Federal learning incentive method, device and system based on reverse auction

Abstract

The invention provides a federated learning excitation method, device and system based on reverse auction. The method comprises the following steps: determining clients participating in federated learning through a reverse auction process; initializing a global model, and distributing the global model to each client; receiving an update model ciphertext sent by each client; generating a new global model according to the updated model ciphertext; and determining rewards of the clients, and issuing the rewards to the corresponding clients. By using the scheme of the invention, not only can the data privacy of the federated learning participants be effectively protected, but also the effective resource allocation under the limited budget can be realized, the enthusiasm of the participants can be effectively improved, and the federated learning efficiency can be improved.

Description

Technical Field

[0001] The present invention relates to the field of federated learning, and in particular to a reverse auction-based federated learning incentive method, device and system. Background Art

[0002] Federated learning (FL), as a distributed machine learning method, has attracted the attention of many researchers in recent years and has been widely used in various fields, such as smart healthcare, smart cities, finance, and smart industry.

[0003] In federated learning, data is retained in the local physical domain, and participating nodes use their own data to train local models separately and jointly build a joint model. The federated learning incentive mechanism is a mechanism used in the federated learning system to encourage and promote participants to actively participate in model training. In federated learning, multiple participants work together to train a shared machine learning model while keeping their respective data privacy private. Since participants need to consume their own computing resources and electricity to train the model, an incentive mechanism is needed to compensate for these costs and encourage participants to contribute high-quality data and computing resources.

[0004] How to more reasonably and effectively enhance the motivation of participants and protect their sensitive information from being leaked is an important issue that the industry needs to solve. Summary of the invention

[0005] The present invention provides a reverse auction-based federated learning incentive method, device and system to solve the problem of insufficient incentives for participants in existing federated learning.

[0006] To this end, the present invention provides the following technical solutions: The present invention provides a reverse auction-based federated learning incentive method, the method comprising: Identify clients participating in federated learning through a reverse auction process; Initialize a global model and distribute the global model to each client; Receive the updated model ciphertext sent by each client; generating a new global model according to the updated model ciphertext; Determine the rewards for each client, and distribute the rewards to the corresponding clients.

[0007] Optionally, determining the client participating in federated learning through the reverse auction process includes: Launching auctions for federated learning; Acquire bidding information submitted by multiple clients, wherein the bidding information includes bid information and client information; Based on the budget constraint and the bidding information, a reverse auction algorithm is used to determine the clients participating in federated learning.

[0008] Optionally, the bidding information is encrypted information; the method further includes: decrypting the encrypted information using a homomorphically encrypted private key to obtain corresponding bidding information and client information.

[0009] Optionally, the determining the client participating in federated learning by using a reverse auction algorithm based on the budget constraint and the bidding information includes: Determine the evaluation objectives; Determine the evaluation indicators according to the evaluation objectives; Scoring each client participating in the bidding according to the evaluation index, and prioritizing the clients according to the scoring results; Clients with higher priorities are selected in turn and the number of clients participating in federated learning is determined according to the budget constraint.

[0010] Optionally, the evaluation index includes one or more; multiple evaluation indexes have the same or different weights.

[0011] Optionally, the multiple evaluation indicators include at least two or more of the following: data quality, computing power, and network bandwidth.

[0012] Optionally, generating a new global model according to the updated model ciphertext includes: Decrypting the update model ciphertext to obtain an update model; Aggregate the updated models of each client to obtain a new global model.

[0013] Optionally, determining the reward for each client includes: Determine the contribution of each client; The reward for each client is determined based on the contribution and budget constraints.

[0014] Optionally, determining the reward for each client includes: Determine the contribution of each client; Obtaining historical behavior records of each client, and determining a behavior score of each client based on the historical behavior records; Calculate the comprehensive score of the client according to the behavior score and contribution of the client; The rewards for each client are determined based on the client's comprehensive score and budget constraints.

[0015] Optionally, determining the contribution of each client includes: Determine the performance of the updated model of each client on the validation set, and determine the contribution of each client according to the performance; or Calculating the difference between the updated model parameters of each client and the new global model parameters, and determining the contribution of the client according to the difference; or A characteristic function representing the cooperation benefit is constructed, and the Shapley value of the client is calculated using the characteristic function, and the contribution of the client is determined according to the Shapley value.

