A blockchain-based federated learning method and device that can be trusted by participants
Through blockchain technology, the server centrality is reduced, the multi-party mutual trust and data security problems in federated learning are solved, the robustness and trust of the model are improved, and higher quality model training is achieved.
Patent Information
- Application Number
- CN202110552254.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-05-20
- Publication Date
- 2024-12-27
- Estimated Expiration
- 2041-05-20
AI Technical Summary
There are multi-party mutual trust problems, data security risks and server centrality in federated learning, which affect the reliability and quality of model training.
Through blockchain technology, participants’ identity registration and authentication are carried out, untrusted participants are screened, and trust mechanisms are used to establish a trust mechanism, reduce server centrality, and optimize the parameters of the supernode aggregation model.
It improves the robustness and trust of the model, reduces the impact of server failure, enhances data security and model quality, and inspires participants to provide high-quality data.
Smart Images
Figure CN113467927B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of federated learning, and specifically relates to a blockchain-based federated learning method and device that is trustworthy to participants. Background Art
[0002] In recent years, artificial intelligence has gradually penetrated into people's lives, such as liveness detection, unmanned driving, face recognition, etc. It is well known that a good artificial intelligence needs a large amount of high-quality data for training to meet the needs. However, in actual scenarios, only a few industry giants can have a large amount of high-quality training data, while the vast majority of companies can only have less or lower-quality data, which often makes it difficult to support the realization of good artificial intelligence technology. In addition, since commercial data often has high commercial value and privacy, companies pay more and more attention to data confidentiality; domestic and foreign measures and policies for data protection are also gradually improving. The above reasons have led to the isolation of data from all parties, forming the so-called data islands, which are not conducive to the realization of large-scale artificial intelligence technology.
[0003] In order to break this deadlock, Google proposed a distributed machine learning technology, namely federated learning, in 2016. It mainly solves the problems of data privacy protection and data silos by adopting technologies such as differential privacy, homomorphic encryption, and multi-party secure computing, so that all parties involved in the training can benefit, thereby achieving data availability without visibility while protecting local data, and promoting the development of technology. However, federated learning also faces some challenges, including: (1) the problem of mutual trust among multiple parties: Since there is a lack of trust between the participants, federated learning needs to run collaboratively without trust between the parties, which may affect the reliability of model training; (2) security risks: the quality of data provided by each participant is uneven, which may affect the quality of the entire model training; (3) the reliability of the communication link: the parameter data uploaded by each party is vulnerable to attack or tampering during transmission, which may affect the training effect of the final model. Therefore, how to make the training participants in federated learning trustworthy and improve the security and trustworthiness of the federated learning model is a key issue.
[0004] In addition, since the model relies on a single central server, when the server model fails, the model updates of other training participants will also be affected. Therefore, how to weaken the centrality of the server in the federated learning framework to reduce the impact of equipment failure is also a top priority.
[0005] Blockchain technology is a security technology that can handle issues between multiple untrusted parties. It has the characteristics of anonymity, immutability, and decentralization. Its technical characteristics such as identity management, distributed ledger, data recording, incentive mechanism, and consensus mechanism make blockchain suitable for scenarios where there is a need for decentralization, multi-party participation and data writing, high requirements for data authenticity, and the need to establish distributed trust among multiple participants who do not trust each other initially. Therefore, blockchain technology can complement federated learning well. Summary of the invention
[0006] In view of the above, the purpose of the present invention is to provide a blockchain-based federated learning system in which participants can trust each other, thereby improving the robustness of the model by screening out untrustworthy participants.
[0007] In a first aspect, an embodiment of the present invention provides a blockchain-based federated learning method that is trustworthy to participants, comprising the following steps:
[0008] Participants’ identities are registered and authenticated, smart contracts are established, and participants are bound to computing devices in the blockchain;
[0009] The participants upload local model parameters and historical credit ratings to the bound computing device in encrypted form. The computing device screens and verifies the participants based on their historical credit ratings and filters out untrustworthy participants for the first time. It then makes a global similarity judgment on the local model parameters of the remaining participants to filter out untrustworthy participants for the second time. It then updates the historical credit ratings of the remaining trusted participants based on global similarity and issues rewards to trusted participants in proportion to their new historical credit ratings.
