Research map / Probabilistic and scientific modeling / Molecules and proteins
Multimarginal flow matching with optimal transport potentials
Raghav Kansal, David Crair, Nghia D. Nguyen, Scott D Pope, Bradley R Parry
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 W2^2 potential instead of W2^inf (ablation) | OTP-FM · EB 100D L2O | mean MMD on held-out times: 0.0813 -> 0.0675 | Improved |
| use W2^2 potential instead of MMD potential (ablation) | OTP-FM · EB 100D L2O | mean MMD on held-out times: 0.0861 -> 0.0675 | Improved |
| use W2^2 potential instead of KL potential (ablation) | OTP-FM · EB 100D L2O | mean MMD on held-out times: 0.0876 -> 0.0675 | Improved |
| raise potential strength from w=10 to w=1000 (ablation) | OTP-FM · EB 100D L2O | mean MMD on held-out times: 0.1431 -> 0.0675 | Improved |
| raise potential strength from w=1000 to w=10000 (ablation) | OTP-FM · EB 100D L2O | mean MMD on held-out times: 0.0675 -> 0.0683 | No clear effect |
| use Gaussian window instead of triangle (ablation) | OTP-FM lambda(t) · EB 100D L2O | mean MMD on held-out times: 0.0723 -> 0.0675 | Improved |
| use Gaussian window instead of box (ablation) | OTP-FM lambda(t) · EB 100D L2O | mean MMD on held-out times: 0.0681 -> 0.0675 | No clear effect |
| use sigmoid curriculum instead of linear (ablation) | OTP-FM · EB 100D L2O | mean MMD on held-out times: 0.0800 -> 0.0675 | Improved |
| use sigmoid curriculum instead of constant (ablation) | OTP-FM · EB 100D L2O | mean MMD on held-out times: 0.0687 -> 0.0675 | Improved |
| shift sigmoid midpoint to mu=1 (ablation) | OTP-FM curriculum · EB 100D L2O | mean MMD on held-out times: 0.0675 -> 0.0758 | Hurt |
| vary window width tau between 0.2 and 0.5 (ablation) | OTP-FM · EB 100D L2O | mean MMD on held-out times: 0.0690, 0.0675, 0.0693 | No clear effect |
| use OTP-FM (W2^2) instead of MMFM | trajectory model · Gulf of Mexico (2D, 9 timepoints) | held-out average W2: 0.150 -> 0.107 | Improved |
Closest papers by meaning
- Curly Flow Matching for Learning Non-gradient Field Dynamics (NeurIPS 2025 · similarity 0.90)
- PrismFlow: Residual Dynamics for Flow Matching in Time-Series Generation (NeurIPS 2026 · similarity 0.89)
- Flow Matching from Viewpoint of Proximal Operators (NeurIPS 2026 · similarity 0.89)
- Pairwise Optimal Transports for Training All-to-All Flow-Based Condition Transfer Model (NeurIPS 2025 · similarity 0.88)
- Optimal Flow Matching: Learning Straight Trajectories in Just One Step (NeurIPS 2024 · similarity 0.88)
- Diff2Flow: Training Flow Matching Models via Diffusion Model Alignment (CVPR 2025 · 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.