[0016] The present invention also provides a reverse auction-based federated learning incentive device, the device comprising: The client selection module is used to determine the clients participating in federated learning through an auction process; A model distribution module, used for initializing a global model and distributing the global model to each client; A receiving module, used for receiving the updated model ciphertext sent by each client; A model updating module, used for generating a new global model according to the updated model ciphertext; A reward determination module, used to determine the rewards for each client; A reward distribution module, used to distribute the reward to the corresponding client; The device sends the entire federated learning process information to the supervision center, and the federated learning process information includes training process information, compliance information of reward issuance, and information of illegal clients.

[0017] Optionally, the client selection module includes: An auction initiation unit, used to initiate auctions for federated learning; An information acquisition unit, used to acquire bidding information submitted by multiple clients, wherein the bidding information includes bid information and client information; The screening unit is used to determine the client participating in the federated learning by using a reverse auction algorithm based on the budget constraint and the bidding information.

[0018] Optionally, the reward determination module includes: A contribution determination unit, used to determine the contribution of each client; The first reward determination unit is used to determine the reward of each client according to the contribution degree.

[0019] Optionally, the reward determination module includes: A contribution determination unit, used to determine the contribution of each client; A behavior scoring unit, used to obtain the historical behavior records of each client and determine the behavior score of each client according to the historical behavior records; A calculation unit, which calculates a comprehensive score of the client according to the behavior score and contribution of the client; The second reward determination unit is used to determine the reward for each client according to the comprehensive score of each client and the budget constraint.

[0020] The present invention also provides a reverse auction-based federated learning incentive system, the system comprising: a central server and multiple clients; the central server comprises the reverse auction-based federated learning incentive device.

[0021] Optionally, the system further includes: a supervision center for monitoring the federated learning process, recording illegal clients, and supervising the compliance of reward issuance.

[0022] On the other hand, the present invention also provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the reverse auction-based federated learning incentive method are executed.

[0023] The reverse auction-based federated learning incentive method, device and system provided by the present invention can ensure the privacy of data during transmission and calculation by encryption, so that participants can jointly train models without leaking original data, thereby achieving effective protection of data privacy of participants in federated learning. Moreover, the clients participating in federated learning are determined through the reverse auction process, allowing participants to bid dynamically according to their resources and needs, ensuring the effective allocation of resources under a limited budget. By incentivizing participants, the problem of insufficient motivation of participants in traditional federated learning is solved. The setting of the reverse auction mechanism enables the system to dynamically adjust rewards according to the resources and needs of participants under a limited budget, ensuring the effective allocation of resources. This mechanism not only improves the overall efficiency of the system, but also increases the satisfaction of participants, because participants can obtain corresponding rewards according to their contributions.

[0024] In some embodiments, when determining rewards, the corresponding rewards can be calculated based on the contribution of each participant's client and the budget constraints; in some embodiments, when determining rewards, the comprehensive score of the client can also be calculated based on the contribution of each participant's client and its historical behavior records, and the corresponding rewards can be determined based on the comprehensive score and budget constraints; thereby, in specific applications, it can be flexibly set as needed to better meet the actual needs of different application environments.

[0025] Furthermore, by introducing a regulatory center to monitor the entire federated learning process, including the training process, the compliance of reward distribution, and the accountability of illegal entities, the compliance of the training process and reward distribution is ensured. The existence of the regulatory center enhances the standardization and trust of the system and ensures that violations can be discovered and handled in a timely manner. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0027] Figure 1 It is a flow chart of the reverse auction-based federated learning incentive method provided by the present invention; Figure 2 is a flow chart of using a reverse auction algorithm to determine a client participating in federated learning in an embodiment of the present invention; Figure 3 It is a structural schematic diagram of a reverse auction-based federated learning incentive device provided by the present invention; Figure 4 It is a structural diagram of the reverse auction-based federated learning incentive system provided by the present invention. DETAILED DESCRIPTION

[0028] The specific implementation of the present invention is described in detail below in conjunction with the accompanying drawings. It should be understood that the specific implementation described here is only used to illustrate and explain the present invention, and is not used to limit the present invention.

