Research map / Reinforcement learning / Visual and representation RL
Ranking Policy Decisions
Hadrien Pouget, Hana Chockler, Youcheng Sun, Daniel Kroening
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 |
|---|---|---|---|
| Rank decisions with the SBFL portfolio instead of random ranking | Pruned policies using repeat-previous-action default · MiniGrid | % of original states restored to recover 90% performance: 99 \pm 00 -> 49 \pm 00 | Improved |
| Rank decisions with the SBFL portfolio instead of random ranking | Pruned policies using repeat-previous-action default · CartPole | % of original states restored to recover 90% performance: 65 \pm 02 -> 31 \pm 04 | Improved |
| Rank decisions with the SBFL portfolio instead of random ranking | Pruned policies using repeat-previous-action default · Breakout | % of original states restored to recover 90% performance: 100 \pm 00 -> 10 \pm 00 | Improved |
| Rank decisions with the SBFL portfolio instead of random ranking | Pruned policies using repeat-previous-action default · Pong | % of original states restored to recover 90% performance: 79 \pm 03 -> 21 \pm 03 | Improved |
| Use state-visit frequency instead of SBFL to rank decisions | Policy pruning · Benchmarks reported in the paper | pruned-policy performance per original-policy step usage | Hurt |
| Replace repeat-previous-action default with random actions (ablation) | SBFL-ranked pruned policies · Benchmarks with both default-action experiments | pruning efficiency at recovering original performance: In most games, the difference was small (\leq 10\%); a few games show a marginal improvement. | Mixed |
| Use a specialized ball-and-paddle coordinate abstraction instead of the generic Breakout abstraction (ablation) | Breakout SBFL ranking · Breakout | % of steps using original policy to recover 90% performance: 13\% fewer steps | Improved |
| Use a specialized ball-and-paddle coordinate abstraction instead of the generic Breakout abstraction (ablation) | Breakout SBFL ranking · Breakout | % of original states restored to recover 90% performance | Hurt |
Closest papers by meaning
- Test Where Decisions Matter: Importance-driven Testing for Deep Reinforcement Learning (NeurIPS 2024 · similarity 0.86)
- Efficient Multi-Policy Evaluation for Reinforcement Learning (AAAI 2025 · similarity 0.86)
- Abstract Reward Processes: Leveraging State Abstraction for Consistent Off-Policy Evaluation (NeurIPS 2024 · similarity 0.86)
- How Ensembles of Distilled Policies Improve Generalisation in Reinforcement Learning (NeurIPS 2025 · similarity 0.86)
- Reflective Prompted Policy Optimization: Trajectory-Grounded Revision and Salience Bias (NeurIPS 2026 · similarity 0.86)
- Doubly Optimal Policy Evaluation for Reinforcement Learning (ICLR 2025 · similarity 0.85)
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.
