Research map / Reinforcement learning / Visual and representation RL
M-Walk: Learning to Walk over Graphs using Monte Carlo Tree Search
Yelong Shen, Jianshu Chen, Po-Sen Huang, Yuqing Guo, Jianfeng 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 |
|---|---|---|---|
| Combine MCTS trajectory generation with off-policy Q-learning policy improvement | M-Walk on Three Glass Puzzle · Three Glass Puzzle | accuracy: (49.0 ± 2.6)% -> (99.0 ± 1.0)% | Improved |
| Replace MCTS with beam search at test time (ablation) | M-Walk · Three Glass Puzzle test | accuracy: 99.0 (1.0) -> 71.7 (4.5) | Hurt |
| Use the M-Walk neural architecture rather than MINERVA's architecture (ablation) | PG-Walk · NELL995 | MAP: 87.6 -> 88.9 | Improved |
| Add MCTS to policy-gradient training and beam-search testing (ablation) | M-Walk on NELL995 · NELL995 | MAP: 88.9 -> 89.9 | Improved |
| Add MCTS trajectory generation to Q-learning-only training (ablation) | M-Walk on NELL995 · NELL995 | MAP: 87.8 -> 89.9 | Improved |
| Add MCTS to policy-gradient training and beam-search testing (ablation) | M-Walk on WN18RR · WN18RR | HITS@1: 39.3 (0.2) -> 41.4 (0.1) | Improved |
| Add MCTS trajectory generation to Q-learning-only training (ablation) | M-Walk on WN18RR · WN18RR | HITS@1: 38.2 (0.3) -> 41.4 (0.1) | Improved |
| Add MCTS to policy-gradient training and beam-search testing (ablation) | M-Walk on FB15k-237 · FB15k-237 | HITS@1: 14.8(0.2) -> 16.5(0.3) | Improved |
| Add MCTS trajectory generation to Q-learning-only training (ablation) | M-Walk on FB15k-237 · FB15k-237 | HITS@1: 15.5(0.2) -> 16.5(0.3) | Improved |
| Use M-Walk rather than ConvE | WN18RR · WN18RR | HITS@3: 44.7 (0.2) vs 44.5 (0.2) | No clear effect |
| Use M-Walk rather than embedding-based baselines | FB15k-237 · FB15k-237 | HITS@1: 16.5(0.3) vs 23.3(0.4) | Hurt |
| Increase the number of MCTS rollouts (ablation) | M-Walk on Three Glass Puzzle · Three Glass Puzzle test | accuracy: 18.0 (1.7) -> 99.0 (1.0) | Improved |
Later papers that compared with this method
1 later accepted paper reports M-walk as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- GaussianPath:A Bayesian Multi-Hop Reasoning Framework for Knowledge Graph Reasoning (AAAI 2021 · better on 2 of 3 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Revisiting Random Walks for Learning on Graphs (ICLR 2025 · similarity 0.86)
- Walk Wisely on Graph: Knowledge Graph Reasoning with Dual Agents via Efficient Guidance-Exploration (AAAI 2025 · similarity 0.84)
- Learning Long Range Dependencies on Graphs via Random Walks (ICLR 2025 · similarity 0.84)
- Random Search Neural Networks for Efficient and Expressive Graph Learning (NeurIPS 2025 · similarity 0.84)
- Decentralized Monte Carlo Tree Search for Partially Observable Multi-Agent Pathfinding (AAAI 2024 · similarity 0.84)
- ExACT: Teaching AI Agents to Explore with Reflective-MCTS and Exploratory Learning (ICLR 2025 · similarity 0.83)
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.
