evaluated to update the global model through the evaluation results and the local training data set, and the evaluation and update operations are iteratively performed according to the privacy protection requirements to obtain the final federated learning model, and at the same time, based on the economic incentive mechanism and the preset mixing server, the target task publisher pays the preset reward fee to the data owner. This application introduces the registration fee and the anonymous authentication protocol of the pledge stage to resist the possible malicious attacks such as the witch attack in the incentive process, thereby significantly improving the participation, security and efficiency of the federated learning system. Thus, the existing technology is easy to cause data privacy leakage, and lacks effective economic incentives, which seriously hinders the data owner from participating in federated learning.

[0020] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through the practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0022] Figure 1 A logical architecture diagram of a decentralized federated learning method based on identity privacy protection and economic incentives provided for this application;

[0023] Figure 2 A flowchart of a decentralized federated learning method based on identity privacy protection and economic incentives provided according to an embodiment of the present application;

[0024] Figure 3 A schematic diagram of a task release phase execution flow provided for an embodiment of the present application;

[0025] Figure 4 A schematic diagram of a model downloading phase execution flow provided for an embodiment of the present application;

[0026] Figure 5 A schematic diagram of the execution flow of anonymous authentication of registration pledge in the model download phase provided for one embodiment of the present application;

[0027] Figure 6 A schematic diagram of a model upload phase execution flow provided for an embodiment of the present application;

[0028] Figure 7 A schematic diagram of the execution flow of anonymous authentication of reward pledge in a model upload phase provided for an embodiment of the present application;

[0029] Figure 8 This is an example diagram of a decentralized federated learning device based on identity privacy protection and economic incentives according to an embodiment of the present application;

[0030] Fig. 9 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application.

[0031] Among them, 10-decentralized federated learning device based on identity privacy protection and economic incentives; 100-task publishing module, 200-download module, 300-reward module; 901-memory, 902-processor, 903-communication interface. DETAILED DESCRIPTION

[0032] Embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application.

[0033] The following describes the decentralized federated learning method based on identity privacy protection and economic incentives of the embodiment of the present application with reference to the accompanying drawings. In response to the problems mentioned in the above background technology, the present application provides a decentralized federated learning method based on identity privacy protection and economic incentives, in which the target federated learning task is published based on a preset decentralized network and a target task publisher, and the economic incentive mechanism and privacy protection requirements of the target federated learning task are determined to determine the data owner receiving the target federated learning task according to the economic incentive mechanism and privacy protection requirements; the data owner downloads the global model, economic incentive mechanism and privacy protection requirements corresponding to the target federated learning task through a preset anonymous communication channel, and pays a preset registration fee to obtain the training certificate corresponding to the global model; the global model is trained using a preset local training data set to obtain an initial federated learning model, and the initial federated learning model is evaluated to update the global model through the evaluation results and the local training data set, and the evaluation and update operations are iteratively performed according to the privacy protection requirements to obtain the final federated learning model, and at the same time, based on the economic incentive mechanism and the preset mixing currency server, the target task publisher pays the preset reward fee to the data owner. This application introduces registration fees and anonymous authentication protocols in the pledge phase to resist possible malicious attacks such as Sybil attacks during the incentive process, thereby significantly improving the participation, security and efficiency of the federated learning system. This solves the problems that the existing technology is prone to data privacy leakage and lacks effective economic incentives, which seriously hinders data owners from participating in federated learning.

[0034] In order to facilitate those skilled in the art to understand the decentralized federated learning method based on identity privacy protection and economic incentives of this application, the entity roles, related algorithms and payment channels designed in this application are explained as follows:

[0035] 1. Entity Role:

[0036] This application conducts decentralized federated learning based on identity privacy protection and economic incentives through two stages: task publishing stage and federated learning task execution stage. The federated learning stage can also be subdivided into model download stage and model upload stage. Figure 1 As shown in the figure, the above task publishing stage, model downloading stage and model uploading stage mainly involve the following three entity roles in the actual execution process:

[0037] 1. Task publisher: In this application, the task publisher can be any node in the P2P scenario, responsible for publishing training tasks, providing initial models and their subsequent aggregate updates. This node is usually the demand side, such as an enterprise or research institution, which is motivated to obtain high-quality models, so it can be regarded as a semi-honest node and may collude with other participants. Because it needs to be recognized and trusted by other participants, the identity of the task publisher is usually public, so as to effectively coordinate and manage the entire training process;

[0038] 2. Data owner (or data owner): The data owner in this application is a participant in federated learning who provides data to train shared models while maintaining their privacy. To ensure privacy protection, data owners use temporary identities or anonymous credentials during the training process to prevent the disclosure of their real identities and ensure that data updates generated in multiple consecutive rounds of training are unlinkable. Although most data owners participate in federated learning in order to obtain incentives to be honest, there is also the risk of malicious behavior such as data poisoning and witch attacks, which may undermine the accuracy and trust of the model.

[0039] 3. Coin mixing server (i.e. coin mixing service provider): In this application, the coin mixing server is responsible for performing the coin mixing operation of blockchain tokens. It is a privacy enhancement technology that protects the user's financial privacy by hiding the source and destination of funds by confusing the transaction chain. In this application, the coin mixing server acts as a key intermediary. Although it technically provides the necessary anonymity, it is unreliable in itself. The coin mixing server may violate the agreement due to profit-driven reasons, or conspire with other entities to leak information. Therefore, additional security measures need to be taken in the design to ensure the transparency and verifiability of its behavior, thereby maintaining the security and trust of the entire system.

[0040] Through the collaboration and interaction of the above three entity roles, this application can effectively promote the process of federated learning while ensuring the privacy of data owners, and encourage the participation of more data owners through economic incentive mechanisms.

[0041] 2. Related algorithms:

[0042] 1. Commitment Scheme (CS):

[0043] Commitment Algorithm (COM): This algorithm takes a message m as input and outputs a commitment value C and a key d associated with the commitment; it can be formally described as: COM(m) → (C, d)

[0044] Verification Algorithm (VCOM): This algorithm accepts a commitment value C, a key d associated with the commitment, and information m as input, and outputs a Boolean value indicating whether the commitment is valid; it can be formally described as: VCOM(C, d, m) → {0, 1}

[0045] In this commitment scheme, algorithm COM is responsible for generating commitments, while VCOM is responsible for verifying whether the provided information m matches the commitment value C; for example, the Pedersen commitment scheme not only ensures the hiding of information (no one can know the information m from C), but also guarantees the binding of the commitment computationally (under the premise of limited computing power, once the commitment is generated, the information m cannot be changed).

