Regret-Based Optimization and Preference Elicitation for Stackelberg Security Games with Uncertainty