PROJECT TITLE :

Learning in Hide-and-Seek

ABSTRACT:

Existing work on pursuit-evasion issues usually either assumes stationary or heuristic behavior of 1 aspect and examines countermeasures of the other, or assumes each sides to be strategic that leads to a game theoretical framework. Results from the previous typically lack robustness against changes in the adversarial behavior, whereas those from the second class, usually as equilibrium answer concepts, might be difficult to justify: either thanks to the implied knowledge of other players' actions/beliefs and knowledge of their information, or due to a scarcity of economical dynamics to realize such equilibria. In this paper, we tend to take a different approach by assuming an intelligent pursuer/evader that's adaptive to the data accessible to it and is capable of learning over time with performance guarantee. Inside this context we have a tendency to investigate two cases. In the first case we assume either the evader or the pursuer is tuned in to the type of learning algorithm utilized by the opponent, while within the second case neither facet has such data and thus must try to find out. We tend to show that the optimal policies in the primary case have a greedy nature. This result is then used to assess the performance of the educational algorithms that each sides employ in the second case, which is shown to be mutually optimal and there is no loss for either side compared to the case when it is aware of perfectly the adaptive pattern employed by the adversary and responses optimally. We further extend our model to check the applying of jamming defense.