[0046] 2. Blinded Randomizable Signature (BRS):

[0047] Key generation (KeyGen): Given a security parameter λ, output a key pair (sk, pk);

[0048] BlindSig: For a message commitment value C, the signer uses the blind signature algorithm to generate a blind signature It can be formally described as:

[0049] UnblindSig: The party with the unblinding information d can use the blind signature is transformed into a valid signature s, which can be formally described as:

[0050] Randomized signature (BRandSig): For a valid signature s, anyone can generate a randomized signature s′, which can be formally described as BRandSig(s)→s′;

[0051] This application may adopt the Pointcheval-Sanders signature scheme, which is based on a Type-III bilinear group and allows blind signatures on commitments.

[0052] 3. Randomizable Signature of UpdatableCommitments (RSUC):

[0053] Parameter Setup: Given a security parameter λ, output the system parameter pp, which can be formally described as: Setup(1 λ )→pp;

[0054] Key generation (KeyGen): Given a system parameter pp, output an authentication-verification key pair (sk, pk), which can be formally described as: KeyGen(pp) → (sk, pk);

[0055] Authentication Commitment (AuthCom): Given an authentication key sk and a value m, output a commitment C and an authentication s on the commitment, which can be formally described as: AuthCom(sk, m) → (C, s);

[0056] ComVerify: Given a commitment C and a value m, verify whether the commitment is valid, which can be formally described as: ComVerify(C, m) → {0, 1};

[0057] Verify authentication commitment (AuthVerify): Given an authentication commitment (C, s) and a verification key pk, verify whether the authentication is valid, which can be formally described as: AuthVerify(C, s, pk) → {0, 1};

[0058] Randomized authentication commitment (RandSig): Given an authentication commitment (C, s), output the randomized result (C′, s′), which can be formally described as: RandSig(C, s) → (C′, s′);

[0059] Update Authentication Commitment (UpSig): Given an authentication commitment (C, s), an update value a, and an authentication key sk, output an updated authentication commitment (C new ,s new ), which can be formally described as: UpSig(C, a, sk)→(C new ,s new );

[0060] AuthUpVerify: Given the original authentication commitment (C, s), the updated value a, and the new authentication commitment (C new ,s new ) and the verification key pk to verify whether the update is correct, which can be formally described as: AuthUpVerify(C, a, C new ,s new ,pk)→0,1.

[0061] 3. Payment Channel (PC):

[0062] In this application, a payment channel is opened by two users (such as Alice and Bob), and on-chain funds are deposited into it, and it is updated as payments are made. The life cycle of a payment channel includes three processes: opening, updating, and closing.

[0063] The opening process is: two users (such as Alice and Bob) open the channel and pledge a sum of money on the chain. These deposits serve as their initial channel balance and will be updated as payments proceed;

[0064] Update process: Users complete payments by updating the state of the payment channel, which consists of the channel balance allocation (i.e., the balance allocation between the two channel users) and the channel identifier (such as the address of the channel contract, which uniquely identifies the channel and remains unchanged throughout the process). When making a payment, a user (such as Alice) generates a new channel state, in which the balance allocation is updated according to the payment; Alice then authenticates the new state and sends it to Bob along with the authentication information; Bob needs to perform correctness checks on the following items:

[0065] (1) Lateness identification: ensuring that the latest channel status is received;

[0066] (2) Balance-sufficiency attestation, which ensures that Alice has sufficient funds to make the payment;

[0067] (3) Authentication-validity verification: whether Alice’s authentication of the new status is valid.

[0068] If all checks pass, Bob can respond with the new state using his authentication; if Alice does not receive a response from Bob, she cannot be sure that the update will take effect; Alice can then submit the new state to the blockchain for acceptance;

[0069] Closing process: Both users of the payment channel can initiate channel closure to refund their balances in the last channel state accepted by the blockchain; the payment channel closure is based on the latest channel state provided by the channel users, which is guaranteed by the consensus of the channel users, and the user's withdrawal amount will not be less than its latest balance in payment.

[0070] A payment channel is a private transaction between two users that allows users to pay each other without the need for blockchain confirmation. Payment information is only posted to the blockchain when the channel is closed or proof of payment is required, which can effectively reduce the number of transactions on the blockchain, increase transaction speed, and reduce transaction fees.

[0071] Specifically, Figure 1A flowchart of a decentralized federated learning method based on identity privacy protection and economic incentives provided in an embodiment of the present application.

[0072] like Figure 2 As shown in the figure, the decentralized federated learning method based on identity privacy protection and economic incentives includes the following steps:

[0073] In step S201, based on the preset decentralized network and the target task publisher, the target federated learning task is published, and the economic incentive mechanism and privacy protection requirements of the target federated learning task are determined to determine the data owner who receives the target federated learning task according to the economic incentive mechanism and privacy protection requirements.

[0074] During the task publishing stage, the embodiments of the present application can publish a federated learning task including an initial model, task description, payment budget, and data requirements on a decentralized network through a task publisher. This stage also includes setting up the task's incentive mechanism (i.e., economic incentive mechanism) and privacy protection requirements, thereby ensuring that all potential data owners fully understand the task requirements and participation conditions.

[0075] Optionally, in one embodiment of the present application, based on a preset decentralized network and a target task publisher, a target federated learning task is published, and the economic incentive mechanism and privacy protection requirements of the target federated learning task are determined, so as to determine the data owner who receives the target federated learning task according to the economic incentive mechanism and the privacy protection requirements, including: determining the privacy protection requirements and the task payment budget of the target federated learning task through the target task publisher; constructing a first transaction channel between the target task publisher and a preset coin mixing server, and allowing the target task publisher to pledge a first one-way pledge fund in the first transaction channel; publishing the target federated learning task on a preset decentralized network through the target task publisher, and broadcasting a first one-way pledge transaction corresponding to the first one-way pledge fund, so that the data owner who receives the first one-way pledge transaction analyzes the target federated learning task according to the first one-way pledge fund and the task payment budget, and generates corresponding analysis results, so as to accept the target federated learning task if the analysis results meet the preset reception requirements.

[0076] It should be noted that in the embodiment of the present application, during the task publishing stage, the task publisher can publish the federated learning task on the decentralized network, and the data owner decides whether to accept or participate in the federated learning task. Figure 3 As shown, the specific workflow of the embodiment of the present application in the task release stage is as follows:

[0077] (1) Task requirements and pre-staking: The task publisher determines the specific requirements of the federated learning task, including data type, data quality, model specifications, and training objectives. At the same time, the task publisher also sets a payment budget for completing the task. In order to initially establish trust in economic incentives, the task publisher establishes a transaction channel (i.e., the first transaction channel) with the currency mixing service provider and pledges a one-way pledge fund (i.e., the first one-way pledge fund) from the task publisher to the currency mixing server in the transaction channel.

