Wireless Federated Learning (FL) has become a promising way to do distributed machine learning that lets edge devices work together to train models without sharing raw data. This keeps privacy and cuts down on communication overhead. However, running FL over wireless networks makes it much more vulnerable, especially to coordinated jamming and poisoning attacks. Jamming attacks stop communication between edge devices and the central server, while poisoning attacks add bad updates to make the global model less accurate. This study presents a cohesive defense framework that amalgamates a contribution-based aggregation algorithm, influenced by Shapley values, with a Bayesian game-theoretic distributed power control system. The proposed system makes the system more resistant to bad behavior while still being energy-efficient and reliable for communication. Compared to traditional FL methods, simulation results show that the global model is more accurate, there are fewer communication failures, and resources are used more efficiently.
Dr.K.Rekhadevi et al. (Thu,) studied this question.