联邦学习双向匿名身份认证及会话密钥协商协议OA

Using federated learning as a novel distributed machine learning paradigm effectively solves problems of secure cross-domain data sharing.However,its distributed architecture makes the system be faced with various threats of privacy and security,such as model inversion and membership inference attacks.This paper proposes a lightweight federated learning protocol for bidirectional anonymous au-thentication and session key negotiation based on the Chinese remainder theorem(CRT),named FL-LCPPA.The protocol leverages CRT-based encryption to distribute valid pseudoidentities to legitimate nodes,enabling bidirectional anonymities to authenticate between server and client,and ensuring the le-gitimacy and authenticity of participants in federated learning.The protocol negotiation session key used for encrypting model parameters is accomplished by integrating elliptic curve cryptography(ECC)and hash functions,ensuring the security and reliability of communication parameters and enhancing the o-verall privacy protection of the system.The experimental result demonstrates that FL-LCPPA can resist various attacks,including replay and man-in-the-middle attacks.Only by meeting with the security needs of confidentiality,integrity,reliability,and anonymity can the computational overheads obviously reduce.The protocol comparatively balances security with efficiency,making it well-suited for federated learning scenarios.