[0029] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0030] The federated learning incentive mechanism is a mechanism used in the federated learning system to encourage and promote participants to actively participate in model training. In federated learning, multiple participants work together to train a shared machine learning model while keeping their respective data privacy private. Since participants need to consume their own computing resources and electricity to train the model, an incentive mechanism is needed to compensate for these costs and encourage participants to contribute high-quality data and computing resources.

[0031] The design of the incentive mechanism needs to take into account multiple factors, including the resource constraints of participants, the amount and quality of data, and the overall goals of the system. Budget constraints are an important consideration in the incentive mechanism, which determines the total amount available for rewards. Under a limited budget, the incentive mechanism needs to distribute rewards fairly and efficiently to ensure the sustainability of the system and the satisfaction of participants.

[0032] In order to enhance the enthusiasm of participants in federated learning, thereby improving the efficiency of model training and maximizing the performance of the model within the budget, the embodiments of the present invention provide a method, device and system for incentivizing federated learning based on reverse auction, which determines the clients participating in federated learning through the reverse auction process, generates a new global model according to the updated model ciphertext after receiving the updated model ciphertext sent by each client, determines the reward of each client, and distributes the reward to the corresponding client.

[0033] Reverse auctions are a method for efficient resource allocation under budget constraints. In federated learning, reverse auctions allow participants to make bids based on their resources and needs, and the central server selects the optimal set of participants for model training based on these bids and budget constraints. This mechanism ensures efficient use of resources and maximizes model performance within the budget.

[0034] In the solution of the present invention, the central server can initialize the federated learning platform, including setting system parameters, auction rules, budget constraints, and further including regulatory policies, etc. In addition, the client group participating in the federated learning is determined through the reverse auction process.

[0035] Clients participating in the reverse auction need to evaluate their own resources, which may include computing power, storage capacity, data volume and data quality, etc.; their bids in the auction are determined based on the evaluation results, reflecting their willingness and conditions for participating in federated learning.

[0036] like Figure 1 FIG. 1 is a flow chart of a reverse auction-based federated learning incentive method provided by the present invention, comprising the following steps: In step 101, clients participating in federated learning are determined through a reverse auction process.

[0037] Specifically, the reverse auction process may be initiated by the central server. The specific process of the reverse auction is as follows: (1) Initiate an auction for federated learning so that participating clients can submit bid information based on their own resource evaluations.

[0038] (2) Obtain bidding information submitted by multiple clients, where the bidding information includes bid information and client information.

[0039] The bidding information may be encrypted information to ensure the privacy of the bidding information of each client. Accordingly, the encrypted information is decrypted to obtain the corresponding bidding information and client information.

[0040] It should be noted that the encryption and decryption of the bidding information may adopt an encryption algorithm agreed upon or negotiated between the central server and the client, and use a corresponding key for encryption and decryption, which is not limited in this embodiment of the present invention.

[0041] For example, in a non-limiting embodiment, a homomorphic encryption algorithm is used, and a central server generates a homomorphic encryption public key and private key pair, and distributes the public key to all clients. Accordingly, the client uses the public key to encrypt the information to be transmitted to the central server, and the central server uses the private key to decrypt the received ciphertext to ensure the confidentiality of the information during transmission.

[0042] (3) Based on the budget constraint and the bidding information, a reverse auction algorithm is used to determine the clients participating in federated learning.

[0043] In some embodiments, after receiving the bidding information submitted by the client, the client may be verified to enhance trust and ensure that the information provided by the client is authentic and reliable. The specific verification may be carried out in one or more ways, such as: (1) Adopt an identity authentication mechanism, such as a blockchain-based identity authentication mechanism, which uses a flexible consensus algorithm with zero-knowledge proof to verify the identity of each participant. This mechanism can achieve lightweight identity authentication while ensuring high model accuracy, solving the problem that identity authentication relies on a fully trusted central server and has high computational costs.

[0044] (2) Data integrity verification mechanism: perform integrity verification on the bidding information submitted by the client.

[0045] It should be noted that, in specific implementation, there may be other verification mechanisms, which can be determined according to actual needs, and this embodiment of the present invention does not limit this.

[0046] like Figure 2 As shown, it is a flow chart of using a reverse auction algorithm to determine a client participating in federated learning in an embodiment of the present invention, including the following steps: Step 201, determine the evaluation target.

