TMAG-FED: 안전한 연합학습을 위한 2-패스 상호인증 가십 프로토콜.

In the realm of artificial intelligence, Federated Learning(FL) has emerged as a promising technique. FL enables AI models to be trained collaboratively without compromising privacy by directly sharing sensitive data. Specifically, FL based on the Gossip protocol in peer-to-peer networks can enhance efficiency by leveraging the local data of each client and optimizing traffic distribution. However, the reliability of FL is not always guaranteed, leading to potential security risks such as data leakage. To address these challenges, this paper proposes a secure push-pull hybrid Gossip protocol based on two-pass mutual authentication. The protocol ensures mutual authentication, integrity, and confidentiality by generating and sharing NONCE values with the receiver in a single push and pull process. This fosters trust among clients in the peer-to-peer network environment, enabling them to generate accurate aggregated models Wagg based on decrypted model updates Ws. This research contributes to the development of advanced FL strategies that prioritize efficient communication, security, and accuracy in peer-to-peer network environments. [ABSTRACT FROM AUTHOR]