Jamming Attack Mitigation in Wireless Federated Learning Networks Using Bayesian Games

Federated Learning (FL), an emerging distributed Artificial Intelligence (AI) technique, is susceptible to jamming attacks during the wireless transmission of trained models. In this letter, we introduce a jamming attack mitigation mechanism for the uplink of wireless FL networks using the power-domain Non-Orthogonal Multiple Access (NOMA) technique. The problem of transmission power allocation for all clients (legitimate and malicious) is formulated and solved distributively as a Bayesian game with incomplete information. The clients aim to successfully transmit their model parameters, minimizing transmission time and consumed power, while having probabilistic knowledge about the malicious behavior of the other clients in the game.