[0010] Voting to select super nodes based on the computing power of the computing device or the historical credit rating of the computing device;
[0011] The super node aggregates the local model parameters of the trusted participants to obtain the global model parameters, and after performing a parameter optimization based on the global model parameters, the optimized global model parameters are sent to each trusted participant to complete a round of federated learning.
[0012] In a second aspect, an embodiment of the present invention provides a blockchain-based federated learning device that is trusted by participants, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the blockchain-based federated learning method that is trusted by participants as described in the first aspect when executing the computer program.
[0013] The beneficial effects of the blockchain-based federated learning method and device provided in the embodiment that can be trusted by the participants are mainly manifested in: (1) By utilizing the decentralized characteristics of the blockchain, the server model function of the federated learning is given to the super node generated by the computing device, thereby weakening the centrality of the server model and reducing the global impact caused by server failure. (2) By utilizing the accounting function and immutability of the blockchain, a trust mechanism is established to enhance the trust between the training participants and improve the quality of the joint training model. (3) Through the blockchain reward mechanism, the training participants are encouraged to provide high-quality model data to improve the effect of joint training. BRIEF DESCRIPTION OF THE DRAWINGS
[0014] 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 or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.
[0015] Figure 1 It is a flowchart of a blockchain-based federated learning method that is trustworthy to participants in one embodiment. DETAILED DESCRIPTION
[0016] To make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific implementation methods described herein are only used to explain the present invention and do not limit the scope of protection of the present invention.
[0017] In view of the problems of distrust among training participants, data security risks and unreliable communication links in the federated learning framework, the embodiments utilize security mechanisms such as the immutability of blockchain to provide a trusted federated learning method and device based on blockchain. By introducing the decentralized characteristics of blockchain technology, the centralization of server models in federated learning can be reduced, reducing the potential risk of performance degradation of the global model due to equipment failure. By utilizing the accounting function and immutability of blockchain technology, a trust mechanism is introduced to perform credit scoring for each participant based on the quality of model parameters uploaded by the participant, improve their own credit according to the individual credit score, improve the training quality of the model, and enhance the user's trust in the model. Reliable super nodes are selected according to the individual credit score and the credit score of the calculation group to aggregate the uploaded local model parameters. The participants are encouraged to upload high-quality parameters through a credit score-based reward mechanism, thereby improving the effect of model joint training.
[0018] Figure 1FIG. 1 is a flow chart of a blockchain-based federated learning method that is trustworthy to participants in one embodiment. Figure 1 As shown, the blockchain-based federated learning method provided by the embodiment and trusted by the participants includes the following steps:
[0019] S101, register and authenticate the identities of participants, establish smart contracts, and bind participants to computing devices in the blockchain.
[0020] The task of federated learning is mainly that multiple participants distributed at the edge use local samples to train the same model structure to optimize the model parameters, and then the center aggregates and optimizes the model parameters of all edge ends to conduct federated learning. In the field of telecommunications, this federated learning can be used to aggregate the model updates of smart devices, and integrate and generate public AI models under the premise of protecting data privacy to improve the generalization ability of the model; in the financial field, financial institution partners jointly build anti-money laundering models, use their own anti-money laundering samples for local training, and optimize a common anti-money laundering model without leaking local data; in the field of medical health, each hospital uses its own patient medical records for local model training, and aggregates the parameters of all parties to jointly train a more effective disease prediction model. In the embodiment, before federated learning, preparation work needs to be done, specifically including: the participants (clients) at the edge of the participating federated learning need to register their identities and conscientiously, establish smart contracts, bind the participants and the computing devices in the blockchain, and after the preparation work is completed, the federated learning process is carried out.
[0021] In the embodiment, the participant identity registration and authentication includes: the participant uploads identity information, and uses the identity ID to verify the identity of the participant, and the initial credit rating of the initial participant. During the participant identity registration and authentication process, the participant will upload the identity information to the administrator, where the identity information includes the identity ID, computing power, and transmission capacity. The transmission capacity generally refers to the transmission rate, the size of the transmitted data, etc. After receiving the identity information, the administrator verifies the identity of the participant based on the participant's identity ID. Only the participants who pass the identity verification can participate in federated learning. The administrator also initializes the initial credit rating for each participant who passes the identity authentication. For example, the initial credit rating can be 0.
