A Secured data exchange mechanism for Federated Learning using blockchain technology

Federated learning is a new distributed network training technique that is gaining popularity in the privacy protection space for its ability to safeguard users' local data during training. Unfortunately, there are still a lot of security issues with current federated learning (FL) models. For example, (1) still there exist attackers to stealing private data via gradients, and (2) when nodes crash, learning the model stops. This work presents a blockchain-based safe federated learning model to solve these issues. The approach is based on the following fundamental ideas: (1) construct a decentralized FL training model that enhances the security of the model training environment by integrating blockchain networks; and (2) establish a system for user failure reaction. The concept improves the decentralized secure network environment, optimizes the compute cost of encryption, and ensures correctness while protecting privacy. Lastly, the suggested model demonstrates practically excellent performance according to experimental validation findings on the CIFAR-10 dataset. On an average, in terms of key generation time the proposed model outperforms existing methods with 8.9%, On transition time the proposed model outperforms 11.1% compared to existing methods. And with respect to aggregation time the proposed model outperforms 6.7% compared to existing methods.