Research On Blockchain-based Decentralized Federated Learning Model And Secure Aggregation Algorithm

With the rapid development of the Internet,the amount of data on the network is growing exponentially,which brings unprecedented development opportunities to artificial intelligence,especially deep learning technology.Deep learning uses large-scale datasets to train complex neural network models,which inevitably include data with privacy attributes,resulting in privacy and security issues.Therefore,how to efficiently share data under the premise of protecting user privacy and data security is of great significance for the development and application of artificial intelligence.Federated learning has been proposed to solve the problem of data sharing.This is an algorithm that uses the data held by all parties to collaboratively train machine learning models on the premise of keeping the data local,which can protect user privacy and data security to a certain extent.However,there are still many problems in the actual application scenarios of federated learning,such as: the central server has a single point of failure and the hidden danger of being attacked;The attacker can infer private information from the parameters or gradients.Federated learning is vulnerable to attacks such as poisoning attack,reconstruction attack and inference attack.The communication efficiency of federated learning is easily affected by network bandwidth.In order to solve the problems in federated learning,this paper studies the decentralized federated learning algorithm and the blockchain-based federated learning model,and also studies the secure aggregation algorithm of federated learning.The main contents and contributions of this paper are as follows:(1)Referring to the relevant algorithm theory in the field of distributed machine learning,Hierarchical Ring All-reduce algorithm and consistent hash algorithm are used to construct a hierarchical ring network topology,and a decentralized federated learning algorithm based on HRA is designed.It can effectively improve the communication efficiency and robustness of decentralization federated learning.At the same time,knowledge distillation and dynamic weight algorithm are introduced to reduce the influence of data poisoning on the global model,and improve the convergence speed and accuracy of the global model.(2)Combining the HRA decentralized federated learning algorithm and the decentralized characteristics of blockchain,the tamper-proof and traceability capabilities of blockchain are used to further enhance the robustness and security of decentralization federated learning.At the same time,the IPFS interplanetary file system is used to provide off-chain storage capabilities for the blockchain,reducing the communication overhead,computing overhead and storage overhead of the blockchain in federated learning.(3)In view of the attack threats faced by the federated learning algorithm,a secure aggregation algorithm based on functional encryption scheme is studied to avoid privacy leakage due to attacks in the aggregation phase of federated learning.At the same time,in order to solve the problem of single point of failure and attack on the server,the decentralized federated learning model based on blockchain and the decentralized function encryption scheme are used to improve the security and robustness of the federated learning system.In summary,aiming at the problems existing in the practical application scenarios of federated learning,this paper studies the decentralized federated learning model and secure aggregation algorithm based on blockchain.The results show that our work improves the communication efficiency,security and robustness of federated learning.