[0078] (2) Task broadcasting and decision-making participation: When the task publisher publishes the federated learning task Task, it broadcasts the pledge transaction id Tid corresponding to the first one-way pledge fund. TI (i.e., the first one-way pledge transaction) for verification by the mixing server and the data owner, and the data owner can evaluate the payment budget corresponding to the first one-way pledge transaction (i.e., the task payment budget), generate the corresponding analysis results or evaluation results, and decide whether to participate in the federated learning training. For example, the data owner in the embodiment of the present application can accept the above-mentioned federated learning task when the analysis results or evaluation results meet the preset reception requirements.

[0079] Therefore, the embodiments of the present application release federated learning tasks through corresponding operations in the task release phase, and set corresponding economic incentive mechanisms and privacy protection requirements, thereby providing a reliable basis and guidance for the execution of subsequent model download phase and model upload phase.

[0080] In step S202, the data owner uses a preset anonymous communication channel to download the global model, economic incentive mechanism, and privacy protection requirements corresponding to the target federated learning task, and pays a preset registration fee to obtain the training certificate corresponding to the global model.

[0081] Furthermore, in order to maintain identity anonymity and security, during the model download stage, the embodiments of the present application may also enable the data owner who receives the above-mentioned federated learning task to download the global model, economic incentive mechanism and privacy protection requirements corresponding to the target federated learning task through a preset anonymous communication channel such as a one-time address or anonymous credential; at the same time, during the model download stage, the data owner must also pay a preset registration fee to obtain the training credential corresponding to the global model.

[0082] Therefore, the embodiments of the present application can not only ensure the effective participation of data owners in the training process, but also effectively prevent witch attacks by paying registration fees, and ensure that only data owner nodes that are truly willing to participate can join the decentralized network.

[0083] Optionally, in one embodiment of the present application, the data owner uses a preset anonymous communication channel to download the global model, economic incentive mechanism and privacy protection requirements corresponding to the target federated learning task, and pays a preset registration fee to obtain the training certificate corresponding to the global model, including: establishing a second one-way pledge transaction from the data owner to the preset coin mixing server, and establishing a third one-way pledge transaction from the preset coin mixing server to the target task publisher through the preset coin mixing server according to the preset registration pledge anonymous authentication strategy, so as to build a second transaction channel according to the second one-way pledge transaction and the third one-way pledge transaction; converting the transaction state of the third one-way pledge transaction into a first hidden state commitment, so that the target task publisher verifies whether the first hidden state commitment meets the preset correctness condition; when the first hidden state commitment meets the preset correctness condition, verifying whether the first hidden state commitment meets the preset legality condition through the data owner; when the first hidden state commitment meets the preset legality condition, randomizing the first hidden state commitment to obtain a random The method comprises the steps of: mechanizing a first hidden state commitment, and using a preset coin mixing server to generate a randomized second hidden state commitment and a preset registration fee according to a preset verification and update strategy; generating a randomized third hidden state commitment based on the randomized second hidden state commitment and the preset randomization processing strategy, and enabling the target task publisher to verify whether the randomized third hidden state commitment meets the preset correctness condition according to the preset training token, and generating a first updated hidden state of the second transaction channel corresponding to the randomized third hidden state commitment when the randomized third hidden state commitment meets the preset correctness condition; blindly signing the randomized third hidden state commitment through the target task publisher, and sending the global model and the blindly signed randomized third hidden state commitment to the data owner, so that the data owner generates a training voucher corresponding to the training token through the preset de-blinding processing strategy; submitting the first updated hidden state through the preset coin mixing server and the target task publisher, and starting the first target payment process, so that the data owner pays the preset registration fee to the preset coin mixing server.

[0084] In the actual implementation process, as a stage in a round of federated learning, the embodiment of the present application can use the data owner to pay a registration fee to the task publisher through the mixing server during the model download stage to obtain the global model and training permissions.

[0085] It is understandable that in the model download phase, the coin-mixing mechanism requires the data owner to pay a registration fee to the coin-mixing server, and the coin-mixing server pays the corresponding registration fee to the task publisher. At this time, based on the security of the coin-mixing, the data owner, the task publisher, and other third parties such as miners cannot determine whether the funds are paid from the data owner to the task publisher. In addition, the coin-mixing server generally requires a service intermediary fee. In order to facilitate the understanding of the decentralized federated learning method based on identity privacy protection and economic incentives of this application by those skilled in the art, this fee can be ignored in the embodiments of this application.

[0086] Specifically, Figure 4 As shown, the specific workflow of the embodiment of the present application in the model download stage is as follows:

[0087] (1) Publishing a pledge transaction: In order to establish a transaction channel, the data owner can establish a one-way pledge transaction from the data owner to the mixing server (i.e., the second one-way pledge transaction), and the mixing server establishes a pledge transaction from the mixing server to the task publisher (i.e., the third one-way pledge transaction).

[0088] It should be noted that in the downloading stage of the traditional simple execution model, the coin mixing server may be attacked by a witch and receive a large number of pledge requests, resulting in a large amount of funds being frozen in the pledge transaction and causing losses. Therefore, the embodiment of the present application designs a registered pledge anonymous authentication (i.e., an anonymous authentication protocol in the pledge stage), thereby ensuring that only when it is proved that the data owner has pledged to the coin mixing server, the coin mixing server will establish a pledge transaction with the task publisher (i.e., the third one-way pledge transaction), wherein the above-mentioned registered pledge anonymous authentication process is as follows: Figure 5 As shown, the issued credentials (tokens) are only used once. The party that issues and verifies the tokens is the coin mixing server. The specific registration and pledge anonymous authentication process is as follows:

[0089] 1) The data owner has established a pledge transaction with the coin-mixing server before the anonymous authentication protocol. The data owner uses the commitment algorithm COM to generate a commitment (C, d) for a random token identifier Rid, and sends the commitment C and the pledge transaction id Tid to the coin-mixing server; the coin-mixing server verifies the legitimacy of the pledge transaction id Tid; then uses the blind signature algorithm to generate a blind signature of the identifier Rid using the commitment C , and Sent to the data owner;

[0090] Among them, the identifier Rid is hidden in the commitment, otherwise, the coin mixing server can easily link the data owner and the corresponding task publisher by comparing the Rid sent by the subsequent task publisher. Therefore, the embodiment of the present application also requires a method that can blindly sign the value hidden in the commitment (such as the Pointcheval-Sanders signature scheme);

[0091] 2) The data owner uses the commitment algorithm to obtain the revealed information d Perform a deblinding operation to obtain the valid signature s corresponding to the token Rid Rid ; Finally, the data owner randomizes Rid To obtain s′ Rid And (Rid, s Rid ) is sent to the task publisher, who verifies the signature and finally sends it to the coin mixing server. Since the PS signature scheme consists of two group elements, and the deblinding operation only affects the second element of the signature, and Rid The first element of the signature is the same. In order to prevent the coin mixing server from linking the data owner and the task publisher through the first element of the signature, it is necessary to randomize s Rid To obtain s′ Rid ;

[0092] 3) The coin mixing server checks whether the token Rid has been used before to prevent replay attacks (i.e. the data owner uses the previously registered token Rid multiple times). before and its signature). If it has not been used, it is determined that the data owner has pledged a transaction, so the coin mixing server also pledges a transaction to the task publisher; in order to prevent replay attacks, the coin mixing server retains a list RS of all previously seen token identifiers; in order to prevent excessive expansion of RS storage, the coin mixing server can periodically update the public and private keys (pk M ,sk M ) and reset the list RS, and the tokens and their signatures used previously are invalid, and the corresponding design will not be repeated here.

