Multi-scale Generative Modeling for Fast Sampling
Abstract
While working within the spatial domain can pose problems associated with ill-conditioned scores caused by power-law decay, recent advances in diffusion-based generative models have shown that transitioning to the wavelet domain offers a promising alternative. However, within the wavelet domain, we encounter unique challenges, especially the sparse representation of high-frequency coefficients, which deviates significantly from the Gaussian assumptions in the diffusion process. To this end, we propose a multi-scale generative modeling in the wavelet domain that employs distinct strategies for handling low and high-frequency bands. In the wavelet domain, we apply score-based generative modeling with well-conditioned scores for low-frequency bands, while utilizing a multi-scale generative adversarial learning for high-frequency bands. As supported by the theoretical analysis and experimental results, our model significantly improve performance and reduce the number of trainable parameters, sampling steps, and time.
Cite
@article{arxiv.2411.09356,
title = {Multi-scale Generative Modeling for Fast Sampling},
author = {Xiongye Xiao and Shixuan Li and Luzhe Huang and Gengshuo Liu and Trung-Kien Nguyen and Yi Huang and Di Chang and Mykel J. Kochenderfer and Paul Bogdan},
journal= {arXiv preprint arXiv:2411.09356},
year = {2024}
}