[0022] In the embodiment, the established smart contract includes a digital signature encryption method, model header information, a method for selecting and associating computing devices in the case of multiple participants, and a reward and punishment mechanism. Among them, the digital signature encryption method constrains the method of digitally signing and encrypting model parameters and model header information. The digital signature encryption method agreed in the smart contract is adopted between the participants participating in federated learning and the bound computing devices, but the parameters used in the signature process can be defined by the participants themselves and transmitted to the bound computing devices for notification. The model header information is mainly model structure information, and the model header information is used as the content of digital signature authentication to ensure that the model parameters uploaded subsequently correspond to the model structure to avoid errors. The method of selecting and associating computing devices in the case of multiple participants stipulates the rules and methods for binding between the participants and the computing devices with computing capabilities in the blockchain, specifically including selecting the node with the strongest computing power as the computing device or randomly selecting a computing device from the participants. The reward mechanism constrains the way participants are rewarded. Specifically, trusted participants with great contributions are rewarded, and untrusted participants who fail to complete on time or upload malicious data or are attacked are punished. This encourages participants to upload high-quality model parameters and discard malicious or poor-quality model parameter data.
[0023] In the embodiment, in order to use blockchain for federated learning, it is necessary to bind the participants to the computing devices in the blockchain. The binding of the participants to the computing devices in the blockchain includes: according to the method of selecting and associating computing devices agreed in the smart contract, binding and associating 1 computing device for each participant, wherein 1 computing device is bound to at least 1 participant, and when 1 computing device is bound to multiple participants, the multiple participants will form a computing group. The bound computing device mainly completes the collection function and judgment and filtering function of local model parameters.
[0024] S102, the participating party encrypts and uploads local model parameters and historical credit ratings to the bound computing device.
[0025] The participants use local data to optimize the model parameters to obtain local model parameters, and encrypt the local model parameters and upload them to the bound computing device. In one embodiment, the participants encrypt and upload the local model parameters to the bound computing device, including:
[0026] The participants encrypt the model header information to form a first signature according to the digital signature encryption method agreed in the smart contract, encrypt the local model parameters to form a ciphertext, and then upload the first signature and the encryption parameters used in the digital signature encryption method to the bound computing device;
[0027] After receiving the first signature and the encryption parameters used in the digital signature encryption method, the computing device uses the encryption parameters used in the digital signature encryption method to encrypt the model header information agreed upon in the smart contract to obtain a second signature. When the second signature is equal to the first signature, the signature verification is passed, and the secure transmission of the ciphertext is achieved.
[0028] In the embodiment, when uploading the local model parameters, the participants use sign=m e modr encrypts the model header information m according to the parameters e and r to form a first signature sign, and uploads the first signature sign, the parameters e and r to the computing device, which uses sign'=m according to the received parameters e and r. e modr signs the model header information m for the second time to obtain the second signature sign'. When sign'=sign, the signature passes, the signature verification is completed, and then the secure transmission of the link ciphertext data is realized. Among them, the ciphertext data C=M e modr, where M is the local model parameter.
[0029] S103, the participant uploads the historical credit rating to the bound computing device, and the computing device screens and verifies the participant based on the historical credit rating and filters out untrustworthy participants for the first time.
[0030] At the same time, in order to ensure the quality of uploaded local model parameters and improve the trustworthiness of participants, participants need to upload historical credit scores when uploading local model parameters. When receiving historical credit scores, the computing device verifies the credit of the participants and filters out participants with poor credit. In one embodiment, screening and verifying based on the historical credit ratings of participants and filtering out untrustworthy participants for the first time includes:
[0031] After receiving the uploaded historical credit rating, the computing device searches the historical credit rating recorded in the historical ledger. When the two historical credit ratings are not equal, the participant is directly judged to be untrustworthy and the untrustworthy participant is filtered out for the first time.
