Decentralized Federated Learning Algorithm And Framework Based On Blockchain

In recent years,people have paid more attention to data privacy protection and the government hash gradually begun to formulate a series of regulations in the field of data privacy protection to protect the security of data privacy.In Europe and America,some public data sets involving sensitive personal information are no longer allowed to be used again.It is conceivable that it will become more difficult for individuals or companies to obtain user privacy data to study artificial intelligence in the near future.At present,the conventional deep learning algorithms in artificial intelligence are all data visible,that is,data must be obtained to use these deep learning algorithms,which poses a great threat to data privacy.Therefore Google first proposed the concept of federated learning in 2016.However,conventional federated learning is a centralized architecture which faces many problems in real scenarios.For example,the security and robustness of conventional federated learning under malicious node attack,the pressure of network communication,the utilization of network bandwidth and the malicious data set poisoning,etc.Therefore,this paper solves the problems caused by centralized federated learning from algorithms and engineering.In terms of algorithms,this paper proposes a ring decentralized federated learning algorithm(RDFL)based on consistent hashing algorithm,knowledge distillation and Ring-allreduce algorithm,which improves network bandwidth utilization and robustness.The RDFL algorithm improves network bandwidth utilization and reduces the impact of malicious data sets.At the same time,the RDFL algorithm improves the accuracy of federated learning on non-independent and identically distributed(Non-IID)data sets.In terms of engineering,this paper proposes a decentralized federated learning framework based on blockchain and RDFL algorithms called GFL.GFL reduces network communication pressure and improves the security and robustness of decentralized federated learning.