Research map / Reinforcement learning / Visual and representation RL
Exploration via Elliptical Episodic Bonuses
Mikael Henaff, Roberta Raileanu, Minqi Jiang, Tim Rocktäschel
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 |
|---|---|---|---|
| use rank-1 inverse covariance updates instead of full matrix inversion (ablation) | E3B · MiniHack-Freeze-Random-Restricted-v0 | FPS: 256 FPS -> 767 FPS | Improved |
| replace the learned inverse-dynamics encoder with a random fixed encoder (ablation) | E3B · MiniHack (16 tasks) | aggregate performance | Hurt |
| tie the feature encoder to the policy network (ablation) | E3B · MiniHack (16 tasks) | aggregate performance | Hurt |
| compute the elliptical bonus across the agent's whole lifetime (non-episodic) (ablation) | E3B · MiniHack (16 tasks) | aggregate performance | Hurt |
| use the full state for the count-based episodic bonus (original NovelD) (ablation) | NovelD · MiniHack (16 tasks) | aggregate performance | Hurt |
| restrict NovelD count bonus to agent (x,y) position (ablation) | NovelD-position · MiniHack navigation tasks | aggregate performance | Improved |
| restrict NovelD count bonus to agent (x,y) position (ablation) | NovelD-position · MiniHack skill-based tasks | aggregate performance | Hurt |
| restrict NovelD count bonus to message only (ablation) | NovelD-message · MiniHack skill-based tasks | aggregate performance | Improved |
| restrict NovelD count bonus to message only (ablation) | NovelD-message · MiniHack navigation tasks | aggregate performance | Hurt |
| use E3B on raw full state instead of NovelD variants with hand-chosen features (ablation) | E3B · MiniHack (16 tasks) | aggregate performance | Improved |
| remove the count-based episodic term (ablation) | RIDE, AGAC, NovelD · additional MiniGrid environments | return | Hurt |
| restrict the 78-action space to task-relevant actions (ablation) | all methods · MiniHack skill-based tasks | task success: 78 actions -> restricted | Improved |
Closest papers by meaning
- Rethinking Exploration in Reinforcement Learning with Effective Metric-Based Exploration Bonus (NeurIPS 2024 · similarity 0.87)
- Just Cluster It: An Approach for Exploration in High-Dimensions using Clustering and Pre-Trained Representations (ICML 2024 · similarity 0.87)
- Episodic Novelty Through Temporal Distance (ICLR 2025 · similarity 0.86)
- Improving Intrinsic Exploration by Creating Stationary Objectives (ICLR 2024 · similarity 0.86)
- Open the Black Box: Step-based Policy Updates for Temporally-Correlated Episodic Reinforcement Learning (ICLR 2024 · similarity 0.86)
- First-Explore, then Exploit: Meta-Learning to Solve Hard Exploration-Exploitation Trade-Offs (NeurIPS 2024 · 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.
