Summary of Extreme Value Monte Carlo Tree Search, by Masataro Asai et al.
Extreme Value Monte Carlo Tree Search
by Masataro Asai, Stephen Wissow
First submitted to arxiv on: 28 May 2024
Categories
- Main: Artificial Intelligence (cs.AI)
- Secondary: None
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary This paper builds upon previous work that combined Monte-Carlo Tree Search (MCTS) with Upper Confidence Bound 1 (UCB1) Multi-Armed Bandit (MAB) for domain-independent planning. The UCB1 algorithm, originally designed for rewards within the [0,1] range, was not suitable for estimating distances-to-go in classical planning, which can be unbounded in R. To address this issue, the authors proposed combining MCTS with MABs designed for Gaussian reward distributions, achieving improved performance. In this paper, the authors further explore ideal bandits for planning tasks by addressing two issues: under-specification of distances and lack of theoretical justification for Full-Bellman backup. The authors propose two new bandits, UCB1-Uniform/Power, which are theoretically justified using extreme value statistics. These bandits are applied to MCTS for classical planning, with formal regret bounds proved and empirical performance demonstrated. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about improving the way computers plan things. Right now, computers use a method called Monte-Carlo Tree Search (MCTS) that’s good at solving puzzles and games. But it can be slow and not very good at making plans for real-world problems. The authors want to make MCTS better by combining it with another technique called Upper Confidence Bound 1 (UCB1). They’ve already tried this before, but it didn’t work perfectly. Now they’re trying again, using a new way of understanding how UCB1 works. This will help computers plan things more efficiently and effectively. |
