DistillKac: Few-Step Image Generation via Damped Wave Equations

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Han, Weiqiao, Meng, Chenlin, Manning, Christopher D., Ermon, Stefano
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917304362598400
author Han, Weiqiao
Meng, Chenlin
Manning, Christopher D.
Ermon, Stefano
author_facet Han, Weiqiao
Meng, Chenlin
Manning, Christopher D.
Ermon, Stefano
contents We present DistillKac, a fast image generator that uses the damped wave equation and its stochastic Kac representation to move probability mass at finite speed. In contrast to diffusion models whose reverse time velocities can become stiff and implicitly allow unbounded propagation speed, Kac dynamics enforce finite speed transport and yield globally bounded kinetic energy. Building on this structure, we introduce classifier-free guidance in velocity space that preserves square integrability under mild conditions. We then propose endpoint only distillation that trains a student to match a frozen teacher over long intervals. We prove a stability result that promotes supervision at the endpoints to closeness along the entire path. Experiments demonstrate DistillKac delivers high quality samples with very few function evaluations while retaining the numerical stability benefits of finite speed probability flows.
format Preprint
id arxiv_https___arxiv_org_abs_2509_21513
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle DistillKac: Few-Step Image Generation via Damped Wave Equations
Han, Weiqiao
Meng, Chenlin
Manning, Christopher D.
Ermon, Stefano
Machine Learning
Artificial Intelligence
Computer Vision and Pattern Recognition
Probability
We present DistillKac, a fast image generator that uses the damped wave equation and its stochastic Kac representation to move probability mass at finite speed. In contrast to diffusion models whose reverse time velocities can become stiff and implicitly allow unbounded propagation speed, Kac dynamics enforce finite speed transport and yield globally bounded kinetic energy. Building on this structure, we introduce classifier-free guidance in velocity space that preserves square integrability under mild conditions. We then propose endpoint only distillation that trains a student to match a frozen teacher over long intervals. We prove a stability result that promotes supervision at the endpoints to closeness along the entire path. Experiments demonstrate DistillKac delivers high quality samples with very few function evaluations while retaining the numerical stability benefits of finite speed probability flows.
title DistillKac: Few-Step Image Generation via Damped Wave Equations
topic Machine Learning
Artificial Intelligence
Computer Vision and Pattern Recognition
Probability
url https://arxiv.org/abs/2509.21513