Sumadhu Rubaiyat (Theory Seminar)

The proposed method bridges Monte Carlo Tree Search (MCTS) with deterministic branch-and-bound techniques and local optimization procedures. To navigate continuous domains, the algorithm recursively partitions the search space into a binary tree. The traditional value function in the Upper Confidence Tree (UCT) is determined through a combination of the best sample and the lower bound computed through interval arithmetic. Through this lens, we can start to understand sample complexity in MCTS settings where value functions can be created to follow a similar structure.