Model Checking for Adversarial Multi-Agent Reinforcement Learning with Reactive Defense Methods.