Research map / Probabilistic and scientific modeling / Variational and generative sampling
Mirror Flow Matching with Heavy-Tailed Priors for Generative Modeling on Convex Domains
Yunrui Guan, Krishna Balasubramanian, Shiqian Ma
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 |
|---|---|---|---|
| replace Gaussian prior with Student-t prior in Mirror Flow (ablation) | Mirror Flow · 10-d polytope (mixture of Gaussians) | MMD (x1e-3): 5.774 -> 5.552 | Improved |
| replace Gaussian prior with Student-t prior in Mirror Flow (ablation) | Mirror Flow · 10-d polytope (mixture of Gaussians) | KL divergence: 0.379 -> 0.339 | Improved |
| replace Gaussian prior with Student-t prior in Mirror Flow (ablation) | Mirror Flow · 6-d L2 ball (mixture of Gaussians) | KL divergence (x1e-2): 9.644 -> 6.944 | Improved |
| replace Gaussian prior with Student-t prior in Mirror Flow (ablation) | Mirror Flow · 6-d L2 ball (mixture of Gaussians) | MMD (x1e-2): 1.221 -> 1.153 | Improved |
| use Mirror t-Flow instead of Reflected Flow Matching | constrained flow matching · 6-d L2 ball (mixture of Gaussians) | KL divergence (x1e-2): 14.827 -> 6.944 | Improved |
| use Mirror t-Flow instead of Reflected Flow Matching | constrained flow matching · 10-d polytope (mixture of Gaussians) | MMD (x1e-3): 5.868 -> 5.552 | Improved |
| use Mirror t-Flow instead of Gauge Vanilla | constrained flow matching · 10-d polytope (mixture of Gaussians) | feasibility: 90.653% -> 100% | Improved |
| use Mirror Flow instead of Mirror Diffusion Model, EDM-initialized | watermarked image generation · AFHQv2 64x64 watermarked (EDM init) | FID (50k): 7.29 -> 4.27 | Improved |
| use Mirror Flow instead of Mirror Diffusion Model, EDM-initialized | watermarked image generation · AFHQv2 64x64 watermarked (EDM init) | CMMD: 0.170 -> 0.023 | Improved |
| use Mirror Flow instead of Mirror Diffusion Model with random initialization | watermarked image generation · AFHQv2 64x64 watermarked (random init, 24h budget) | CMMD: 0.152 -> 0.177 | No clear effect |
Closest papers by meaning
- On the Closed-Form of Flow Matching: Generalization Does Not Arise from Target Stochasticity (NeurIPS 2025 · similarity 0.87)
- Optimal Flow Matching: Learning Straight Trajectories in Just One Step (NeurIPS 2024 · similarity 0.87)
- Coreset-Induced Conditional Velocity Flow Matching (NeurIPS 2026 · similarity 0.87)
- Reflected Flow Matching (ICML 2024 · similarity 0.87)
- CAR-Flow: Condition-Aware Reparameterization Aligns Source and Target for Better Flow Matching (NeurIPS 2025 · similarity 0.86)
- Energy Matching: Unifying Flow Matching and Energy-Based Models for Generative Modeling (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.