[0047] The evaluation objective refers to the objective to be achieved by the reverse auction, such as one or more of the following objectives, but not limited to: maximizing model performance, minimizing costs, improving data diversity, etc.

[0048] Step 202: determining an evaluation index according to the evaluation target.

[0049] The evaluation indicators corresponding to each evaluation target may include one or more. For example, in some embodiments, the evaluation indicators may include but are not limited to any one or more of the following: data quality (such as data accuracy, completeness, consistency, timeliness, etc.), computing power (such as CPU / GPU speed, memory size, etc.), network bandwidth, etc. Different evaluation indicators may have the same or different weights, which is not limited in the embodiments of the present invention.

[0050] Step 203: score each client participating in the bidding according to the evaluation index, and prioritize the clients according to the scoring results.

[0051] The bidding information submitted by the client includes bid information and client information, wherein the client information may include but is not limited to the content covered by the above-mentioned evaluation indicators, for example, the client information includes the client's capabilities and resource information.

[0052] For each evaluation indicator, the client is scored, and then the total score is calculated through simple weighting or complex combination functions, depending on the evaluation requirements.

[0053] The weights of different evaluation indicators may be different, and may be determined based on the contribution of each evaluation indicator to the overall evaluation result. For example, they may be determined through expert opinions, historical data analysis, experimental adjustments, etc.

[0054] The clients are prioritized according to the total scores. Specifically, the priorities may be sorted from high to low or from low to high, which is not limited in the embodiment of the present invention.

[0055] It should be noted that for the same type of indicators, the corresponding information provided by different clients may have differences in dimensions, numerical ranges, etc. In view of such differences, the same type of indicators provided by different clients can be standardized to eliminate the impact of differences in dimensions and numerical ranges. Normalization and standardization methods can be used to convert the indicator values ​​provided by the client to a unified scale.

[0056] Step 204 , selecting clients with higher priorities in turn and determining the number of clients participating in federated learning according to the budget constraint.

[0057] Based on the scores and budget constraints of each client, the list of clients participating in federated learning is finally determined, which can ensure that the selected clients can meet the needs of federated learning tasks and have good performance in terms of data quality, computing power, and network conditions.

[0058] Continue to participate Figure 1 , in step 102, the global model is initialized and distributed to each client.

[0059] Specifically, the global model can be initialized by a central server and distributed to selected clients.

[0060] To ensure data security, the global model distributed to each client can be encrypted. For example, a homomorphic encryption algorithm can be used. Homomorphic encryption allows direct calculations on encrypted data, so that each client can directly use the encrypted global model for training locally to generate an updated model. This means that participants can train and update the model without leaking the original data, ensuring the security of data transmission and improving data processing efficiency.

[0061] Similarly, the client can use the public key to homomorphically encrypt the updated model and send it to the central server.

[0062] In step 103, the updated model ciphertext sent by each client is received.

[0063] In step 104, a new global model is generated according to the updated model ciphertext.

[0064] Specifically, the central server can use the homomorphically encrypted private key to decrypt the updated model ciphertext, and then aggregate the updated models of each client to obtain a new global model; or the central server directly aggregates the received updated model ciphertext to obtain a new global model.

[0065] In step 105, the reward for each client is determined, and the reward is distributed to the corresponding client.

[0066] In some embodiments, the reward for each client may be determined based on the contribution of each client and a budget constraint.

[0067] In the embodiment of the present invention, the contribution of the client may be calculated in a variety of ways.

[0068] For example, in a non-limiting embodiment, the performance of the updated model of each client on the validation set is determined, and the contribution of each client is determined based on the performance. The specific calculation method is as follows: Determine the weight of each sample, and then evaluate the client's performance on the weighted samples. The weight of the validation sample is determined by the prediction power of all clients, and the prediction power is determined by the probability of the prediction result being the true label.

[0069] Assumptions U For the client set, i yes U A client in ; D is the validation set, j yes D A sample of i Correctly predicted samplesj The accuracy is defined as Pi,j , then the sample j The weight is: ; The contribution of each client in a global round can be calculated based on the accuracy of each client's prediction of each sample and the weight of the sample. : ; The client i In round t The contribution in is expressed as: ; For the above client i In round t The contribution value of Standardize and get the client i Contribution to the task : ; Where T is the client i All global rounds participated in.

