Research map / Reinforcement learning / Visual and representation RL
CURL: Contrastive Unsupervised Representations for Reinforcement Learning
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 |
|---|---|---|---|
| Add contrastive instance discrimination to SAC | SAC pixel control · DMControl100k | median performance vs Dreamer: 1.9x | Helped |
| Add contrastive instance discrimination to Efficient Rainbow | Efficient Rainbow · Atari100k | median performance vs SimPLe: 1.2x | Helped |
| Use stacked frames rather than individual frames for contrastive discrimination (ablation) | CURL · Most DMControl environments | episode return | Helped |
| Use stacked frames rather than individual frames for contrastive discrimination (ablation) | CURL · Reacher easy and Ball in cup catch | episode return | Hurt |
| Increase gradient updates per agent step from 1:1 to 3:1 (ablation) | CURL · DMControl | episode return | Helped |
| Stop critic gradients before the convolutional encoder (ablation) | CURL · Most DMControl environments | policy performance | No clear effect |
| Stop critic gradients before the convolutional encoder (ablation) | CURL · Cheetah run | policy performance | Hurt |
| Remove data augmentation from inputs to the SAC agent while retaining it for contrastive learning (ablation) | CURL + SAC · DMControl500k | performance vs pixel SAC: mean of 2.0x / median of 1.7x | Helped |
| Remove data augmentation from inputs to the SAC agent (ablation) | CURL + SAC · DMControl | performance | Hurt |
| Use the query encoder rather than a momentum-averaged key encoder (ablation) | CURL · Cheetah run | episode return | Hurt |
| Replace the bilinear similarity with cosine similarity (ablation) | CURL · Cheetah run | episode return | Hurt |
Closest papers by meaning
- VI-CuRL: Stabilizing Verifier-Independent RL Reasoning via Confidence-Guided Variance Reduction (NeurIPS 2026 · similarity 0.85)
- Robust Task Representations for Offline Meta-Reinforcement Learning via Contrastive Learning (ICML 2022 · similarity 0.85)
- Foundation Policies with Hilbert Representations (ICML 2024 · similarity 0.83)
- Q-Supervised Contrastive Representation: A State Decoupling Framework for Safe Offline Reinforcement Learning (ICML 2025 · similarity 0.83)
- Contrastive Representation for Data Filtering in Cross-Domain Offline Reinforcement Learning (ICML 2024 · similarity 0.83)
- Sliding Puzzles Gym: A Scalable Benchmark for State Representation in Visual Reinforcement Learning (ICML 2025 · similarity 0.83)
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.