DRL and Blockchain Empowered Federated Learning Framework for Genetic Data Engineering

Abstract

AbstractWith the development of genetic data engineering and edge intelligence, more and more intelligent applications and services are trained in the edge side. However, the centralized training mode has the problems of high transmission delay and user privacy disclosure, while federated learning (FL) can protect the privacy of users, and reduce data transmission costs by distributing the training work. Existing FL schemes often ignore the impact of low-quality training nodes and the security issues in the data transmission process. To improve the accuracy of the FL model, we design a node selection algorithm based on deep reinforcement learning (DRL). In addition, we use blockchain for model transmission to complete the global aggregation of FL to enhance the security and reliability of model parameters. We design a blockchain empowered FL framework and further propose a two-layer consensus algorithm based on PBFT to improve consensus efficiency, reduce consensus delay and reduce communication resource consumption. Simulation results show that the proposed node selection algorithm outperforms other compared algorithms, and can well improve the accuracy of the model and reduce the loss function. The proposed consensus algorithm can balance the consensus efficiency and communication resource consumption.KeywordsBlockchainConsensus algorithmDRLGenetic data engineering