Research map / Reinforcement learning / Visual and representation RL
Neuro-algorithmic Policies Enable Fast Combinatorial Generalization
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 NAP instead of PPO | Maze · Maze, unseen levels | success rate | Helped |
| Use NAP instead of behavior cloning | Crash Jewel Hunt 5×5 · Crash Jewel Hunt 5×5, 1000 test levels | levels solved: 30 trajectories; a third of the 1000 test-levels; less than 50 out of the 1000 | Helped |
| Use NAP instead of PPO or DrAC | ProcGen environments · Procedurally generated test levels | generalization performance: <500 levels; 10,000 levels; 1000 levels | Helped |
| Use longer planning horizons in environments with more dynamic interactions (ablation) | NAP · Dynamic environments | success rate | Helped |
| Use a shorter planning horizon (ablation) | NAP · Dynamic environments | success rate | Hurt |
| Use a longer planning horizon (ablation) | NAP · Maze | success rate | No clear effect |
| Use NAP instead of DrAC* | Leaper(Grid) · Leaper(Grid), unseen levels | path length | Helped |
| Replace learned start and goal predictors with ground-truth vertices (ablation) | NAP · Evaluated environments except Chaser | performance | No clear effect |
Closest papers by meaning
- ASAP: Exploiting the Satisficing Generalization Edge in Neural Combinatorial Optimization (ICML 2026 · similarity 0.88)
- Optimizing Solution-Samplers for Combinatorial Problems: The Landscape of Policy-Gradient Method (NeurIPS 2023 · similarity 0.88)
- Towards Generalization and Simplicity in Continuous Control (NeurIPS 2017 · similarity 0.88)
- Efficient Reinforcement Learning by Discovering Neural Pathways (NeurIPS 2024 · similarity 0.86)
- DNCs Require More Planning Steps (ICML 2024 · similarity 0.86)
- GOAL: A Generalist Combinatorial Optimization Agent Learner (ICLR 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.