[0032] In one embodiment, the process of filtering out untrustworthy participants based on historical credit ratings is implemented before signature verification, and signature verification is performed on the remaining participants. After the signature verification is passed, secure transmission of the ciphertext is achieved. This method of filtering out untrustworthy participants before uploading local model parameters can greatly save communication resources and communication consumption.
[0033] When uploading historical credit ratings, participants also upload parameter validity proof P, where the validity proof P can be the difference between the task effect of this training and the previous training. The validity proof P is also used to filter out untrustworthy participants for the second time.
[0034] S104, performing global similarity judgment on the local model parameters of the remaining participants to filter out untrustworthy participants for the second time and obtain trustworthy participants.
[0035] In one embodiment, the global similarity judgment of the local model parameters of the remaining participants to filter out untrustworthy participants for the second time includes:
[0036] When the number of iterations is 1, the average of all local model parameters uploaded by the remaining participants is used as the benchmark parameter;
[0037] When the number of iterations is greater than 1, the corresponding participants with the parameter validity certificates with a high preset ratio before screening are selected based on the parameter validity certificates uploaded by the remaining participants; and the average value of the local model parameters of the screened participants is used as the benchmark parameter;
[0038] In each round, the global similarity between each remaining participant and the benchmark parameter is calculated, and the participants with global similarity less than the threshold are filtered out for the second time to obtain the trusted participants.
[0039] In the embodiment, for n participants who have completed uploading the local model parameter data, when the number of iterations is 1, the local model parameter M of the n participants is used as i The mean of t ,Right now
[0040]
[0041] The preset ratio is mainly used as a dividing line for screening high-quality local model parameters, and can be preferably 1 / 2. When the number of iterations is greater than 1, the local model parameter validity proofs P uploaded by each participant will be sorted from high to low, and the top 1 / 2 of the local model parameters will be taken to form a high-quality parameter set M. h , for the high-quality parameter set M h The average of all local model parameters in is taken as the reference parameter μ t ,Right now
[0042]
[0043] In each round, the global similarity d between the participant and the benchmark parameter is calculated as:
[0044]
[0045] Among them, the dimensions of the benchmark parameter μ and the local model parameter M are the same, both of which are K×L, and k and l are the indexes of the elements in the benchmark parameter μ and the local model parameter M. Then, the participants whose local model parameter similarity is less than the threshold or the lowest are considered to be the participants who may be attacked or malicious. When the computing device collects the local model parameters of each participant, it discards the local model parameters, and the remaining participants are trusted participants, and the corresponding local model parameters are used for aggregation.
[0046] S105, updating the historical credit rating of the trusted parties based on the global similarity.
[0047] After obtaining the trusted party, the computing device queries the historical ledger again to obtain the historical credit rating of the trusted party, and then updates the historical credit rating of the trusted party based on the global similarity, including:
[0048] The historical credit rating is updated using the following formula:
[0049]
[0050] Among them, represents the global similarity, K t-1 , K t Represent the historical credit ratings of time t-1 and time t respectively.
[0051] S106: Rewards are distributed to credible participants in proportion to the new historical credit rating.
[0052] In the embodiment, rewards are also issued to trusted parameters and parties. The computing device runs the proof-of-work mechanism POW. When the computing device completes POW, a new block is generated, which records the local model parameter updates uploaded by the credit-verified participants and the participants' historical credit score updates. The computing device obtains rewards and issues rewards to each participant based on the historical credit rating of each participant in the computing device:
[0053]
[0054] Among them, R total is the total reward value, R i The reward obtained by the i-th participant is used to encourage each participant to upload better local model parameter data.
[0055] S107, voting to select a super node based on the computing power of the computing device or the historical credit rating of the computing device.
[0056] In the embodiment, the generation of super nodes weakens the centrality of the server model in the federated learning framework, which can effectively reduce the global impact caused by server model failure. Even if a super node fails due to a failure, a new super node can be screened and generated from the remaining computing devices, so that the federated model can continue to train. The selection of super nodes can be completed in two ways:
[0057] Method 1: The publishing node of the computing power test question publishes the computing power test question to the computing power pool, and the candidate computing devices obtain the test question from the computing power pool to solve it and return the solution result. The publishing node of the computing power test question publishes the answer and selects the candidate computing device with the closest answer as the super node.