[0093] (2) Initialize channel transaction status: After the pledge transaction (i.e., the third one-way pledge transaction) is securely published, the coin mixing server runs the RSUC algorithm to convert the transaction status Sta between the coin mixing server and the task publisher into a hidden state commitment (i.e. the first hidden state commitment) and sent to the task publisher, who checks Ensure the correctness of its starting balance;

[0094] (3) Payment process of the data owner: When the data owner needs to pay the registration fee to the task publisher, the task publisher will commit the transaction status of the channel between the data owner and the mixing server. (i.e., the first hidden state commitment) is sent to the data owner, who verifies After the legitimacy of (i.e. randomize the first hidden state commitment) and send it to the mixing server; after the mixing server verifies it, it uses the RSUC algorithm to update the hidden channel state (i.e. randomized second hidden state commitment), the update amount is the registration fee RigAmt required to achieve the agreement, and then through the atomicity of the transaction channel, the data owner pays RigAnt to the mixing server and gets the updated

[0095] (4) Model download and training certificate issuance: After the data owner verifies the legitimacy of the updated status, it is randomized using the RSUC randomization algorithm. The commitment (i.e., the randomized third hidden state commitment) calculated on the randomly selected training token TrainRid is sent to the task publisher; the task publisher verifies (i.e., randomizing the third hidden state commitment) to check the correctness of the update and store it as the new channel state; as a training authorization, the task publisher commits C Train (i.e., randomized third hidden state commitment) is blindly signed and sent to the data owner together with the global model. The data owner obtains the signature s′ corresponding to TrainRid through deblinding and randomization Train (i.e. training certificate);

[0096] (5) Closing the channel and settling the transaction: In order to maximize their profits, the coin mixing server and the task publisher close the second transaction channel and complete the payment by submitting the latest hidden state (i.e., the first updated hidden state) and initiating the payment process of the coins in the channel (i.e., the first target payment process).

[0097] It should be noted that the pledge anonymous authentication protocol designed in the embodiment of the present application can utilize the commitment scheme and blind signature technology to enable the mixing server to safely verify the reliability of the transaction and perform the pledge without revealing the identity of the data owner, so as to establish a trusted transaction channel. The pledge anonymous authentication protocol further defends against Sybil attacks and denial of service attacks (DoS), and enhances the robustness of the overall incentive mechanism.

[0098] In the above model downloading phase, the data owner can pay a registration fee to the task publisher through the coin mixing mechanism, and use the nature of the blind signature to destroy the link above the transaction, thereby ensuring identity privacy. It should be noted that all information required for participation is exchanged in the anonymous communication network, ensuring anonymity from the network communication.

[0099] Therefore, the embodiments of the present application provide for the introduction of a registration fee mechanism, which allows data owners to pay a certain registration fee to participate in training, thereby effectively increasing the cost for malicious participants and reducing their motivation to participate; at the same time, since the registration process remains anonymous, it not only protects the privacy of data owners, but also improves the security of the entire system, effectively prevents witch attacks and other forms of malicious behavior, and enhances the ability to resist malicious behavior.

[0100] In step S203, the global model is trained using the preset local training data set to obtain the initial federated learning model, and the initial federated learning model is evaluated to update the global model through the evaluation results and the local training data set, and the evaluation and update operations are iteratively performed according to the privacy protection requirements to obtain the final federated learning model. At the same time, based on the economic incentive mechanism and the preset coin mixing server, the target task publisher pays the preset reward fee to the data owner.

[0101] Afterwards, in order to maintain identity anonymity and security, during the model upload stage, the embodiments of the present application may also enable the data owner to use the local training data set to train the global model, and evaluate the trained model (i.e., the initial federated learning model), so as to update the global model through the evaluation results and the local training data set, and iteratively perform the evaluation and update operations according to the privacy protection requirements to obtain the final federated learning model, and upload it back to the task publisher through an anonymous communication channel; anonymous or one-time identity protection measures are also used during the upload process to ensure that the uploaded data updates cannot be associated with a specific identity or previous activities. After the upload is completed, the participants will obtain anonymous incentives (i.e., preset reward fees) through blockchain transactions, and the incentives will be guaranteed through the mixing server. The anonymity of both parties to the transaction.

[0102] Therefore, the embodiments of the present application, through a federated learning framework based on blockchain transaction channels and a hybrid currency architecture, can not only enable data owners to participate in model training in a completely anonymous manner, but also provide them with direct economic incentives, effectively combining privacy protection and public chain economic incentives; in addition, by expanding off-chain transactions, the embodiments of the present application significantly reduce the additional communication and transaction costs caused by frequent transactions between data owners and task publishers in federated learning.

[0103] Optionally, in one embodiment of the present application, based on the economic incentive mechanism and the preset coin mixing server, the target task publisher pays the preset reward fee to the data owner, including: establishing a fourth one-way pledge transaction between the preset coin mixing server and the data publisher according to the preset reward pledge anonymous authentication mechanism; based on the preset coin mixing server, converting the transaction state corresponding to the fourth one-way pledge transaction into a fourth hidden state commitment, and verifying and randomizing the fourth hidden state commitment by the data owner to obtain a randomized fourth hidden state commitment corresponding to the fourth hidden state commitment; sending the randomized fourth hidden state commitment to the preset coin mixing server, so that the preset coin mixing server generates a second updated hidden state and a preset reward fee corresponding to the randomized fourth hidden state commitment according to the randomized fourth hidden state commitment and a preset verification update strategy; verifying whether the second updated hidden state meets the preset legitimacy condition by the target task publisher, and if the second updated hidden state meets the preset legitimacy condition, randomizing the second updated hidden state by the data owner to generate a third updated hidden state corresponding to the second updated hidden state; submitting the third updated hidden state by the preset coin mixing server and the data owner, and starting the second target payment process, so that the target task publisher pays the preset reward fee to the data owner.

