Research map / Reinforcement learning / Visual and representation RL
Masked Contrastive Representation Learning for Reinforcement Learning
Jinhua Zhu, Yingce Xia, Lijun Wu, Jiajun Deng, Wengang Zhou, Tao Qin, Tie‐Yan Liu, Houqiang Li
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 |
|---|---|---|---|
| add masked contrastive learning with Transformer on top of Eff. Rainbow (M-CURL vs CURL) | M-CURL · Atari 100k | median HNS: 17.5% -> 24.2% | Improved |
| add masked contrastive learning to Efficient Rainbow | Eff. Rainbow · Atari 100k | median HNS: 16.1% -> 24.2% | Improved |
| use M-CURL instead of CURL on DMControl | SAC · DMControl 100k/500k | return: outperforms CURL on 14 out of 16 | Improved |
| feed consecutive sequential inputs to CURL (seq-CURL) instead of random sampling (ablation) | CURL · DMControl Hopper-Stand | return | Hurt |
| add a Transformer module on the key side (sym-CURL) (ablation) | M-CURL · DMControl Hopper-Stand | return | Hurt |
| set masked probability to 0.9 instead of 0.5 (ablation) | M-CURL · DMControl Hopper-Stand | return | Hurt |
| set masked probability to 0.1 instead of 0.5 (ablation) | M-CURL · DMControl Hopper-Stand | return | Hurt |
| vary sequence length T from 16 to 256 (ablation) | M-CURL · DMControl Hopper-Stand | return | No clear effect |
| increase Transformer layers from 3 to 4 (ablation) | M-CURL · DMControl Hopper-Stand | return | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- CURL — CURL: Contrastive Unsupervised Representations for Reinforcement Learning (ICML 2020) · better on 16 of 18 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- CURL: Contrastive Unsupervised Representations for Reinforcement Learning (ICML 2020 · similarity 0.91)
- Towards Principled Representation Learning from Videos for Reinforcement Learning (ICLR 2024 · similarity 0.86)
- Contrastive Representation for Data Filtering in Cross-Domain Offline Reinforcement Learning (ICML 2024 · similarity 0.84)
- VI-CuRL: Stabilizing Verifier-Independent RL Reasoning via Confidence-Guided Variance Reduction (NeurIPS 2026 · similarity 0.84)
- Skill-aware Mutual Information Optimisation for Zero-shot Generalisation in Reinforcement Learning (NeurIPS 2024 · similarity 0.83)
- Efficient Reinforcement Learning Through Adaptively Pretrained Visual Encoder (AAAI 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.
