Muneef A. Razaz

The evolving landscape of cybersecurity threats, characterized by increasingly sophisticated and adaptive attackers, poses major challenges to traditional static network defense mechanisms. To address these limitations, this paper proposes an adaptive cyber defense framework that integrates Reinforcement Learning (RL) with Attack Graph (AG) modeling. The interaction between attacker and defender is formulated as a repeated zero-sum stochastic game over a partially observable Attack Graph-guided environment, allowing both agents to adapt their strategies through repeated interaction. Two value-based learning approaches are investigated, namely tabular Q-learning and Deep Q-Networks (DQN), under a unified attacker–defender setting. Experimental results across multiple training scenarios show that defender performance improves substantially as the training budget increases. Under limited training, Q-learning provides a computationally efficient and stable baseline, while DQN requires more training and careful tuning to achieve strong performance. However, with extended training, the DQN-based defender attains the highest win rate, albeit at a significantly greater computational cost. In addition, multi-run statistical comparisons highlight a clear trade-off between defensive effectiveness and runtime efficiency: Q-learning remains far more lightweight, whereas DQN offers stronger asymptotic performance when sufficient resources are available. These findings demonstrate the promise of learning-based adaptive defense over attack graphs while also emphasizing the importance of training budget, computational constraints, and model selection in practical cyber defense deployment.

This paper investigates secure and energy-efficient multi-user symbiotic radio (SR) communication assisted by an active reconfigurable intelligent surface (ARIS). We consider a downlink multi-user unicast scenario where a multi-antenna primary transmitter (PTx) serves multiple primary users (PUs), while the ARIS embeds a secondary BPSK symbol into the reflected waveform to deliver sensitive information to a secondary user (SU). Two classes of eavesdroppers are present: EVs-I attempt to intercept the PU messages, whereas EVs-II aim to decode the ARIS–SU link and are typically located in the vicinity of the SU. Our goal is to minimize the total network power consumption, including the PTx transmit power and the ARIS circuit and amplification power, under QoS and secrecy constraints for both primary and secondary transmissions. To account for the lack of reliable instantaneous CSI for the eavesdroppers, we adopt a statistical CSI error model and formulate probabilistic secrecy constraints, which are then converted into tractable convex forms using Bernstein-type inequalities. The resulting non-convex optimization problem is solved via an alternating optimization framework that iteratively updates the PTx beamformers and the ARIS reflection vector, employing a sequential rank-one constraint relaxation (SROCR) to recover feasible rank-one solutions. Numerical results show that the proposed robust ARIS-assisted SR design significantly reduces the total system power consumption compared with PRIS and the considered benchmarks, while meeting the QoS and secrecy-outage constraints under statistical eavesdropper CSI uncertainty.