[0104] During the specific execution process, in the model upload phase of the embodiment of the present application, the federated learning model obtained through local training can be sent to the task publisher through the data owner. After evaluating the quality of the model, the task publisher updates the global model and pays a reward to the data owner. Similar to the model download phase, in the model upload phase, the embodiment of the present application also requires the task publisher to pay a reward to the data owner through the mixing server.

[0105] Specifically, Figure 6 As shown, the specific workflow of the embodiment of the present application in the model download stage is as follows:

[0106] (1) Reward pledge anonymous authentication: In the reward stage, since the task publisher has pledged a fund to establish a transaction channel in the task publishing stage, the coin mixing server only needs to establish a pledge transaction from the coin mixing server to the data owner (i.e., the fourth one-way pledge transaction); in order to prevent Sybil attacks, the embodiment of the present application also designs a reward pledge anonymous authentication to ensure that the coin mixing server will only establish a pledge transaction with the data publisher when it is proved that the data owner has indeed received the reward.

[0107] Among them, Figure 7 As shown in the figure, the specific execution process of reward pledge anonymous authentication is as follows:

[0108] 1) The data owner obtains the training certificate (TrainRid, s′) before Train), and then use their own local data (i.e., local training data set) to train and obtain the federated learning model LM; the data owner randomly selects a reward token RwRid and promises to commit C rw It is sent to the task publisher together with the federated learning model. The task publisher verifies the legitimacy of the training certificate to confirm that the data owner has registered before and is a legitimate user. The task publisher then evaluates the value of the federated learning model and aggregates it. It should be noted that the task publisher in the embodiment of the present application can use the corresponding robust model detection strategy to detect whether the federated learning model is a low-quality model, such as a poisonous model attacked by poisoning, a low-utility model obtained by a free-rider attack, a duplicate model obtained by a witch attack, etc. If the federated learning model is a low-quality model, no reward will be issued for it, resulting in no income for the task publisher, but instead causing certain losses due to the registration fee, which ultimately plays a role in motivating the elimination of malicious participants and improving the quality of the model.

[0109] 2) The task publisher sends the accompanying commitment to the rewarded data owner to the coin-mixing server to obtain the reward pledge certificate; since the incentive fund pledged by the task publisher (i.e., the first one-way pledge fund) is publicly verifiable, each application for reward pledge by the task publisher is equivalent to freezing a part of the pledge fund. The coin-mixing server verifies whether the unfrozen funds of the task publisher's incentive fund are sufficient. If the verification is sufficient, the coin-mixing server generates a blind signature of the identifier RwRid And send it to the task publisher, who then forwards it to the data owner;

[0110] 3) Data owner Deblind to obtain the valid signature s corresponding to the token RwRid Rw Finally, the data owner randomizes the signature to obtain s′ Rw And send it and the token to the mixing server; it should be noted that the token RwRid is still selected by the data owner, so the task publisher and the mixing server still do not know the specific information of the token, thus preventing the identity of the data owner from being leaked due to collusion between the two;

[0111] 4) The coin mixing server checks whether the token RwRid has been used before to prevent replay attacks; if it has not been used, the coin mixing server also pledges a transaction to the data owner;

[0112] (2) Initializing channel transaction status: After establishing the pledge transaction from the coin mixer to the data owner, the coin mixer also runs the RSUC algorithm to convert the transaction status Sta between the coin mixer and the data owner into a hidden state commitment (i.e., the fourth hidden state commitment) and sends it to the data owner for verification; after verification, the data owner sends the transaction status Sta to the data owner to prevent the coin mixer from colluding with the task publisher to cause the data owner to be transferred through C Sta The identity of the link data owner is obtained by randomizing the commitment (i.e. randomizing the fourth hidden state commitment);

[0113] (3) Payment process of the task publisher: The data owner sends The task publisher sends it to the coin mixing server; after verification, the coin mixing server uses the RSUC algorithm to update the hidden channel status (i.e., the second updated hidden state), the updated amount is the reward fee RwAmt required to achieve the agreement, and then through the atomicity of the transaction channel, the task publisher pays RwAmt to the mixing server and gets the updated

[0114] (4) Distribute economic incentives: Task publishers will verify the update status The legitimacy of the second updated hidden state is verified, and when the second updated hidden state meets the legitimacy condition, it is sent to the data owner; the data owner uses the randomization algorithm to randomize it into (i.e., the third updated hidden state) and stored as the new channel state;

[0115] (5) Closing the channel and settling the transaction: In order to maximize their profits, the mixing server and the data owner close the channel (i.e., the first transaction channel) by submitting the latest hidden state (i.e., the third updated hidden state) and initiating the payment process of the coins in the channel (i.e., the second target payment process) to complete the reward payment.

[0116] It should be noted that during the model uploading phase, the task publisher paid a bonus to the data owner as an incentive, which also ensured the identity privacy of the data owner. In addition, the two fees mentioned above, the registration fee RigAmt and the reward RwAmt, will be known by the coin mixing server. Therefore, in order to ensure that the transactions between different data owners and task publishers are indistinguishable, the registration fees and rewards for all data owners in each round should be consistent and agreed upon. If you want to give more incentives, you can do so by issuing two or more rewards of the same amount.

[0117] In summary, the embodiments of the present application integrate blockchain technology and anonymous communication networks, and utilize the three main entity roles of task publishers, data owners and coin mixing servers, combined with pre-designed collaboration and interaction mechanisms, which not only ensures the identity privacy and security of data owners when participating in model training, but also introduces a blockchain-based incentive mechanism to encourage more data owners to participate, thereby achieving efficient operation of the system and reducing communication costs; in addition, the embodiments of the present application also design anonymous registration and transaction channels, and introduce anonymous authentication protocols for registration fees and pledge stages, while ensuring the anonymity of the identities of participants and the security of transactions through coin mixing technology to resist possible malicious attacks such as witch attacks during the incentive process, further enhancing the robustness against witch attacks in an anonymous environment, significantly improving the participation, security and efficiency of the federated learning system, and expanding off-chain transactions with the help of the nature of the transaction channel, reducing the additional costs caused by frequent transactions.

[0118] According to the decentralized federated learning method based on identity privacy protection and economic incentives proposed in the embodiment of the present application, the target federated learning task is published based on a preset decentralized network and a target task publisher, and the economic incentive mechanism and privacy protection requirements of the target federated learning task are determined, so as to determine the data owner who receives the target federated learning task according to the economic incentive mechanism and privacy protection requirements; the data owner downloads the global model, economic incentive mechanism and privacy protection requirements corresponding to the target federated learning task through a preset anonymous communication channel, and pays a preset registration fee to obtain the training certificate corresponding to the global model; the global model is trained using a preset local training data set to obtain an initial federated learning model, and the initial federated learning model is evaluated to update the global model through the evaluation results and the local training data set, and the evaluation and update operations are iteratively performed according to the privacy protection requirements to obtain the final federated learning model, and at the same time, based on the economic incentive mechanism and the preset coin mixing server, the target task publisher pays the preset reward fee to the data owner. This application introduces an anonymous authentication protocol in the registration fee and pledge phase to resist malicious attacks such as possible witch attacks during the incentive process, thereby significantly improving the participation, security and efficiency of the federated learning system.

