Research map / Reinforcement learning / Policy gradient and actor-critic
Bridging Successor Measure and Online Policy Learning with Flow Matching-Based Representations
Haosen Shi, Jianda Chen, Sinno Jialin Pan
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 SF2 representation (gamma=0.99) to TD3 and SAC | TD3 / SAC · 7 DMC tasks, normalized AUC | IQM normalized AUC | Improved |
| use full successor horizon gamma=0.99 instead of transition-only gamma=0 (ablation) | SF2 · 7 DMC tasks | normalized AUC | Improved |
| SF2 gains for TD3 versus SAC | SF2 · 7 DMC tasks | normalized AUC | Mixed |
| add SF2 compared with successor-feature baselines in sparse reward tasks | TD3 · sparse reward DMC tasks | return | Improved |
| increase EMA coefficient tau toward 1.0 (ablation) | SF2 · AcrobotSwingup | return | Hurt |
| vary denoising steps 1 to 4 (ablation) | SF2 · AcrobotSwingup | return | No clear effect |
| increase denoising steps (ablation) | SF2 · AcrobotSwingup | run time | Hurt |
| increase feature size (ablation) | SF2 with TD3 · AcrobotSwingup | return | Improved |
| increase feature size (ablation) | SF2 with SAC · AcrobotSwingup | return | No clear effect |
| add SF2 to TD3 costs training time | TD3 · AcrobotSwingup | run time: 659 -> 1300 seconds | Hurt |
| more gradient steps for the feature updater (ablation) | SF2 · DMC IQM | IQM score | Improved |
| use explicit mixture flow matching instead of standard FM (ablation) | flow matching · 1D Gaussian mixture, gamma=0.99 | Wasserstein distance | Improved |
Closest papers by meaning
- Learning Successor Features the Simple Way (NeurIPS 2024 · similarity 0.90)
- A Distributional Analogue to the Successor Representation (ICML 2024 · similarity 0.88)
- Distributional Successor Features Enable Zero-Shot Policy Optimization (NeurIPS 2024 · similarity 0.88)
- $\pi$2vec: Policy Representation with Successor Features (ICLR 2024 · similarity 0.87)
- Reward-Aware Proto-Representations in Reinforcement Learning (NeurIPS 2025 · similarity 0.87)
- Flow-Based Policy for Online Reinforcement Learning (NeurIPS 2025 · similarity 0.86)
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.
