Research map / Probabilistic and scientific modeling / Variational and generative sampling
Multi-Marginal Flow Matching with Adversarially Learnt Interpolants
Oskar Kviman, Kirill Tamogashev, Nicola Branchini, Víctor Elvira, Jens Lagergren, Nikolay Malkin
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 adversarially learnt interpolants instead of piecewise linear interpolants (OT-CFM) | OT-CFM · spatial transcriptomics | EMD: 109.76 -> 98.91 | Improved |
| use adversarially learnt interpolants instead of the LAND-metric interpolants (OT-MFM) | OT-MFM · spatial transcriptomics | EMD: 183.88 -> 98.91 | Improved |
| use ALI instead of OT-CFM on Cite 5D | OT-CFM · Cite scRNA-seq (5D PCA) | EMD: 1.142 -> 0.910 | Improved |
| use ALI instead of OT-CFM on EB 5D | OT-CFM · EB scRNA-seq (5D PCA) | EMD: 0.809 -> 0.742 | Improved |
| use ALI instead of OT-MFM on Cite 5D | OT-MFM · Cite scRNA-seq (5D PCA) | EMD: 0.793 -> 0.910 | Hurt |
| use ALI instead of cubic spline interpolants (OT-MMFM) on EB 5D | OT-MMFM · EB scRNA-seq (5D PCA) | EMD: 3.530 -> 0.742 | Improved |
| use ALI instead of OT-MFM on Cite 50D | OT-MFM · Cite scRNA-seq (50D PCA) | EMD: 36.471 -> 41.449 | Hurt |
| use ALI instead of OT-CFM on Cite 50D | OT-CFM · Cite scRNA-seq (50D PCA) | EMD: 38.367 -> 41.449 | Hurt |
| use ALI instead of OT-CFM on Multi 100D | OT-CFM · Multi scRNA-seq (100D PCA) | EMD: 54.630 -> 54.554 | No clear effect |
| increase piecewise-linear regulariser coefficient lambda from 0.01 to 0.05 (ablation) | OT-ALI-CFM · Cite 50D, left-out t=1 | CFM EMD: 47.29 -> 41.07 | Improved |
| increase piecewise-linear regulariser coefficient lambda from 0.01 to 10 (ablation) | OT-ALI-CFM · Cite 50D, left-out t=1 | CFM path energy: 300.01 -> 87052.14 | Hurt |
| increase second-derivative regulariser coefficient from 1e-9 to 1e-6 (ablation) | OT-ALI-CFM · Cite 50D, left-out t=1 | CFM path energy: 56.39 -> 12.30 | Improved |
Closest papers by meaning
- Metric Flow Matching for Smooth Interpolations on the Data Manifold (NeurIPS 2024 · similarity 0.88)
- Blockwise Flow Matching: Improving Flow Matching Models For Efficient High-Quality Generation (NeurIPS 2025 · similarity 0.87)
- Shortcutting Pre-trained Flow Matching Diffusion Models is Almost Free Lunch (NeurIPS 2025 · similarity 0.86)
- Diff2Flow: Training Flow Matching Models via Diffusion Model Alignment (CVPR 2025 · similarity 0.86)
- On the Closed-Form of Flow Matching: Generalization Does Not Arise from Target Stochasticity (NeurIPS 2025 · similarity 0.86)
- Structured Coupling for Flow Matching (NeurIPS 2026 · 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.