[0119] Secondly, a decentralized federated learning device based on identity privacy protection and economic incentives proposed according to an embodiment of the present application is described with reference to the accompanying drawings.

[0120] Figure 8 It is a block diagram of a decentralized federated learning device based on identity privacy protection and economic incentives in an embodiment of the present application.

[0121] like Figure 8As shown, the decentralized federated learning device 10 based on identity privacy protection and economic incentives includes: a task publishing module 100, a download module 200 and a reward module 300.

[0122] Among them, the task publishing module 100 is used to publish the target federated learning task based on the preset decentralized network and the target task publisher, and determine the economic incentive mechanism and privacy protection requirements of the target federated learning task, so as to determine the data owner who receives the target federated learning task according to the economic incentive mechanism and privacy protection requirements.

[0123] The download module 200 is used to download the global model, economic incentive mechanism and privacy protection requirements corresponding to the target federated learning task through the data owner using the preset anonymous communication channel, and pay the preset registration fee to obtain the training certificate corresponding to the global model.

[0124] The reward module 300 is used to train the global model using a preset local training data set to obtain an initial federated learning model, and to evaluate the initial federated learning model so as to update the global model through the evaluation results and the local training data set, and to iteratively perform the evaluation and update operations according to the privacy protection requirements to obtain the final federated learning model. At the same time, based on the economic incentive mechanism and the preset coin mixing server, the target task publisher pays the preset reward fee to the data owner.

[0125] Optionally, in one embodiment of the present application, the task issuing module 100 includes: a determining unit, a constructing unit and an analyzing unit.

[0126] Among them, the determination unit is used to determine the privacy protection requirements and task payment budget of the target federated learning task through the target task publisher.

[0127] The construction unit is used to construct a first transaction channel between the target task publisher and the preset currency mixing server, and enable the target task publisher to pledge a first one-way pledge fund in the first transaction channel.

[0128] An analysis unit is used to publish a target federated learning task on a preset decentralized network through a target task publisher, and broadcast a first one-way pledge transaction corresponding to a first one-way pledge fund, so that a data owner who receives the first one-way pledge transaction analyzes the target federated learning task according to the first one-way pledge fund and the task payment budget, and generates a corresponding analysis result, so as to accept the target federated learning task if the analysis result meets the preset reception requirements.

[0129] Optionally, in one embodiment of the present application, the download module 200 includes: a first establishment unit, a first conversion unit, a first verification unit, an update unit, a first randomization unit, a blind signature unit and a first submission unit.

[0130] Among them, the first establishment unit is used to establish a second one-way pledge transaction from the data owner to the preset coin mixing server, and to establish a third one-way pledge transaction from the preset coin mixing server to the target task publisher through the preset coin mixing server according to the preset registration pledge anonymous authentication strategy, so as to build a second transaction channel according to the second one-way pledge transaction and the third one-way pledge transaction.

[0131] The first conversion unit is used to convert the transaction state of the third one-way pledge transaction into a first hidden state commitment, so that the target task publisher verifies whether the first hidden state commitment meets a preset correctness condition.

[0132] The first verification unit is used to verify whether the first hidden state commitment meets the preset legality condition through the data owner when the first hidden state commitment meets the preset correctness condition.

[0133] The updating unit is used to randomize the first hidden state commitment to obtain a randomized first hidden state commitment when the first hidden state commitment meets a preset legality condition, and to generate a randomized second hidden state commitment and a preset registration fee according to a preset verification and update strategy using a preset coin mixing server.

[0134] The first randomization unit is used to generate a randomized third hidden state commitment based on the randomized second hidden state commitment and a preset randomization processing strategy, and at the same time enable the target task publisher to verify whether the randomized third hidden state commitment meets the preset correctness condition according to the preset training token, and generate the first updated hidden state of the second transaction channel corresponding to the randomized third hidden state commitment when the randomized third hidden state commitment meets the preset correctness condition.

[0135] The blind signature unit is used to blindly sign the randomized third hidden state commitment through the target task publisher, and send the global model and the randomized third hidden state commitment after the blind signature to the data owner, so that the data owner can generate the training credential corresponding to the training token through the preset de-blinding processing strategy.

[0136] The first submission unit is used to submit the first updated hidden state through the preset coin mixing server and the target task publisher, and start the first target payment process, so that the data owner pays the preset registration fee to the preset coin mixing server.

[0137] Optionally, in one embodiment of the present application, the reward module 300 includes: a second establishment unit, a second conversion unit, a second verification unit, a second randomization unit and a second submission unit.

[0138] Among them, the second establishing unit is used to establish a fourth one-way pledge transaction between the preset mixing server and the data publisher according to the preset reward pledge anonymous authentication mechanism.

[0139] The second conversion unit is used to convert the transaction state corresponding to the fourth one-way pledge transaction into a fourth hidden state commitment based on a preset coin mixing server, and verify and randomize the fourth hidden state commitment through the data owner to obtain a randomized fourth hidden state commitment corresponding to the fourth hidden state commitment.

[0140] The second verification unit is used to send the randomized fourth hidden state commitment to a preset coin mixing server, so that the preset coin mixing server generates a second updated hidden state and a preset reward fee corresponding to the randomized fourth hidden state commitment according to the randomized fourth hidden state commitment and a preset verification update strategy.

[0141] The second randomization unit is used to verify whether the second updated hidden state meets the preset legitimacy conditions through the target task publisher, and if the second updated hidden state meets the preset legitimacy conditions, the second updated hidden state is randomized by the data owner to generate a third updated hidden state corresponding to the second updated hidden state.

[0142] The second submission unit is used to submit the third updated hidden state through the preset coin mixing server and the data owner, and start the second target payment process, so that the target task publisher pays the preset reward fee to the data owner.

[0143] It should be noted that the aforementioned explanation of the embodiment of the decentralized federated learning method based on identity privacy protection and economic incentives is also applicable to the decentralized federated learning device based on identity privacy protection and economic incentives in this embodiment, and will not be repeated here.

