Research map / Reinforcement learning / MDPs and planning
Learning the Reward Function for a Misspecified Model
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 |
|---|---|---|---|
| Train the reward model on hallucinated model states using rewards from parallel environment rollouts (ablation) | H-DAgger-MC · Shooter, no model limitations | discounted return | Helped |
| Train the reward model only on environment states instead of hallucinated states (ablation) | H-DAgger-MC · Shooter, no model limitations | discounted return | Mixed |
| Use a perfect reward model instead of a learned reward model (ablation) | H-DAgger-MC · Shooter, no model limitations | discounted return | Hurt |
| Train the reward model on hallucinated model states using rewards from parallel environment rollouts (ablation) | DAgger-MC · Shooter, moving bullseyes | discounted return | Helped |
| Use a perfect reward model instead of a learned reward model (ablation) | DAgger-MC · Shooter, moving bullseyes | discounted return | Hurt |
| Train the reward model on hallucinated model states using rewards from parallel environment rollouts (ablation) | H-DAgger-MC · Shooter, moving bullseyes | discounted return | Helped |
| Train the reward model on hallucinated model states using rewards from parallel environment rollouts (ablation) | DAgger-MC · Shooter, flawed factored structure | discounted return | Helped |
| Use a perfect reward model instead of a learned reward model (ablation) | H-DAgger-MC · Shooter, flawed factored structure | discounted return | Hurt |
| Train the reward model only on environment states instead of hallucinated states (ablation) | H-DAgger-MC · Shooter, flawed factored structure | discounted return | Hurt |
Closest papers by meaning
- Reward-Consistent Dynamics Models are Strongly Generalizable for Offline Reinforcement Learning (ICLR 2024 · similarity 0.87)
- Improved Sample Complexity for Reward-free Reinforcement Learning under Low-rank MDPs (ICLR 2023 · similarity 0.87)
- Algorithmic Framework for Model-based Deep Reinforcement Learning with Theoretical Guarantees (ICLR 2019 · similarity 0.87)
- Model-Based Offline Reinforcement Learning with Local Misspecification (AAAI 2023 · similarity 0.86)
- The Perils of Optimizing Learned Reward Functions: Low Training Error Does Not Guarantee Low Regret (ICML 2025 · similarity 0.86)
- Inverting the Bellman Equation: From $Q$-Values to World Models (NeurIPS 2026 · 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.