Monte-Carlo tree search with tree shape control

摘要

The paper, basing on analysis of the Monte-Carlo Tree Search (MCTS) method and specific features of its behavior for various cases of usage, proposes a new variant of the method, which was called as Monte-Carlo Tree Search with Tree Shape Control (MCTS-TSC) and which uses original Depth-Width Criteria (DWCs) for both tree shape estimation and control during search and for estimation and selection of potentially better options for search continuation. Proposed Tree Shape Control (TSC) technique can be used with some other tuning, pruning, and learning techniques. Besides, it can provide better scheduling of MCTS parallelization.