ADAPTIVE STRATEGY OPTIMIZATION WITH MULTI-AGENT MACHINE LEARNING IN THE GAME OF RADAR COUNTERMEASURE

Traditional radar countermeasure usually acts with some pre-defined strategies, ignoring the dynamic changes of both sides. In this paper, the scenario of radar countermeasure is represented as a two-player zero-sum dynamic game, where each player adaptively optimizes its own strategy. Specifically, according to game theory, two effective multi-agent machine learning methods, i.e. multi-stage minimax backward induction and deep counterfactual regret minimization are utilized to obtain the final strategies for both players. The experimental results demonstrate that the learned strategies for both players are more effective and reasonable than some simple strategies.