[0070] For another example, in a non-limiting embodiment, the difference between the updated model parameters of each client and the new global model parameters is calculated, and the contribution of the client is determined based on the difference. The difference in parameters can be determined based on Euclidean distance or Mahalanobis distance.

[0071] For another example, in a non-limiting embodiment, a characteristic function representing the benefits of cooperation is constructed, and the Shapley value of the client is calculated using the characteristic function, and the contribution of the client is determined according to the Shapley value.

[0072] Furthermore, when calculating the contribution of each client, factors such as the quality of each client model update and the speed of task completion may also be comprehensively considered, which is not limited in the embodiment of the present invention.

[0073] Based on the contribution score of each client, considering the budget constraint, it can ensure that the selected client combination can maximize the overall incentive goal while meeting the budget limit. It should be noted that if the budget is limited, clients with higher contribution scores can be prioritized and selected in turn until the budget limit is reached.

[0074] In some embodiments, the historical behavior of the client can also be comprehensively considered for comprehensive scoring. Specifically, determine the contribution of each client; obtain the historical behavior record of each client, and determine the behavior score of each client based on the historical behavior record; calculate the comprehensive score of the client based on the behavior score and contribution of the client; determine the reward of each client based on the comprehensive score of each client and the budget constraint. The historical behavior of the client can be recorded by the central server or by other servers such as the supervision center, which is not limited in this embodiment of the present invention.

[0075] In specific implementations, the reward may be monetary, service, or other forms of incentives.

[0076] It should be noted that in order to ensure the compliance of the model training process and the issuance of rewards, in some embodiments, an independent supervision center can also be set up. The supervision center is responsible for monitoring the entire federated learning process, including the training process, the compliance of the issuance of rewards, and the accountability of the violating entities, to ensure that participants comply with the regulations and prevent violations, such as data leakage or model tampering. Once a violation is discovered, the supervision center can take measures to hold the responsible party accountable. By setting up a supervision center, the standardization and trustworthiness of the system can be enhanced, ensuring that violations can be discovered and handled in a timely manner.

[0077] The regulatory center can supervise the reward distribution process to ensure that the calculation and distribution of rewards are fair and transparent, such as verifying whether the rewards match the client's contribution and checking whether there are any violations. In addition, the regulatory center records the process and results of reward distribution to facilitate subsequent audits and evaluations.

[0078] The reverse auction-based federated learning incentive method provided in the embodiment of the present invention can effectively incentivize client participation and reasonably allocate limited budget resources while ensuring data privacy; moreover, the reverse auction mechanism can effectively improve client resource utilization.

[0079] Furthermore, the use of homomorphic encryption algorithms can ensure the security of data transmission and improve the effect of federated learning.

[0080] Furthermore, by setting up a regulatory center, the transparency and credibility of the system can be increased, ensuring the compliance of the training process and reward distribution. At the same time, illegal entities can be held accountable, ensuring the fairness and legitimacy of the system.

[0081] Accordingly, the present invention also provides a federated learning incentive device based on reverse auction, such as Figure 3 Shown is a structural schematic diagram of the device.

[0082] The reverse auction-based federated learning incentive device 300 includes the following modules: The client selection module 301 is used to determine the client participating in the federated learning through an auction process; A model distribution module 302 is used to initialize a global model and distribute the global model to each client; The receiving module 303 is used to receive the updated model ciphertext sent by each client; A model updating module 304, configured to generate a new global model according to the updated model ciphertext; A reward determination module 305, for determining rewards for each client; The reward distribution module 306 is used to distribute the reward to the corresponding client.

[0083] A non-limiting embodiment of the client selection module 301 may include the following units: An auction initiation unit, used to initiate auctions for federated learning; An information acquisition unit, used to acquire bidding information submitted by multiple clients, wherein the bidding information includes bid information and client information; The screening unit is used to determine the client participating in the federated learning by using a reverse auction algorithm based on the budget constraint and the bidding information.

[0084] In specific implementation, the above-mentioned reward determination module 305 can be implemented in multiple ways.

[0085] For example, in some embodiments, the reward determination module 305 may include: a contribution determination unit and a first reward determination unit. The contribution determination unit is used to determine the contribution of each client; and the first reward determination unit is used to determine the reward of each client according to the contribution.

