A Decentralized Asynchronous Federated Learning Framework Based on Blockchain Technology

Federated learning is an innovative distributed machine learning technology. Each participant saves private data locally and only sends parameter information during model training to the aggregation server without directly sharing the original data so as to ensure data security and privacy. However, in the process of model training, a single aggregation server is vulnerable to poisoning attacks or single points of failure, and malicious server adversaries can also have a substantial negative impact on the model's prediction. Therefore, in this paper, we propose a decentralized asynchronous federated learning framework based on blockchain technology, which uses blockchain to replace the parameter server in traditional federated learning, and uses a homomorphic encryption algorithm to encrypt the model gradient and other parameters in the training process. In addition, this paper also proposes a model performance verification mechanism to ensure the accuracy of model training. After experimental verification, the framework achieves better training results on the MNIST dataset and CIFAR-10 dataset and effectively avoids malicious attacks from the central server.