[0058] Method 2: Based on the historical credit rating of the computing device, it is generated by voting among the computing devices.
[0059] S108, the super node aggregates the local model parameters of the trusted participants to obtain the global model parameters, and after performing a parameter optimization based on the global model parameters, sends the optimized global model parameters to each trusted participant to complete a round of federated learning.
[0060] In one embodiment, the super node aggregates the local model parameters of the trusted participants to obtain the global model parameters, including:
[0061] Calculate the average of the historical credit ratings of all participants in each computing group as the trustworthiness of the computing group;
[0062] The local model parameters of the computing group are weighted by the proportion of the trustworthiness of each computing group, so as to obtain the weighted local model parameters of the computing group;
[0063] The weighted local model parameters of all computing groups are averaged or weightedly aggregated to obtain the global model parameters.
[0064] In the embodiment, the historical account book records of the computing group are queried to determine the trustworthiness of the computing group. That is, the super node first queries the historical credit scores of each participant of each computing device, and calculates the trustworthiness of each computing group. Before aggregating local model parameters, assume that there are m computing groups. The model parameters M uploaded by the computing groups are aggregated according to the group trustworthiness. j Weighted j is the index of the computation group.
[0065] When the super node receives the weighted local model parameter data M provided by all computing groups within a given time, j 'After that, the local model parameter M j'Decrypt to obtain the real local model parameters, and aggregate the decrypted local model parameters to obtain the global model. The aggregation operation here can adopt the aggregation method commonly used in federated learning, such as average aggregation or weighted aggregation Among them, w j As trainable parameters, the local model parameters uploaded by each participant can be adaptively weighted.
[0066] After obtaining the global model, the super node trains the global model for the first time using the gradient descent method, etc. After a parameter optimization, the optimized global model parameters, the historical credit scores of each computing group, and the historical credit scores of the participants are packaged and stored in the block, broadcast, bookkeeping, etc., and the updated block is added to the blockchain. Each participant can download relevant data from it to train the next round of local models. Repeat the above process, and when the iteration end condition is met, the federated model achieves the expected effect and the training ends.
[0067] The above-mentioned embodiment provides a blockchain-based federated learning method that is trustworthy to participants. On the one hand, it uses the decentralization of blockchain to weaken the centrality of the server model in the federated learning framework, which can reduce the global impact caused by equipment failure; on the other hand, it uses blockchain ledger tools and the characteristics of non-tamperability to establish a trust mechanism, enhance the trust between the training participants, and improve the quality of the joint training model. On the other hand, through the blockchain reward mechanism, it encourages the training participants to provide high-quality model data and improves the effect of joint training.
[0068] The embodiment also provides a blockchain-based federated learning device that is trusted by the participants, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the above-mentioned blockchain-based federated learning method that is trusted by the participants is implemented.
[0069] In specific applications, the memory can be a volatile memory at the near end, such as RAM, or a non-volatile memory, such as ROM, FLASH, floppy disk, mechanical hard disk, etc., or a remote storage cloud. The processor can be a central processing unit (CPU), a microprocessor (MPU), a digital signal processor (DSP), or a field programmable gate array (FPGA), that is, these processors can be used to implement the steps of federated learning based on blockchain that can be trusted by participants.
[0070] The specific implementation methods described above provide a detailed description of the technical solutions and beneficial effects of the present invention. It should be understood that the above is only the most preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, supplements and equivalent substitutions made within the scope of the principles of the present invention should be included in the protection scope of the present invention.
Claims
1. A blockchain-based federated learning method that is trustworthy to all participants, characterized in that: The following steps are involved: Participants’ identities are registered and authenticated, smart contracts are established, and participants are bound to computing devices in the blockchain; The participants upload the local model parameters and historical credit ratings to the bound computing device in encrypted form. The computing device screens and verifies the participants based on their historical credit ratings and filters out untrustworthy participants for the first time. It then performs a global similarity judgment on the local model parameters of the remaining participants to filter out untrustworthy participants for the second time. It then uses the following formula to update the historical credit ratings of the remaining trusted participants based on global similarity, and issues rewards to trusted participants in proportion to their new historical credit ratings. ； in, d represents the global similarity, K t-1 , K t Represent the historical credit ratings of times t-1 and t respectively; Voting to select super nodes based on the computing power of the computing device or the historical credit rating of the computing device; The super node aggregates the local model parameters of the trusted participants to obtain the global model parameters, and after performing a parameter optimization based on the global model parameters, the optimized global model parameters are sent to each trusted participant to complete a round of federated learning.
