Research map / Reinforcement learning / Policy gradient and actor-critic
Improve Agents without Retraining: Parallel Tree Search with Off-Policy Correction
Gal Dalal, Assaf Hallak, Steven Dalton, iuri frosio, Shie Mannor, Gal Chechik
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 |
|---|---|---|---|
| Apply vanilla tree search at inference to pretrained Rainbow | Rainbow agent · Atari (Frostbite, Breakout, BeamRider, SpaceInvaders, Asteroids) | reward | Hurt |
| Add off-policy correction (BCTS) to tree search over vanilla TS | Rainbow agent with TS · Atari Frostbite, Breakout, BeamRider, SpaceInvaders | reward: d = 4: score more than doubles; BeamRider about x5 | Improved |
| Increase tree depth with BCTS | Rainbow with BCTS · Atari inference | reward | Improved |
| BCTS vs vanilla TS on Asteroids | Rainbow with TS · Atari Asteroids | reward | No clear effect |
| Increase tree depth in DQN training with TS (d=1 to d=4) | DQN with TS · Atari Asteroids | average score: 2, 093 -> 17, 929 | Improved |
| Increase tree depth in DQN training with TS (d=1 to d=4) | DQN with TS · Atari Breakout | average score: 385 -> 620 | Mixed |
| Increase tree depth in DQN training with TS (d=1 to d=4) | DQN with TS · Atari MsPacman | average score: 1, 644 -> 4, 021 | Improved |
| Increase tree depth in DQN training with TS (d=1 to d=4) | DQN with TS · Atari SpaceInvaders | average score: 675 -> 2, 550 | Improved |
| Increase tree depth in DQN training with TS (d=1 to d=4) | DQN with TS · Atari VideoPinball | average score: 229, 129 -> 345, 742 | Mixed |
| DQN with TS (d=4) vs plain DQN under equal 2-day compute | DQN with TS · Atari Asteroids | average score: 1, 664 -> 17, 929 | Improved |
| DQN with TS (d=4) vs Rainbow under equal compute | DQN with TS · Atari VideoPinball | average score: 641, 235 -> 345, 742 | Hurt |
| DQN with TS (d=4) vs Rainbow under equal compute | DQN with TS · Atari MsPacman | average score: 3, 600 -> 4, 021 | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Rainbow — Rainbow: Combining Improvements in Deep Reinforcement Learning (AAAI 2018) · better on 4 of 5 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
- Tree Search for LLM Agent Reinforcement Learning (ICLR 2026 · similarity 0.89)
- Feedback-Based Tree Search for Reinforcement Learning (ICML 2018 · similarity 0.88)
- Decoupling Exploration and Policy Optimization: Uncertainty Guided Tree Search for Hard Exploration (NeurIPS 2026 · similarity 0.88)
- Policy Gradient with Tree Expansion (ICML 2025 · similarity 0.87)
- MALinZero: Efficient Low-Dimensional Search for Mastering Complex Multi-Agent Planning (NeurIPS 2025 · similarity 0.87)
- ContraDiff: Planning Towards High Return States via Contrastive Learning (ICLR 2025 · similarity 0.87)
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.
