Local Curvature Smoothing with Stein's Identity for Efficient Score Matching

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Osada, Genki, Shing, Makoto, Nishide, Takashi
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917858791915520
author Osada, Genki
Shing, Makoto
Nishide, Takashi
author_facet Osada, Genki
Shing, Makoto
Nishide, Takashi
contents The training of score-based diffusion models (SDMs) is based on score matching. The challenge of score matching is that it includes a computationally expensive Jacobian trace. While several methods have been proposed to avoid this computation, each has drawbacks, such as instability during training and approximating the learning as learning a denoising vector field rather than a true score. We propose a novel score matching variant, local curvature smoothing with Stein's identity (LCSS). The LCSS bypasses the Jacobian trace by applying Stein's identity, enabling regularization effectiveness and efficient computation. We show that LCSS surpasses existing methods in sample generation performance and matches the performance of denoising score matching, widely adopted by most SDMs, in evaluations such as FID, Inception score, and bits per dimension. Furthermore, we show that LCSS enables realistic image generation even at a high resolution of $1024 \times 1024$.
format Preprint
id arxiv_https___arxiv_org_abs_2412_03962
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Local Curvature Smoothing with Stein's Identity for Efficient Score Matching
Osada, Genki
Shing, Makoto
Nishide, Takashi
Machine Learning
Computer Vision and Pattern Recognition
The training of score-based diffusion models (SDMs) is based on score matching. The challenge of score matching is that it includes a computationally expensive Jacobian trace. While several methods have been proposed to avoid this computation, each has drawbacks, such as instability during training and approximating the learning as learning a denoising vector field rather than a true score. We propose a novel score matching variant, local curvature smoothing with Stein's identity (LCSS). The LCSS bypasses the Jacobian trace by applying Stein's identity, enabling regularization effectiveness and efficient computation. We show that LCSS surpasses existing methods in sample generation performance and matches the performance of denoising score matching, widely adopted by most SDMs, in evaluations such as FID, Inception score, and bits per dimension. Furthermore, we show that LCSS enables realistic image generation even at a high resolution of $1024 \times 1024$.
title Local Curvature Smoothing with Stein's Identity for Efficient Score Matching
topic Machine Learning
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2412.03962