Guardado en:
Detalles Bibliográficos
Autores principales: Pandey, Kushagra, Pathak, Jaideep, Xu, Yilun, Mandt, Stephan, Pritchard, Michael, Vahdat, Arash, Mardani, Morteza
Formato: Preprint
Publicado: 2024
Materias:
Acceso en línea:https://arxiv.org/abs/2410.14171
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866910676480425984
author Pandey, Kushagra
Pathak, Jaideep
Xu, Yilun
Mandt, Stephan
Pritchard, Michael
Vahdat, Arash
Mardani, Morteza
author_facet Pandey, Kushagra
Pathak, Jaideep
Xu, Yilun
Mandt, Stephan
Pritchard, Michael
Vahdat, Arash
Mardani, Morteza
contents Diffusion models achieve state-of-the-art generation quality across many applications, but their ability to capture rare or extreme events in heavy-tailed distributions remains unclear. In this work, we show that traditional diffusion and flow-matching models with standard Gaussian priors fail to capture heavy-tailed behavior. We address this by repurposing the diffusion framework for heavy-tail estimation using multivariate Student-t distributions. We develop a tailored perturbation kernel and derive the denoising posterior based on the conditional Student-t distribution for the backward process. Inspired by $γ$-divergence for heavy-tailed distributions, we derive a training objective for heavy-tailed denoisers. The resulting framework introduces controllable tail generation using only a single scalar hyperparameter, making it easily tunable for diverse real-world distributions. As specific instantiations of our framework, we introduce t-EDM and t-Flow, extensions of existing diffusion and flow models that employ a Student-t prior. Remarkably, our approach is readily compatible with standard Gaussian diffusion models and requires only minimal code changes. Empirically, we show that our t-EDM and t-Flow outperform standard diffusion models in heavy-tail estimation on high-resolution weather datasets in which generating rare and extreme events is crucial.
format Preprint
id arxiv_https___arxiv_org_abs_2410_14171
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Heavy-Tailed Diffusion Models
Pandey, Kushagra
Pathak, Jaideep
Xu, Yilun
Mandt, Stephan
Pritchard, Michael
Vahdat, Arash
Mardani, Morteza
Machine Learning
Diffusion models achieve state-of-the-art generation quality across many applications, but their ability to capture rare or extreme events in heavy-tailed distributions remains unclear. In this work, we show that traditional diffusion and flow-matching models with standard Gaussian priors fail to capture heavy-tailed behavior. We address this by repurposing the diffusion framework for heavy-tail estimation using multivariate Student-t distributions. We develop a tailored perturbation kernel and derive the denoising posterior based on the conditional Student-t distribution for the backward process. Inspired by $γ$-divergence for heavy-tailed distributions, we derive a training objective for heavy-tailed denoisers. The resulting framework introduces controllable tail generation using only a single scalar hyperparameter, making it easily tunable for diverse real-world distributions. As specific instantiations of our framework, we introduce t-EDM and t-Flow, extensions of existing diffusion and flow models that employ a Student-t prior. Remarkably, our approach is readily compatible with standard Gaussian diffusion models and requires only minimal code changes. Empirically, we show that our t-EDM and t-Flow outperform standard diffusion models in heavy-tail estimation on high-resolution weather datasets in which generating rare and extreme events is crucial.
title Heavy-Tailed Diffusion Models
topic Machine Learning
url https://arxiv.org/abs/2410.14171