[0144] According to the decentralized federated learning device based on identity privacy protection and economic incentives proposed in the embodiment of the present application, it includes a task publishing module, which is used to publish the target federated learning task based on the preset decentralized network and the target task publisher, and determine the economic incentive mechanism and privacy protection requirements of the target federated learning task, so as to determine the data owner who receives the target federated learning task according to the economic incentive mechanism and privacy protection requirements; a download module, which is used to download the global model, economic incentive mechanism and privacy protection requirements corresponding to the target federated learning task through the data owner using the preset anonymous communication channel, and pay the preset registration fee to obtain the training certificate corresponding to the global model; a reward module, which is used to train the global model using the preset local training data set to obtain the initial federated learning model, and evaluate the initial federated learning model, so as to update the global model through the evaluation results and the local training data set, and iteratively perform the evaluation and update operations according to the privacy protection requirements to obtain the final federated learning model, and at the same time, based on the economic incentive mechanism and the preset mixing currency server, the target task publisher pays the preset reward fee to the data owner. This application introduces the registration fee and the anonymous authentication protocol in the pledge stage to resist the possible malicious attacks such as witch attacks in the incentive process, thereby significantly improving the participation, security and efficiency of the federated learning system.

[0145] Fig. 9 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. The electronic device may include:

[0146] A memory 901 , a processor 902 , and a computer program stored in the memory 901 and executable on the processor 902 .

[0147] When the processor 902 executes the program, the decentralized federated learning method based on identity privacy protection and economic incentives provided in the above embodiment is implemented.

[0148] Furthermore, the electronic device further comprises:

[0149] The communication interface 903 is used for communication between the memory 901 and the processor 902 .

[0150] The memory 901 is used to store computer programs that can be executed on the processor 902 .

[0151] The memory 901 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0152] If the memory 901, the processor 902 and the communication interface 903 are implemented independently, the communication interface 903, the memory 901 and the processor 902 can be connected to each other through a bus and communicate with each other. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component (PCI) bus or an Extended Industry Standard Architecture (EISA) bus. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Fig. 9 Only one thick line is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0153] Optionally, in a specific implementation, if the memory 901, the processor 902 and the communication interface 903 are integrated on a chip, the memory 901, the processor 902 and the communication interface 903 can communicate with each other through an internal interface.

[0154] The processor 902 may be a central processing unit (CPU), or an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present application.

[0155] An embodiment of the present application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-mentioned decentralized federated learning method based on identity privacy protection and economic incentives.

[0156] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or N embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0157] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of the features. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0158] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, fragment or portion of code comprising one or N executable instructions for implementing the steps of a custom logical function or process, and the scope of the preferred embodiments of the present application includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in reverse order depending on the functions involved, which should be understood by technicians in the technical field to which the embodiments of the present application belong.

[0159] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in combination with these instruction execution systems, devices or apparatuses. For the purpose of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in combination with these instruction execution systems, devices or apparatuses. More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or N wirings (electronic devices), a portable computer disk box (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically by optically scanning the paper or other medium and then editing, interpreting or processing in other suitable ways as necessary and then storing it in a computer memory.

[0160] It should be understood that the various parts of the present application can be implemented by hardware, software, firmware or a combination thereof. In the above embodiment, the N steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. If implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0161] A person skilled in the art may understand that all or part of the steps in the method for implementing the above-mentioned embodiment may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiment.

[0162] In addition, each functional unit in each embodiment of the present application may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0163] The storage medium mentioned above may be a read-only memory, a disk or an optical disk, etc. Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limiting the present application. A person of ordinary skill in the art may change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. A decentralized federated learning method based on identity privacy protection and economic incentives, characterized in that: The following steps are involved: Based on a preset decentralized network and a target task publisher, a target federated learning task is published, and an economic incentive mechanism and privacy protection requirements of the target federated learning task are determined, so as to determine a data owner who receives the target federated learning task according to the economic incentive mechanism and the privacy protection requirements; The data owner downloads the global model corresponding to the target federated learning task, the economic incentive mechanism, and the privacy protection requirements through a preset anonymous communication channel, and pays a preset registration fee to obtain a training certificate corresponding to the global model; The global model is trained using a preset local training data set to obtain an initial federated learning model, and the initial federated learning model is evaluated to update the global model through the evaluation results and the local training data set, and the evaluation and update operations are iteratively performed according to the privacy protection requirements to obtain the final federated learning model. At the same time, based on the economic incentive mechanism and the preset coin mixing server, the target task publisher pays a preset reward fee to the data owner.

2. The method according to claim 1, characterized in that The method of publishing a target federated learning task based on a preset decentralized network and a target task publisher, and determining an economic incentive mechanism and a privacy protection requirement for the target federated learning task, so as to determine a data owner who receives the target federated learning task according to the economic incentive mechanism and the privacy protection requirement, includes: Determine the privacy protection requirements and task payment budget of the target federated learning task through the target task publisher; Establishing a first transaction channel between the target task publisher and the preset currency mixing server, and allowing the target task publisher to pledge a first one-way pledge fund in the first transaction channel; The target federated learning task is published on the preset decentralized network by the target task publisher, and the first one-way pledge transaction corresponding to the first one-way pledge fund is broadcast, so that the data owner who receives the first one-way pledge transaction analyzes the target federated learning task according to the first one-way pledge fund and the task payment budget, and generates a corresponding analysis result, so as to accept the target federated learning task if the analysis result meets the preset reception requirements.

3. The method according to claim 2, characterized in that The downloading of the global model corresponding to the target federated learning task, the economic incentive mechanism and the privacy protection requirements by the data owner through a preset anonymous communication channel, and paying a preset registration fee to obtain the training certificate corresponding to the global model includes: Establishing a second one-way pledge transaction from the data owner to the preset coin mixing server, and establishing a third one-way pledge transaction from the preset coin mixing server to the target task publisher through the preset coin mixing server according to a preset registration pledge anonymous authentication policy, so as to construct a second transaction channel according to the second one-way pledge transaction and the third one-way pledge transaction; Converting the transaction state of the third one-way pledge transaction into a first hidden state commitment, so that the target task publisher verifies whether the first hidden state commitment meets a preset correctness condition; In the case where the first hidden state commitment satisfies the preset correctness condition, verifying by the data owner whether the first hidden state commitment satisfies the preset legality condition; In the case where the first hidden state commitment satisfies the preset legality condition, the first hidden state commitment is randomized to obtain a randomized first hidden state commitment, and the preset coin mixing server is used to generate a randomized second hidden state commitment and the preset registration fee according to a preset verification and update strategy; Based on the randomized second hidden state commitment and the preset randomization processing strategy, a randomized third hidden state commitment is generated, and at the same time, the target task publisher verifies whether the randomized third hidden state commitment satisfies the preset correctness condition according to the preset training token, and generates a first updated hidden state of the second transaction channel corresponding to the randomized third hidden state commitment when the randomized third hidden state commitment satisfies the preset correctness condition; The target task publisher blindly signs the randomized third hidden state commitment, and sends the global model and the blindly signed randomized third hidden state commitment to the data owner, so that the data owner generates a training credential corresponding to the training token through a preset de-blinding processing strategy; The first updated hidden state is submitted through the preset coin mixing server and the target task publisher, and a first target payment process is initiated, so that the data owner pays a preset registration fee to the preset coin mixing server.

