Research map / Probabilistic and scientific modeling / Molecules and proteins
Learning to Stop: Dynamic Simulation Monte-Carlo Tree Search
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 State-UN to stop at the beginning of search | AlphaZero NoGo · NoGo vs PV-MCTS with 1600 simulations | Win Rate: 49% win rate with 801 average simulations; less than 40% of the resource still gives 43% win rate | Mixed |
| Use calibrated PV-NN policy confidence instead of State-UN (ablation) | DS-MCTS on NoGo · NoGo vs PV-MCTS | Win Rate: 48.6% win rate with 925 average simulations; below 0.4 average simulation ratio it performs worse than the baseline | Hurt |
| Add MCTS-UN checkpoint at 160 (ablation) | DS-MCTS on NoGo · NoGo vs PV-MCTS with 1600 simulations | Win Rate: 49.2% win rate with 715 average simulations | Helped |
| Add MCTS-UN checkpoint at 800 (ablation) | DS-MCTS on NoGo · NoGo vs PV-MCTS with 1600 simulations | Win Rate: 48.5% win rate with 591 average simulations | Helped |
| Add the STOP strategy (ablation) | DS-MCTS on NoGo · NoGo vs PV-MCTS with 1600 simulations | Avg Sim: 591 -> 555 | Helped |
| Use dynamic uncertainty stopping rather than PV-MCTS with STOP | NoGo agent · NoGo vs PV-MCTS | Avg Sim: DS-MCTS: 555 simulations; PV-MCTS needs 964 simulations to exceed 48% win rate | Helped |
| Use multiple MCTS-UN checkpoints (ablation) | DS-MCTS on NoGo · NoGo, 4000 vs 1600 | Win Rate: 61.4% at checkpoint 800; 54.8% at checkpoints 160/400/800/1600 | Mixed |
| Scale the search maximum from 1600 to 4000 and use MCTS-UN | DS-MCTS on NoGo · NoGo, 4000 vs 1600 | Win Rate: 61.4% | Helped |
| Use checkpoints 400/800/1600 (ablation) | DS-MCTS on NoGo · NoGo, 6400 vs 6400 | Avg Sim: 2453 average simulations with 46.5% win rate | Helped |
| Add checkpoints 160/400/800 to State-UN stopping (ablation) | DS-MCTS on 9x9 Go · 9x9 Go, 1600 vs 1600 | Avg Sim Ratio: 79.2% -> 59.4% | Helped |
| Use checkpoints 160/400/800 at a larger search maximum (ablation) | DS-MCTS on 9x9 Go · 9x9 Go, 3200 vs 3200 | Avg Sim Ratio: 81.98% -> 58.0% | Helped |
Closest papers by meaning
- Learning to search with MCTSnets (ICML 2018 · similarity 0.90)
- Beyond Monte Carlo Tree Search: Playing Go with Deep Alternative Neural Network and Long-Term Evaluation (AAAI 2017 · similarity 0.90)
- Speculative Monte-Carlo Tree Search (NeurIPS 2024 · similarity 0.88)
- Monte Carlo Tree Search in the Presence of Transition Uncertainty (AAAI 2024 · similarity 0.87)
- Online Robust Reinforcement Learning Through Monte-Carlo Planning (ICML 2025 · similarity 0.87)
- Improving Monte Carlo Tree Search for Symbolic Regression (NeurIPS 2025 · 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.