[0086] For another example, in some embodiments, the reward determination module 305 may include: a contribution determination unit, a behavior scoring unit, a calculation unit, and a second reward determination unit. The contribution determination unit is used to determine the contribution of each client; The behavior scoring unit is used to obtain the historical behavior records of each client and determine the behavior score of each client according to the historical behavior records; The calculation unit calculates the comprehensive score of the client according to the behavior score and contribution of the client; The second reward determination unit is used to determine the reward for each client according to the comprehensive score of each client and the budget constraint.

[0087] The specific implementation methods of the above modules and units can be found in the description of the various embodiments of the reverse auction-based federated learning incentive method of the present invention, which will not be repeated here.

[0088] Accordingly, the present invention also provides a reverse auction-based federated learning incentive system, such as Figure 4 As shown, the system includes: a central server 400, and multiple clients 401; the central server includes the above-mentioned reverse auction-based federated learning incentive device 300.

[0089] Among them, the central server 400 determines the clients participating in federated learning through a reverse auction process; initializes the global model and distributes the global model to each client; receives the updated model ciphertext sent by each client; generates a new global model based on the updated model ciphertext; determines the reward for each client, and distributes the reward to the corresponding client.

[0090] In some embodiments, the reverse auction-based federated learning incentive system 400 provided by the present invention may further include a supervision center (not shown) for monitoring the federated learning process, recording illegal clients, and supervising the compliance of reward issuance.

[0091] Compared with the prior art, one or more embodiments of the reverse auction-based federated learning incentive method, device, and system provided in the embodiments of the present invention have the following beneficial results: 1. The solution of the present invention applies reverse auction to the design of incentive mechanism of federated learning, allowing participants to bid dynamically according to their resources and needs, ensuring the effective allocation of resources under limited budget.

[0092] 2. The use of homomorphic encryption technology ensures the privacy and security of data during transmission and calculation, allowing participants to jointly train models without leaking original data.

[0093] 3. Introduce a regulatory center as an independent supervisory body to monitor the entire federated learning process and ensure the compliance of the training process and reward distribution.

[0094] 4. The accountability mechanism of the regulatory center ensures that violations can be discovered and dealt with in a timely manner, enhancing the standardization and trustworthiness of the system.

[0095] It should be noted that, for the above-mentioned method embodiments, for the sake of simplicity, they are all described as a series of action combinations, but those skilled in the art should know that the present invention is not limited by the described action sequence, because according to the present invention, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present invention.

[0096] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0097] In several embodiments provided by the present invention, it should be understood that the disclosed device can be implemented in other ways.

[0098] The present invention also provides a storage medium, which is a computer-readable storage medium, on which a computer program is stored, and the computer program can be executed when it is run. Figure 1 The storage medium may include a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc. The storage medium may also include a non-volatile memory (non-volatile) or a non-transitory memory, etc.

[0099] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination thereof. When implemented using software, the above embodiments may be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions may be transmitted from one website site, computer, server or data center to another website site, computer, server or data center by wired or wireless means.

[0100] The embodiments of the present invention are described in detail above. The present invention is described in detail using specific implementation methods herein. The description of the above embodiments is only used to help understand the method and system of the present invention. It is only a part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should belong to the scope of protection of the present invention, and the content of this specification should not be understood as limiting the present invention. Therefore, any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A reverse auction-based federated learning incentive method, characterized in that: The method comprises: Identify clients participating in federated learning through a reverse auction process; Initialize a global model and distribute the global model to each client; Receive the updated model ciphertext sent by each client; Generate a new global model according to the updated model ciphertext; Determine the reward for each client, and distribute the reward to the corresponding client; The clients participating in federated learning determined through the reverse auction process include: Launching auctions for federated learning; Acquire bidding information submitted by multiple clients, wherein the bidding information includes bid information and client information; Based on the budget constraint and the bidding information, a reverse auction algorithm is used to determine the client participating in the federated learning; It also includes setting up a supervision center, which is used to monitor the federated learning process, record illegal clients, and supervise the compliance of reward issuance.

2. The reverse auction-based federated learning incentive method according to claim 1, characterized in that: The bidding information is encrypted information; the method further includes: The encrypted information is decrypted using the homomorphically encrypted private key to obtain the corresponding bid information and client information.

