Research map / Reinforcement learning / Offline RL
Mutual Information Regularized Offline Reinforcement Learning
Xiao Ma, Bingyi Kang, Zhongwen Xu, Min Lin, Shuicheng YAN
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 |
|---|---|---|---|
| Increase Monte-Carlo samples from 5 to 50 (ablation) | MISA · gym-locomotion-v2 (total) | normalized score: 711 -> 742.9 | Improved |
| Increase Monte-Carlo samples from 20 to 50 (ablation) | MISA · gym-locomotion-v2 (total) | normalized score: 721.6 -> 742.9 | Improved |
| Increase MCMC burn-in steps from 1 to 5 (ablation) | MISA · gym-locomotion-v2 (total) | normalized score: 687.2 -> 742.9 | Improved |
| Add the Barber-Agakov term (ablation) | MISA · gym-locomotion-v2 (total) | normalized score: 715 -> 742.9 | Improved |
| Use tighter MI bound than Barber-Agakov bound alone (ablation) | MISA · gym-locomotion-v2 (total) | normalized score: 254.4 -> 742.9 | Improved |
| Use tighter bound than f-divergence variant (ablation) | MISA · gym-locomotion-v2 (total) | normalized score: 504.7 -> 742.9 | Improved |
| Use tighter bound than Donsker-Varadhan variant (ablation) | MISA · gym-locomotion-v2 (total) | normalized score: 673.7 -> 742.9 | Improved |
| Add bias correction with extra MCMC samples (ablation) | MISA · gym-locomotion-v2 (total) | normalized score: 720.9 -> 742.9 | Improved |
| Use Q function as energy instead of separate T network (ablation) | MISA · gym-locomotion-v2 (total) | normalized score: 271.0 -> 742.9 | Improved |
| Learn separate T network (MISA-T) instead of Q (ablation) | MISA · halfcheetah-medium-v2 | normalized score: 60.8 vs 47.4 | Hurt |
| Remove Barber-Agakov term (ablation) | MISA · hopper-medium-replay-v2 | normalized score: 98.6 -> 100.3 | No clear effect |
| MISA instead of IQL on antmaze | IQL · antmaze-v0 (total) | normalized score: 378 -> 348.1 | Hurt |
Closest papers by meaning
- Policy Regularization with Dataset Constraint for Offline Reinforcement Learning (ICML 2023 · similarity 0.91)
- Iteratively Refined Behavior Regularization for Offline Reinforcement Learning (NeurIPS 2024 · similarity 0.90)
- Doubly Mild Generalization for Offline Reinforcement Learning (NeurIPS 2024 · similarity 0.90)
- Fewer May Be Better: Enhancing Offline Reinforcement Learning with Reduced Dataset (ICLR 2025 · similarity 0.89)
- Beyond OOD State Actions: Supported Cross-Domain Offline Reinforcement Learning (AAAI 2024 · similarity 0.89)
- A Perspective of Q-value Estimation on Offline-to-Online Reinforcement Learning (AAAI 2024 · similarity 0.89)
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.
