A Distributed Personalized Federated Learning Method based on Siamese Neural Networks

Federated learning allows multiple users to collaboratively train models while protecting data privacy. However, for some users, the non-independent identically distributed nature of user data often reduces the accuracy of the global model. Existing personalized federated learning methods usually focus on individual users, which leads to problems such as bias and overfitting. This paper proposes a new distributed personalized federated learning framework based on Siamese neural networks (DPFL-SNN). First, a novel similarity calculation method is designed using the dual-branch structure of the Siamese neural network to effectively identify local users with similar data. Second, by combining this similarity calculation with blockchain technology, a new consensus algorithm is developed to achieve decentralization and reduce security risks. Simulations conducted on publicly available datasets demonstrate that the DPFL-SNN achieves higher accuracy compared to state-of-the-art personalized federated learning methods, thanks to enhanced collaboration among users with similar data.