Research map / Reinforcement learning / Offline RL
Flow Actor-Critic for Offline Reinforcement Learning
Jongseong Chae, Jongeui Park, Yongjae Shin, Gyeongmin Kim, Seungyul Han, Youngchul Sung
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 flow-density-based critic penalization to flow actor regularization | FQL · OGBench (state-based average) | success rate: 43.6 -> 60.3 | Improved |
| add flow-based critic penalization | FQL · OGBench puzzle-3x3-play | success rate: 29.6 -> 100.0 | Improved |
| add flow-based critic penalization | FQL · D4RL Antmaze average | normalized score: 83.5 -> 90.5 | Improved |
| add flow-based critic penalization | FQL · D4RL MuJoCo average | normalized score: 71.8 -> 88.2 | Improved |
| use FAC instead of ReBRAC | ReBRAC · D4RL MuJoCo average | normalized score: 88.5 vs 88.2 | No clear effect |
| use FAC instead of ReBRAC | ReBRAC · OGBench antmaze-giant-navigate | success rate: 26.2 vs 23.0 | Hurt |
| increase candidate actions N (argmax Q) with unpenalized critic (ablation) | FQL · OGBench | performance | Hurt |
| increase candidate actions N with penalized critic (ablation) | FAC · OGBench | performance | No clear effect |
| use fewer Euler steps for flow behavior proxy (T<10) (ablation) | FAC · puzzle-3x3-play | performance | Hurt |
| use convex weight function instead of linear (ablation) | FAC · Critic penalization weight design | performance | Hurt |
| use concave weight function instead of linear (ablation) | FAC · Critic penalization weight design | performance | No clear effect |
| apply FAC online fine-tuning | FAC · D4RL Adroit | online gains | Hurt |
Closest papers by meaning
- Guided Flow Policy: Learning from High-Value Actions in Offline Reinforcement Learning (ICLR 2026 · similarity 0.92)
- Flow Q-Learning (ICML 2025 · similarity 0.91)
- Offline Actor-Critic Reinforcement Learning Scales to Large Models (ICML 2024 · similarity 0.90)
- Offline-Boosted Actor-Critic: Adaptively Blending Optimal Historical Behaviors in Deep Off-Policy RL (ICML 2024 · similarity 0.90)
- Flow-Based Policy for Online Reinforcement Learning (NeurIPS 2025 · similarity 0.90)
- Fewer May Be Better: Enhancing Offline Reinforcement Learning with Reduced Dataset (ICLR 2025 · 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.
