Research map / Probabilistic and scientific modeling / Molecules and proteins
PolyFlow: Safe and Efficient Polytope-Constrained Flow Matching with Constraint Embedding and Projection-free Update
jianming ma, Qiyue Yang, Yang Zhang, Liyun Yan, Zhanxiang Cao, Yazhou Zhang, Yue Gao
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 |
|---|---|---|---|
| use MLP-pooling constraint encoder instead of attention encoder (ablation) | PolyFlow · 2D Maze (N=10 steps) | TotalTime (s): 2.11 -> 0.58 | Improved |
| use MLP encoder instead of attention encoder (ablation) | PolyFlow · 2D Maze (default steps) | MMD: 6.20 × 10−6 -> 2.62 × 10−6 | Improved |
| embed polytope constraints in architecture with ray shooting instead of post-hoc CBF correction | flow matching · 2D Maze (N=10 steps) | safety rate: 0.530 (SafeFlow) -> 1.0 | Improved |
| add PolyFlow constraint embedding to flow matching | Flow · 2D Maze (default steps) | MMD: 2.57 × 10−5 -> 2.62 × 10−6 | Improved |
| add PolyFlow safety constraints in Hopper-Simple | Flow · Hopper-Simple | rollout return: 2450 -> 3187 | Improved |
| add PolyFlow safety constraints in HalfCheetah | Flow · HalfCheetah | rollout return: 724 -> 2977 | Improved |
| add PolyFlow in Hopper-Complex | Flow · Hopper-Complex | safety rate: 0.005 -> 1.000 | Improved |
| remove explicit constraint embedding block (w/o) on static-constraint tasks (ablation) | PolyFlow · Maze2d, HalfCheetah | performance | No clear effect |
| add explicit constraint embedding on time-varying constraints (ablation) | PolyFlow · Unitree Go2 quadruped | performance | Improved |
| condition weight predictor on direction predictor output (coupled) (ablation) | PolyFlow · Gym tasks | performance | No clear effect |
| use smoothed ray shooting (Softmin, Boltzmann) instead of exact min (ablation) | PolyFlow · Gym tasks | performance and safety | Hurt |
| add OT-based batch coupling during training (ablation) | PolyFlow · Gym tasks | rollout return | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Flow — Flow++: Improving Flow-Based Generative Models with Variational Dequantization and Architecture Design (ICML 2019) · better on 8 of 9 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Physics-Constrained Flow Matching: Sampling Generative Models with Hard Constraints (NeurIPS 2025 · similarity 0.89)
- Gradient-Free Generation for Hard-Constrained Systems (ICLR 2025 · similarity 0.87)
- Optimal Flow Matching: Learning Straight Trajectories in Just One Step (NeurIPS 2024 · similarity 0.86)
- Primal-Dual Flow Matching for Sample-Wise Constrained Generation (NeurIPS 2026 · similarity 0.86)
- Chance-constrained Flow Matching for High-Fidelity Constraint-aware Generation (NeurIPS 2026 · similarity 0.86)
- On the Guidance of Flow Matching (ICML 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.
