Research map / Reinforcement learning / Visual and representation RL
Learning Generalizable Representations for Reinforcement Learning via Adaptive Meta-learner of Behavioral Similarities
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 |
|---|---|---|---|
| Use AMBS instead of DrQ + PSEs | RL representation learning · DAVIS unseen backgrounds, C-swingup | score: 749 ± 19 -> 807 ± 41 | Helped |
| Use AMBS instead of DrQ + PSEs | RL representation learning · DAVIS unseen backgrounds, F-spin | score: 779 ± 49 -> 933 ± 96 | Helped |
| Use AMBS instead of DrQ + PSEs | RL representation learning · DAVIS unseen backgrounds, C-run | score: 308 ± 12 -> 332 ± 27 | Helped |
| Use AMBS instead of DrQ + PSEs | RL representation learning · DAVIS unseen backgrounds, W-walk | score: 789 ± 28 -> 893 ± 85 | Helped |
| Use AMBS instead of DrQ + PSEs | RL representation learning · DAVIS unseen backgrounds, R-easy | score: 955 ± 10 -> 980 ± 11 | Helped |
| Use AMBS instead of DrQ + PSEs | RL representation learning · DAVIS unseen backgrounds, BiC-catch | score: 821 ± 17 -> 944 ± 59 | Helped |
| Add learned meta-learners for reward and dynamics similarities instead of L1 distance (ablation) | AMBS · DMC natural video background, cheetah-run and walker-walk | RL performance | Helped |
| Add data augmentation (ablation) | AMBS · DMC natural video background, cheetah-run and walker-walk | RL performance | Helped |
| Adaptively balance reward and dynamics representations with a learned weight (ablation) | AMBS · DMC natural video background, cheetah-run and walker-walk | RL performance | Helped |
| Learn the reward-dynamics weight instead of fixing it at 0.5 (ablation) | AMBS · DMC natural video background, cheetah-run and walker-walk | RL performance | Helped |
| Learn the reward-dynamics weight instead of fixing reward weight to 1 and dynamics weight to γ (ablation) | AMBS · DMC natural video background, cheetah-run and walker-walk | RL performance | Helped |
| Share only convolutional layers between actor and critic instead of sharing the full encoders (ablation) | AMBS · Walker-walk, natural video background | RL performance | Helped |
Closest papers by meaning
- Learning Markov State Abstractions for Deep Reinforcement Learning (NeurIPS 2021 · similarity 0.90)
- Generalizable Task Representation Learning for Offline Meta-Reinforcement Learning with Data Limitations (AAAI 2024 · similarity 0.89)
- The Surprising Ineffectiveness of Pre-Trained Visual Representations for Model-Based Reinforcement Learning (NeurIPS 2024 · similarity 0.88)
- MetaCARD: Meta-Reinforcement Learning with Task Uncertainty Feedback via Decoupled Context-Aware Reward and Dynamics Components (AAAI 2024 · similarity 0.88)
- Towards Principled Representation Learning from Videos for Reinforcement Learning (ICLR 2024 · similarity 0.88)
- Entropy Regularized Task Representation Learning for Offline Meta-Reinforcement Learning (AAAI 2025 · similarity 0.88)
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.