Research map / Reinforcement learning / Offline RL
Flow Matching with Injected Noise for Offline-to-Online Reinforcement Learning
Yongjae Shin, Jongseong Chae, Jongeui Park, 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 |
|---|---|---|---|
| inject noise into flow matching and add entropy-guided sampling (FINO vs FQL) | FQL · OGBench humanoidmaze-medium-navigate | normalized score after online fine-tuning: 61 -> 97 | Improved |
| FINO vs FQL on humanoidmaze-large | FQL · OGBench humanoidmaze-large-navigate | normalized score: 10 -> 33 | Improved |
| FINO vs FQL on antmaze-large | FQL · OGBench antmaze-large-navigate | normalized score: 92 -> 99 | Improved |
| FINO vs FQL on D4RL antmaze | FQL · D4RL antmaze | normalized score: 95 -> 96 | No clear effect |
| FINO vs FQL on D4RL adroit | FQL · D4RL adroit | normalized score: 100 -> 112 | Improved |
| FINO vs ReBRAC on antmaze-giant | ReBRAC · OGBench antmaze-giant-navigate | normalized score: 96 (ReBRAC) vs 79 (FINO) | Hurt |
| FINO vs FQL on antsoccer-arena | FQL · OGBench antsoccer-arena-navigate | normalized score: 74 -> 77 | No clear effect |
| FINO vs FQL on puzzle-4x4 | FQL · OGBench puzzle-4x4-play | normalized score: 45 -> 56 | Improved |
| noise injection in offline pre-training does not degrade offline performance | FQL · OGBench humanoidmaze-medium-navigate (offline) | normalized score: 53 -> 50 | No clear effect |
| add noise to flow matching rather than to output actions (Action Noise) (ablation) | FINO · OGBench and D4RL door-cloned | normalized score | Improved |
| add entropy-guided sampling instead of entropy-regulated noise scaling (ER-Noise) (ablation) | FINO · OGBench humanoidmaze | normalized score | Improved |
| add noise injection (vs w/o Noise) (ablation) | FINO · humanoidmaze-medium-navigate | normalized score | Improved |
Closest papers by meaning
- Flow-Based Policy for Online Reinforcement Learning (NeurIPS 2025 · similarity 0.92)
- Robust Policy Expansion for Offline-to-Online RL under Diverse Data Corruption (NeurIPS 2025 · similarity 0.90)
- Composite Flow Matching for Reinforcement Learning with Shifted-Dynamics Data (NeurIPS 2025 · similarity 0.90)
- Flow Matching Policy Optimization with Mirror Descent and Entropy Constraints (NeurIPS 2026 · similarity 0.90)
- Flow Matching for Offline Reinforcement Learning with Discrete Actions (NeurIPS 2026 · similarity 0.89)
- Flow Q-Learning (ICML 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.
