Research map / Reinforcement learning / Policy gradient and actor-critic
GoldenStart: Q-Guided Priors and Entropy Control for Distilling Flow Policies
He Zhang, Ying Sun, Hui Xiong
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 Q-guided prior and entropy-controlled distillation | FQL · OGBench offline average | success rate (%): 38.5 -> 47.1 | Improved |
| add Q-guided prior and entropy-controlled distillation | FQL · Cube Double Play offline | success rate (%): 36 -> 51.3 | Improved |
| add Q-guided prior and entropy-controlled distillation | FQL · Puzzle-4x4 offline-to-online | success rate (%): 38 -> 100 | Improved |
| add Q-guided prior and entropy-controlled distillation | FQL · OGBench offline-to-online average | success rate (%): 67.6 -> 88.6 | Improved |
| add Q-guided prior and entropy-controlled distillation | FQL · Visual environments offline average | success rate (%): 65.4 -> 70.9 | Improved |
| add Q-guided prior and entropy-controlled distillation | FQL · HumanoidMaze Large offline | success rate (%): 7 -> 4.6 | Hurt |
| add Q-guided prior and entropy-controlled distillation | FQL · Visual Puzzle-3x3 offline | success rate (%): 94 -> 88.67 | Hurt |
| add learned Q-guided prior (more candidates 0 -> 5 -> 15) (ablation) | GoldenStart · Multi-Crescent online | return | Improved |
| add controllable entropy (vs deterministic w/o CE) (ablation) | GoldenStart · Puzzle-4x4 / Multi-Crescent online | return | Improved |
| add Q-guided prior (ablation) | GoldenStart (high exploration) · Multi-Crescent online | return: 5.9 -> 14.0 | Improved |
| lower target entropy (mult 0.1 -> 0.75) (ablation) | GoldenStart · Multi-Crescent online | return: 14.0 -> 9.3 | Hurt |
| add rejection sampling (N=32) to FQL (ablation) | FQL · Cube-Double | success rate (%): 36 -> 37 | No clear effect |
Closest papers by meaning
- Flow Q-Learning (ICML 2025 · similarity 0.89)
- Flow Matching Policy Optimization with Mirror Descent and Entropy Constraints (NeurIPS 2026 · similarity 0.89)
- Aligning Flow Map Policies with Optimal $Q$-Guidance (NeurIPS 2026 · similarity 0.89)
- Random Policy Evaluation Uncovers Policies of Generative Flow Networks (ICML 2025 · similarity 0.89)
- Flow Matching for Offline Reinforcement Learning with Discrete Actions (NeurIPS 2026 · similarity 0.88)
- Trust Region Q Adjoint Matching (NeurIPS 2026 · similarity 0.88)
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.
