Securing decentralized federated learning: cryptographic mechanisms for privacy and trust

In an era where collaborative data analysis and privacy protection are paramount, federated learning emerges as a transformative paradigm. This paper delves into integrating cryptographic techniques to fortify security and privacy in federated learning environments. To protect sensitive data and uphold privacy standards, Attribute-Based Encryption (ABE) and Homomorphic Encryption (HE) offer granular access control and secure computation, respectively. Additionally, secret sharing techniques safeguard model weights by preventing single points of failure. Our approach combines CP-ABE and CKKS encryption, enhancing privacy without sacrificing performance or accuracy, unlike differential privacy and traditional homomorphic encryption methods. To this end, we propose an efficient, secure aggregation method within the Federated Learning framework, using CP-ABE to secure initial global models and CKKS encryption for model parameters. Compared to existing state-of-the-art approaches, our method demonstrates promising results while ensuring data privacy.