Research map / Reinforcement learning / Visual and representation RL
Decoupling Representation Learning from Reinforcement Learning
Adam Stooke, Kimin Lee, Pieter Abbeel, Michael Laskin
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 |
|---|---|---|---|
| Train the encoder independently with ATC instead of using RL gradients | Online RL agents · DMControl | return | Improved |
| Train the encoder independently with ATC instead of using RL gradients | Online RL agents · DMLab | return | Improved |
| Train the encoder independently with ATC instead of using RL gradients | Online RL agents · Atari | return | Mixed |
| Add ATC as an auxiliary loss or use it to initialize encoder weights | Atari online RL · Atari | return | Improved |
| Pre-train the encoder with ATC on demonstrations and freeze it during RL | RL agents · DMControl | return | Improved |
| Pre-train the encoder with ATC on demonstrations and freeze it during RL | RL agents · DMLab Lasertag | return | Improved |
| Pre-train the encoder with ATC on demonstrations and freeze it during RL | RL agents · Atari | return | Improved |
| Prioritize ATC replay samples using empirical absolute returns (ablation) | ATC online encoder in Explore Goal Locations · DMLab Explore | return | Improved |
| Double the rate of ATC updates (ablation) | ATC online encoder in Lasertag · DMLab Lasertag | return | Improved |
| Add random shift augmentation during encoder pre-training (ablation) | ATC encoders · Offline pre-training across domains | return | Mixed |
| Reduce random-shift probability from 1.0 to 0.1 (ablation) | ATC encoder in Atari · Atari | return: 1.0 to 0.1 | Improved |
| Add random-shift augmentation during RL, including subpixel random shift on latent images (ablation) | RL agents with frozen DMControl encoders · DMControl | return | Improved |
Closest papers by meaning
- Efficient Reinforcement Learning Through Adaptively Pretrained Visual Encoder (AAAI 2025 · similarity 0.89)
- Reinforcement Learning with Augmented Data (NeurIPS 2020 · similarity 0.89)
- Improving Computational Efficiency in Visual Reinforcement Learning via Stored Embeddings (NeurIPS 2021 · similarity 0.89)
- Contrastive Learning as Goal-Conditioned Reinforcement Learning (NeurIPS 2022 · similarity 0.89)
- TTRV: Test-Time Reinforcement Learning for Vision Language Models (CVPR 2026 · similarity 0.88)
- Towards Principled Representation Learning from Videos for Reinforcement Learning (ICLR 2024 · 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.
