Research On The Security Mechanism Of Federated Learning Based On Blockchain

With the rapid development of the Internet industry in recent years,people have entered the era of big data.The continuous mining and utilization of potential value in massive data has promoted the development of artificial intelligence and the progress of human society.However,in this process,data privacy leakage has become a major pain point,and information industries such as social networks and Internet e-commerce have frequently exposed security incidents of user data leakage.The increase in people’s awareness of privacy protection and the promulgation and implementation of laws such as the General Data Protection Regulation have further restricted the access and use of personal data.In fields such as healthcare and the Internet of Things,a large amount of sensitive data is widely distributed in different locations,and there is a need for joint modeling of multiple data entities.However,due to privacy protection considerations,there are barriers that are difficult to break between data sources.The proposal and application of federated learning technology can ensure distributed machine learning without collecting user data to solve the problem of data islands.However,traditional federated learning relies on a central server for model storage and parameter aggregation.The centralized model training process makes the system vulnerable to single-point failure and malicious node attacks;plaintext intermediate parameters passed by federated learning participants can also be used to infer the private information in data,federated learning is increasingly challenged in system security and privacy protection.In addition,federated learning does not provide a suitable incentive mechanism for participants to contribute more training data and computing resources.In order to solve the above challenges,this paper proposes a decentralized,secure and fair federated learning model based on blockchain,which uses homomorphic encryption technology to protect the privacy of the intermediate parameters of the collaborative training parties,and conducts model aggregation and collaborative decryption through an elected federated learning committee.The decryption process achieves secure key management through a secret sharing scheme,and utilizes a bilinear map accumulator to provide correctness verification for the secret share.The model also introduces reputation as an indicator for evaluating the reliability of the participants,and uses the subjective logic model to realize distrust-enhanced reputation calculation as the basis for the election of the federated learning committee.The model information and reputation value realize data tamper-proof and non-repudiation through the blockchain.The blockchain-based privacy-preserving federated learning model proposed in this paper is implemented using the FISCO BCOS blockchain platform and performs collaborative learning on the MNIST dataset.Experiments show that the proposed federated learning model can achieve a decentralized model training method with a slight loss of training accuracy compared to the centralized learning model,and effectively realize the privacy protection for each participant.