Research map / Learning theory and algorithms / Combinatorial and search optimization
Spending Thinking Time Wisely: Accelerating MCTS with Virtual Expansions
Weirui Ye, Pieter Abbeel, Yang Gao
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use adaptive virtual-expansion termination instead of a fixed 150-iteration search | EfficientZero MCTS · Go 9×9 vs GNU Go level 10 | average budget: 150 ± 0.0 -> 76 ± 10.8 | Improved |
| Use adaptive virtual-expansion termination instead of a fixed 150-iteration search | EfficientZero MCTS · Go 9×9 vs GNU Go level 10 | winning rate: 75% ± 3.0% -> 71% ± 4.7% | Hurt |
| Use adaptive virtual-expansion termination instead of 50 fixed iterations | EfficientZero MCTS · Pong | score: 19.7 ± 1.6 -> 18.8 ± 2.8 | No clear effect |
| Use adaptive virtual-expansion termination instead of 50 fixed iterations | EfficientZero MCTS · Breakout | score: 410.7 ± 15.1 -> 372.8 ± 18.3 | Hurt |
| Use adaptive virtual-expansion termination instead of 50 fixed iterations | EfficientZero MCTS · Seaquest | score: 1159.9 ± 90.7 -> 970.0 ± 339.5 | Hurt |
| Use adaptive virtual-expansion termination instead of 50 fixed iterations | EfficientZero MCTS · Hero | score: 9992.1 ± 2059.4 -> 8928.1 ± 2922.1 | Hurt |
| Use adaptive virtual-expansion termination instead of 50 fixed iterations | EfficientZero MCTS · Qbert | score: 14495.8 ± 683.9 -> 11476.6 ± 978.2 | Hurt |
| Add virtual expansion rather than stopping at 30 vanilla iterations (ablation) | V-MCTS · Go 9×9 | winning rate: 17% ± 3.2% -> 32% ± 3.5% | Improved |
| Add virtual expansion rather than greedy expansion for the remaining simulations (ablation) | V-MCTS · Go 9×9 | winning rate: 3% ± 2.0% -> 32% ± 3.5% | Improved |
| Replace the L1 termination criterion with L2 at ε=0.1 (ablation) | V-MCTS · Go 9×9 | winning rate: 81.5% -> 79.8% | Hurt |
| Replace the L1 termination criterion with L2 at ε=0.1 (ablation) | V-MCTS · Go 9×9 | average budget: 96.2 -> 97.1 | Hurt |
| Use L2 termination and reduce ε from 0.1 to 0.05 (ablation) | V-MCTS · Go 9×9 | winning rate: 79.8% -> 81.0% | Improved |
Closest papers by meaning
- Memory-Augmented Monte Carlo Tree Search (AAAI 2018 · similarity 0.90)
- Speculative Monte-Carlo Tree Search (NeurIPS 2024 · similarity 0.88)
- Online Robust Reinforcement Learning Through Monte-Carlo Planning (ICML 2025 · similarity 0.87)
- Monte Carlo Tree Search in the Presence of Transition Uncertainty (AAAI 2024 · similarity 0.87)
- ExACT: Teaching AI Agents to Explore with Reflective-MCTS and Exploratory Learning (ICLR 2025 · similarity 0.86)
- SeeA*: Efficient Exploration-Enhanced A* Search by Selective Sampling (NeurIPS 2024 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