4. The method according to claim 3, characterized in that: Based on the economic incentive mechanism and the preset currency mixing server, the target task publisher pays the preset reward fee to the data owner, including: Establishing a fourth one-way pledge transaction between the preset coin mixing server and the data publisher according to the preset reward pledge anonymous authentication mechanism; Based on the preset coin mixing server, converting the transaction state corresponding to the fourth one-way pledge transaction into a fourth hidden state commitment, and verifying and randomizing the fourth hidden state commitment by the data owner to obtain a randomized fourth hidden state commitment corresponding to the fourth hidden state commitment; Sending the randomized fourth hidden state commitment to the preset coin mixing server, so that the preset coin mixing server generates a second updated hidden state corresponding to the randomized fourth hidden state commitment and the preset reward fee according to the randomized fourth hidden state commitment and a preset verification update strategy; Verifying, by the target task publisher, whether the second updated hidden state satisfies the preset legality condition, and, if the second updated hidden state satisfies the preset legality condition, performing random processing on the second updated hidden state by the data owner to generate a third updated hidden state corresponding to the second updated hidden state; The third updated hidden state is submitted through the preset coin mixing server and the data owner, and a second target payment process is initiated, so that the target task publisher pays the preset reward fee to the data owner.

5. A decentralized federated learning device based on identity privacy protection and economic incentives, characterized in that: include: A task publishing module, used to publish a target federated learning task based on a preset decentralized network and a target task publisher, and determine the economic incentive mechanism and privacy protection requirements of the target federated learning task, so as to determine the data owner who receives the target federated learning task according to the economic incentive mechanism and the privacy protection requirements; A download module, used to download the global model, the economic incentive mechanism and the privacy protection requirements corresponding to the target federated learning task through the data owner using a preset anonymous communication channel, and pay a preset registration fee to obtain a training certificate corresponding to the global model; A reward module is used to train the global model with a preset local training data set to obtain an initial federated learning model, and to evaluate the initial federated learning model, so as to update the global model through the evaluation results and the local training data set, and to iteratively perform the evaluation and update operations according to the privacy protection requirements to obtain the final federated learning model. At the same time, based on the economic incentive mechanism and the preset coin mixing server, the target task publisher pays the preset reward fee to the data owner.

6. The device according to claim 5, characterized in that The task publishing module includes: A determination unit, configured to determine the privacy protection requirements and task payment budget of the target federated learning task through the target task publisher; A construction unit, configured to construct a first transaction channel between the target task publisher and the preset currency mixing server, and enable the target task publisher to pledge a first one-way pledge fund in the first transaction channel; An analysis unit is used to publish the target federated learning task on the preset decentralized network through the target task publisher, and broadcast the first one-way pledge transaction corresponding to the first one-way pledge fund, so that the data owner who receives the first one-way pledge transaction analyzes the target federated learning task according to the first one-way pledge fund and the task payment budget, and generates a corresponding analysis result, so as to accept the target federated learning task if the analysis result meets the preset reception requirements.

7. The device according to claim 6, characterized in that The download module includes: A first establishing unit is used to establish a second one-way pledge transaction from the data owner to the preset coin mixing server, and establish a third one-way pledge transaction from the preset coin mixing server to the target task publisher through the preset coin mixing server according to a preset registration pledge anonymous authentication strategy, so as to construct a second transaction channel according to the second one-way pledge transaction and the third one-way pledge transaction; A first conversion unit, configured to convert the transaction state of the third one-way pledge transaction into a first hidden state commitment, so that the target task publisher verifies whether the first hidden state commitment satisfies a preset correctness condition; a first verification unit, configured to verify, through the data owner, whether the first hidden state commitment satisfies a preset legality condition if the first hidden state commitment satisfies the preset correctness condition; an updating unit, configured to randomize the first hidden state commitment to obtain a randomized first hidden state commitment when the first hidden state commitment satisfies the preset legality condition, and generate a randomized second hidden state commitment and the preset registration fee by using the preset coin mixing server according to a preset verification and update strategy; A first randomization unit is used to generate a randomized third hidden state commitment based on the randomized second hidden state commitment and a preset randomization processing strategy, and at the same time, enable the target task publisher to verify whether the randomized third hidden state commitment satisfies the preset correctness condition according to a preset training token, and generate a first updated hidden state of the second transaction channel corresponding to the randomized third hidden state commitment when the randomized third hidden state commitment satisfies the preset correctness condition; A blind signature unit, configured to blindly sign the randomized third hidden state commitment through the target task publisher, and send the global model and the blindly signed randomized third hidden state commitment to the data owner, so that the data owner generates a training credential corresponding to the training token through a preset de-blinding processing strategy; The first submitting unit is used to submit the first updated hidden state through the preset coin mixing server and the target task publisher, and start a first target payment process so that the data owner pays a preset registration fee to the preset coin mixing server.

8. The device according to claim 7, characterized in that The reward module includes: A second establishing unit is used to establish a fourth one-way pledge transaction between the preset coin mixing server and the data publisher according to a preset reward pledge anonymous authentication mechanism; A second conversion unit is configured to convert the transaction state corresponding to the fourth one-way pledge transaction into a fourth hidden state commitment based on the preset coin mixing server, and verify and randomize the fourth hidden state commitment through the data owner to obtain a randomized fourth hidden state commitment corresponding to the fourth hidden state commitment; A second verification unit is used to send the randomized fourth hidden state commitment to the preset coin mixing server, so that the preset coin mixing server generates a second updated hidden state corresponding to the randomized fourth hidden state commitment and the preset reward fee according to the randomized fourth hidden state commitment and a preset verification update strategy; A second randomization unit is used to verify whether the second updated hidden state satisfies the preset legality condition through the target task publisher, and if the second updated hidden state satisfies the preset legality condition, randomize the second updated hidden state through the data owner to generate a third updated hidden state corresponding to the second updated hidden state; The second submission unit is used to submit the third updated hidden state through the preset coin mixing server and the data owner, and start a second target payment process so that the target task publisher pays the preset reward fee to the data owner.

9. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the decentralized federated learning method based on identity privacy protection and economic incentives as described in any one of claims 1 to 4.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: The program is executed by a processor to implement the decentralized federated learning method based on identity privacy protection and economic incentives as described in any one of claims 1 to 4.

Images