AI-driven Wireless Attack Detection and Escape (AI-WADE)

Spectrum battlefield is characterized by a complex multi-domain environment with fast and unknown dynamics. Wireless channel, network topology, user mobility and traffic, and interference effects, which includes in-network and out-network interference as well as adversarial jamming, changes over time. Conventional statistical and rule-based methods cannot be effectively applied for attack detection and prediction due to the complexity of the spectrum environment. Deep Learning (DL), particularly Deep Reinforcement Learning (RL) has emerged as a powerful means for spectrum awareness by learning from and adapting to spectrum dynamics. However, DL models can forget the behavior and the appropriate responses to a known threats as they learn to adapt to a new unseen type of wireless threat. Continual learning provides the capability to continuously learn and acquire new skills building on previously acquired knowledge. The goal of AI-WADE is to develop a wireless network security solution with continual RL capabilities that can learn on-the-fly the appropriate response to a jammer and does not suffer from catastrophic forgetting.