Adversarial Online Learning with Variable Plays in the Pursuit-Evasion Game: Theoretical Foundations and Application in Connected and Automated Vehicle Cybersecurity

We extend the adversarial/non-stochastic multi-play multi-armed bandit (MPMAB) to case where number of arms play is variable. The work motivated by fact that resources allocated scan different critical locations in an interconnected transportation system change dynamically over time and depending on environment. By modeling malicious hacker intrusion monitoring as attacker defender, respectively, we formulate problem for two players a sequential pursuit-evasion game. derive condition under which Nash equilibrium strategic game exists. For defender side, provide exponential-weighted based algorithm with sublinear pseudo-regret. further our model heterogeneous rewards both players, obtain lower upper bounds average reward attacker. numerical experiments demonstrate effectiveness variable-arm play.