Mirror Flow Matching with Heavy-Tailed Priors for Generative Modeling on Convex Domains
Authors:
Preprint
Abstract
We study generative modeling on convex domains using flow matching and mirror maps, and identify two fundamental challenges. First, standard log-barrier mirror maps induce heavy-tailed dual distributions, leading to ill-posed dynamics. Second, coupling with Gaussian priors performs poorly when matching heavy-tailed targets. To address these issues, we propose Mirror Flow Matching based on a \emph{regularized mirror map} that controls dual tail behavior and guarantees finite moments, together with coupling to a Student-$t$ prior that aligns with heavy-tailed targets and stabilizes training. We provide theoretical guarantees, including spatial Lipschitzness and temporal regularity of the velocity field, Wasserstein convergence rates for flow matching with Student-$t$ priors and primal-space guarantees for constrained generation, under $\varepsilon$-accurate learned velocity fields. Empirically, our method outperforms baselines in synthetic convex-domain simulations and achieves competitive sample quality on real-world constrained generative tasks.Keywords
All Related Versions
This publication has 8 references indexed in Scilit:
- Preference Alignment with Flow Matching2024
- Constrained Diffusion Models via Dual Training2024
- Constrained Synthesis with Projected Diffusion Models2024
- Improving the Training of Rectified Flows2024
- Metropolis Sampling for Constrained Diffusion Models2023
- Mirror Diffusion Models for Constrained and Watermarked Generation2023
- The probability flow ODE is provably fast2023
- Elucidating the Design Space of Diffusion-Based Generative Models2022