3. The reverse auction-based federated learning incentive method according to claim 2, characterized in that: The method of determining the client participating in federated learning by using a reverse auction algorithm based on the budget constraint and the bidding information includes: Determine the evaluation objectives; Determine the evaluation indicators according to the evaluation objectives; Scoring each client participating in the bidding according to the evaluation index, and prioritizing the clients according to the scoring results; Clients with higher priorities are selected in turn and the number of clients participating in federated learning is determined according to the budget constraint.

4. The reverse auction-based federated learning incentive method according to claim 3, characterized in that: The evaluation index includes one or more; multiple evaluation indexes have the same or different weights.

5. The reverse auction-based federated learning incentive method according to claim 4, characterized in that: The multiple evaluation indicators include at least two or more of the following: data quality, computing power, and network bandwidth.

6. The reverse auction-based federated learning incentive method according to claim 1, characterized in that: Generating a new global model according to the updated model ciphertext comprises: Decrypting the update model ciphertext to obtain an update model; Aggregate the updated models of each client to obtain a new global model.

7. The reverse auction-based federated learning incentive method according to claim 1, characterized in that: Determining the reward for each client includes: Determine the contribution of each client; The reward for each client is determined based on the contribution and budget constraints.

8. The reverse auction-based federated learning incentive method according to claim 1, characterized in that: Determining the reward for each client includes: Determine the contribution of each client; Obtaining historical behavior records of each client, and determining a behavior score of each client based on the historical behavior records; Calculate the comprehensive score of the client according to the behavior score and contribution of the client; The rewards for each client are determined based on the client's comprehensive score and budget constraints.

9. The reverse auction-based federated learning incentive method according to claim 7 or 8, characterized in that: Determining the contribution of each client includes: Determine the performance of each client's updated model on the validation set, and determine the contribution of each client based on the performance; or, calculate the difference between each client's updated model parameters and the new global model parameters, and determine the contribution of the client based on the difference; or, construct a characteristic function that characterizes the benefits of cooperation, and use the characteristic function to calculate the Shapley value of the client, and determine the contribution of the client based on the Shapley value.

10. A federated learning incentive device based on reverse auction, characterized in that: The device comprises: The client selection module is used to determine the clients participating in federated learning through an auction process; A model distribution module, used for initializing a global model and distributing the global model to each client; A receiving module, used for receiving the updated model ciphertext sent by each client; A model updating module, used for generating a new global model according to the updated model ciphertext; A reward determination module, used to determine the rewards for each client; A reward distribution module, used to distribute the reward to the corresponding client; The client selection module includes: An auction initiation unit, used to initiate auctions for federated learning; An information acquisition unit, used to acquire bidding information submitted by multiple clients, wherein the bidding information includes bid information and client information; A screening unit, configured to determine a client participating in federated learning by using a reverse auction algorithm based on a budget constraint and the bidding information; The device sends the entire federated learning process information to the supervision center, and the federated learning process information includes the training process, compliance information of reward issuance, and information of illegal clients.

11. The reverse auction-based federated learning incentive device according to claim 10, characterized in that: The reward determination module comprises: A contribution determination unit, used to determine the contribution of each client; The first reward determination unit is used to determine the reward of each client according to the contribution degree.

12. The reverse auction-based federated learning incentive device according to claim 11, characterized in that: The reward determination module comprises: A contribution determination unit, used to determine the contribution of each client; A behavior scoring unit, used to obtain the historical behavior records of each client and determine the behavior score of each client according to the historical behavior records; A calculation unit, which calculates a comprehensive score of the client according to the behavior score and contribution of the client; The second reward determination unit is used to determine the reward for each client according to the comprehensive score of each client and the budget constraint.

13. A reverse auction-based federated learning incentive system, characterized in that: The system includes: a central server and multiple clients; the central server includes the reverse auction-based federated learning incentive device as described in any one of claims 10 to 12.

14. The reverse auction-based federated learning incentive system according to claim 13, characterized in that: The system further comprises: The supervision center is used to monitor the federated learning process, record illegal clients, and supervise the compliance of reward issuance.

15. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the reverse auction-based federated learning incentive method described in any one of claims 1 to 9 are performed.

Images