2. The method for federated learning based on blockchain and trusted by participants as claimed in claim 1, characterized in that: The participant identity registration and authentication includes: the participant uploads identity information, and uses the identity ID to verify the identity of the participant, and the initial credit rating of the initial participant; the identity information includes identity ID, computing capacity, and transmission capacity; The smart contract includes digital signature encryption method, model header information, method of selecting and associating computing devices in the case of multiple parties, and reward and punishment mechanism.
3. The method for federated learning based on blockchain and trusted by participants as claimed in claim 2, characterized in that: The binding of the participants and the computing devices in the blockchain includes: according to the method of selecting and associating computing devices agreed in the smart contract, binding and associating one computing device for each participant, wherein one computing device is bound to at least one participant, and when one computing device is bound to multiple participants, the multiple participants will form a computing group.
4. The method for federated learning based on blockchain and trusted by participants as claimed in claim 1, characterized in that: Participants encrypt and upload local model parameters to bound computing devices including: The participants encrypt the model header information to form a first signature according to the digital signature encryption method agreed in the smart contract, encrypt the local model parameters to form a ciphertext, and then upload the first signature and the encryption parameters used in the digital signature encryption method to the bound computing device; After receiving the first signature and the encryption parameters used in the digital signature encryption method, the computing device uses the encryption parameters used in the digital signature encryption method to encrypt the model header information agreed upon in the smart contract to obtain a second signature. When the second signature is equal to the first signature, the signature verification is passed, and the secure transmission of the ciphertext is achieved.
5. The method for federated learning based on blockchain and trusted by participants as claimed in claim 1, characterized in that: The computing device screens and verifies the participants based on their historical credit ratings and filters out untrustworthy participants for the first time, including: After receiving the uploaded historical credit rating, the computing device searches the historical credit rating recorded in the historical ledger. When the two historical credit ratings are not equal, the participant is directly judged to be untrustworthy and the untrustworthy participant is filtered out for the first time.
6. The method for federated learning based on blockchain and trusted by participants as claimed in claim 4, characterized in that: The process of filtering out untrustworthy participants based on historical credit ratings is implemented before signature verification. The signatures of the remaining participants are verified. After the signature verification is passed, the secure transmission of the ciphertext is achieved.
7. The method for federated learning based on blockchain and trusted by participants as claimed in claim 1, characterized in that: The performing global similarity judgment on the local model parameters of the remaining participants to filter out untrustworthy participants for the second time includes: When the number of iterations is 1, the average of all local model parameters uploaded by the remaining participants is used as the benchmark parameter; When the number of iterations is greater than 1, the corresponding participants with the parameter validity certificates with a high preset ratio before screening are selected based on the parameter validity certificates uploaded by the remaining participants; and the average value of the local model parameters of the screened participants is used as the benchmark parameter; In each round, the global similarity between each remaining participant and the benchmark parameter is calculated, and the participants with global similarity less than the threshold are filtered out for the second time to obtain the trusted participants.
8. The method for federated learning based on blockchain and trusted by participants as claimed in claim 1, characterized in that: The super node aggregates the local model parameters of the trusted participants to obtain the global model parameters including: Calculate the average of the historical credit ratings of all participants in each computing group as the trustworthiness of the computing group; The local model parameters of the computing group are weighted by the proportion of the trustworthiness of each computing group, so as to obtain the weighted local model parameters of the computing group; The weighted local model parameters of all computing groups are averaged or weightedly aggregated to obtain the global model parameters.
9. A blockchain-based federated learning device that is trusted by participants, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the blockchain-based federated learning method that is trusted by participants as described in any one of claims 1 to 8 is implemented.