Research map / Reinforcement learning / MDPs and planning
Why Generalization in RL is Difficult: Epistemic POMDPs and Implicit Partial Observability
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 LEEP instead of PPO | PPO on Procgen · Maze | test return | Helped |
| Use LEEP instead of PPO | PPO on Procgen · Heist | test return | Helped |
| Use LEEP instead of PPO | PPO on Procgen · Dodgeball | test return | Helped |
| Use LEEP instead of PPO | PPO on Procgen · BigFish | test return | No clear effect |
| Use LEEP instead of PPO with only 50 training levels | PPO on Procgen · Maze | test return | Helped |
| Remove the ensemble disagreement regularization (ablation) | LEEP · Maze | test-time performance | Hurt |
| Replace the optimistic max link with an average-policy link (ablation) | LEEP · Maze | test-time performance | Hurt |
| Use Distral instead of LEEP (ablation) | LEEP-style multi-task learning · Procgen generalization | generalization performance | Hurt |
| Use a deterministic policy that repeats the same label after an incorrect guess | DQN on sequential FashionMNIST classification · FashionMNIST test set | return | Hurt |
| Use process-of-elimination guessing after each incorrect guess | Sequential FashionMNIST classification policy · FashionMNIST test set | return | Helped |
Closest papers by meaning
- Instance-based Generalization in Reinforcement Learning (NeurIPS 2020 · similarity 0.91)
- On the Generalization Gap in Reparameterizable Reinforcement Learning (ICML 2019 · similarity 0.90)
- Provable Representation with Efficient Planning for Partially Observable Reinforcement Learning (ICML 2024 · similarity 0.89)
- Generalization of RLVR Using Causal Reasoning as a Testbed (ICLR 2026 · similarity 0.89)
- Provable Zero-Shot Generalization in Offline Reinforcement Learning (ICML 2025 · similarity 0.88)
- Mitigating Partial Observability in Sequential Decision Processes via the Lambda Discrepancy (NeurIPS 2024 · similarity 0.88)
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.