University of Tokyo

Adapting Improved Upper Confidence Bounds for Monte-Carlo Tree Search

This research introduces the Mi-UCT algorithm, a novel variant of Monte-Carlo Tree Search (MCTS) that integrates a modified version of the improved Upper Confidence Bound (UCB) algorithm.

• Mi-UCT is a new Monte-Carlo Tree Search variant that adapts the improved Upper Confidence Bound (UCB) algorithm to provide tighter regret bounds in game tree searches.
• Mi-UCT consistently outperforms standard UCT in resource-constrained environments with low playout counts.
• The performance advantage of Mi-UCT diminishes as playout counts increase, eventually performing at a level comparable to plain UCT at higher volumes.