Research map / Reinforcement learning / Offline RL
Look Beneath the Surface: Exploiting Fundamental Symmetry for Sample-Efficient Offline RL
Peng Cheng, Xianyuan Zhan, Zhihao Wu, Wenjia Zhang, Youfang Lin, Shou cheng Song, Han Wang, Li Jiang
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 TSRL instead of DOGE on 10k datasets | TSRL · Hopper-medium 10k | normalized score: 44.2 ± 10.2 -> 62.0±3.7 | Improved |
| use TSRL instead of best baseline on 10k Halfcheetah-expert | TSRL · Halfcheetah-expert 10k | normalized score: 4.2±0.94 -> 40.6±24.4 | Improved |
| use TSRL on full-size Hopper-medium vs DOGE | TSRL · Hopper-medium full 1M | normalized score: 98.6 ± 2.1 -> 86.7±8.7 | Hurt |
| add T-symmetry data augmentation vs Gaussian noise (S4RL-N) | TSRL · Walker2d-expert 10k | normalized score: 56.5 ± 26.7 -> 102.2 ± 11.3 | Improved |
| replace TDM-based augmentation with CABI | TSRL · Hopper-expert 10k | normalized score: 34.6 ± 24.4 -> 82.7 ± 21.9 | Improved |
| weaken TDM regularization (loss weights 0.01) (ablation) | TSRL · Hopper-medium 10k | normalized score: 62.0±3.7 -> 50.7±13.6 | Hurt |
| strengthen TDM regularization (loss weights 1) (ablation) | TSRL · Hopper-medium 10k | normalized score: 62.0±3.7 -> 43.6±14.3 | Hurt |
| weaken TDM regularization on Halfcheetah-medium (ablation) | TSRL · Halfcheetah-medium 10k | normalized score: 38.4±3.1 -> 39.1±3.6 | No clear effect |
| weaken TDM regularization on Walker2d-medium-expert (ablation) | TSRL · Walker2d-medium-expert 10k | normalized score: 46.4±17.4 -> 25.9±20.7 | Hurt |
| remove ODE structure from TDM (TDM-no-ODE) (ablation) | TSRL · 10k MuJoCo datasets | normalized score | Hurt |
| use T-symmetry regularized TDM representation instead of AE or SimSiam (ablation) | TSRL · 10k MuJoCo datasets | normalized score | Improved |
| remove T-symmetry representation, policy constraint or augmentation (no-R, no-P, no-A) (ablation) | TSRL · Halfcheetah-medium and medium-expert 10k | normalized score | Hurt |
Closest papers by meaning
- Beyond OOD State Actions: Supported Cross-Domain Offline Reinforcement Learning (AAAI 2024 · similarity 0.91)
- A Clean Slate for Offline Reinforcement Learning (NeurIPS 2025 · similarity 0.90)
- Offline RL with Observation Histories: Analyzing and Improving Sample Complexity (ICLR 2024 · similarity 0.90)
- Compositional Conservatism: A Transductive Approach in Offline Reinforcement Learning (ICLR 2024 · similarity 0.90)
- Simple Ingredients for Offline Reinforcement Learning (ICML 2024 · similarity 0.90)
- Reward-Consistent Dynamics Models are Strongly Generalizable for Offline Reinforcement Learning (ICLR 2024 · similarity 0.90)
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.
