Research map / Reinforcement learning / Visual and representation RL
Learning Human Objectives by Evaluating Hypothetical Behavior
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 |
|---|---|---|---|
| Synthesize hypothetical behaviors instead of querying only real training-environment rollouts | ReQueST · MNIST, 2D navigation, and Car Racing test environments | test performance: at least 2x | Helped |
| Synthesize hypothetical trajectories rather than querying only real rollouts | ReQueST · 2D navigation test environment | success rate | Helped |
| Add reward-maximizing synthetic queries to correct false positives | ReQueST · 2D navigation | false positive rate | Helped |
| Add reward-minimizing synthetic queries to elicit unsafe-state labels | ReQueST · 2D navigation | unsafe-state true negative rate: 100% | Helped |
| Use synthetic queries rather than real environment visits to learn unsafe states | ReQueST · 2D navigation deployment | agent failure rate: 75% | Helped |
| Optimize acquisition functions rather than randomly sampling from the generative model | ReQueST · MNIST, 2D navigation, and Car Racing | queries needed for performance | Helped |
| Remove uncertainty-maximizing queries (ablation) | ReQueST · MNIST | test performance | Hurt |
| Remove novelty-maximizing queries (ablation) | ReQueST · MNIST with few queries | test performance | Hurt |
| Remove novelty-maximizing queries (ablation) | ReQueST · MNIST after enough queries | query efficiency | Helped |
| Remove uncertainty-maximizing queries (ablation) | ReQueST · 2D navigation | test performance | Hurt |
| Remove reward-minimizing queries (ablation) | ReQueST · Car Racing | true reward | Hurt |
| Remove any one of reward-maximizing, reward-minimizing, or novelty-maximizing queries (ablation) | ReQueST · 2D navigation | test performance | No clear effect |
Closest papers by meaning
- PEBBLE: Feedback-Efficient Interactive Reinforcement Learning via Relabeling Experience and Unsupervised Pre-training (ICML 2021 · similarity 0.88)
- DUO: Diverse, Uncertain, On-Policy Query Generation and Selection for Reinforcement Learning from Human Feedback (AAAI 2025 · similarity 0.88)
- Reinforcement Learning with Lookahead Information (NeurIPS 2024 · similarity 0.87)
- Scaling Autonomous Agents via Automatic Reward Modeling And Planning (ICLR 2025 · similarity 0.87)
- The Distributional Reward Critic Framework for Reinforcement Learning Under Perturbed Rewards (AAAI 2025 · similarity 0.87)
- RLZero: Direct Policy Inference from Language Without In-Domain Supervision (NeurIPS 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.