Anuraag Shankar

This project studies decision problems where an outer choice's value depends on a hard inner game involving hidden information, simultaneous moves, and stochasticity. This differs from the simpler inner problems usually assumed in existing bilevel and Stackelberg optimization research. The project begins by reviewing prior work on bilevel frameworks, equilibrium-finding in complex stochastic games, and AI systems for related games, to identify what has been done and where gaps remain. It then formalizes this problem class and estimates its complexity, comparing it to previously solved benchmark games. The aim is to to show that it combines properties that no existing solved game or method fully addresses. The project also considers whether the problem could simply be solved using existing techniques, and explains why current evidence suggests otherwise. A concrete application domain will be used to test these ideas in practice, though the exact domain is not yet fixed.