SECC: A Secure and Efficient Communication Channel for Federated Learning System

Federated Learning (FL) successfully provides a privacy-preserving mechanism where participants are not required to share their private datasets to train specific Artificial Intelligence (AI) models. However, FL techniques still face limitations, particularly in the parameter updating mechanism. State-of-the-art methods primarily utilize one method to secure AI model parameter updates, but these still pose drawbacks in certain contexts. In this work, we introduce a Secure and Efficient Communication Channel (SECC) for FL systems. The proposed SECC utilizes various cryptographic methods, such as parameter conversion, digital signatures, and blockchain networks. Evaluations under different levels of poisoning attacks, ranging from 10% to 20%, combined with the cost and time efficiency presented in this work, indicate the superiority of the proposed method compared to traditional FL systems.