Saved in:
Bibliographic Details
Main Authors: Gu, Yuxuan, Feng, Xiaocheng, Huang, Lei, Wu, Yingsheng, Zhou, Zekun, Zhong, Weihong, Zhu, Kun, Qin, Bing
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2410.22380
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866929567305826304
author Gu, Yuxuan
Feng, Xiaocheng
Huang, Lei
Wu, Yingsheng
Zhou, Zekun
Zhong, Weihong
Zhu, Kun
Qin, Bing
author_facet Gu, Yuxuan
Feng, Xiaocheng
Huang, Lei
Wu, Yingsheng
Zhou, Zekun
Zhong, Weihong
Zhu, Kun
Qin, Bing
contents We present an novel framework for efficiently and effectively extending the powerful continuous diffusion processes to discrete modeling. Previous approaches have suffered from the discrepancy between discrete data and continuous modeling. Our study reveals that the absence of guidance from discrete boundaries in learning probability contours is one of the main reasons. To address this issue, we propose a two-step forward process that first estimates the boundary as a prior distribution and then rescales the forward trajectory to construct a boundary conditional diffusion model. The reverse process is proportionally adjusted to guarantee that the learned contours yield more precise discrete data. Experimental results indicate that our approach achieves strong performance in both language modeling and discrete image generation tasks. In language modeling, our approach surpasses previous state-of-the-art continuous diffusion language models in three translation tasks and a summarization task, while also demonstrating competitive performance compared to auto-regressive transformers. Moreover, our method achieves comparable results to continuous diffusion models when using discrete ordinal pixels and establishes a new state-of-the-art for categorical image generation on the Cifar-10 dataset.
format Preprint
id arxiv_https___arxiv_org_abs_2410_22380
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Discrete Modeling via Boundary Conditional Diffusion Processes
Gu, Yuxuan
Feng, Xiaocheng
Huang, Lei
Wu, Yingsheng
Zhou, Zekun
Zhong, Weihong
Zhu, Kun
Qin, Bing
Machine Learning
Artificial Intelligence
We present an novel framework for efficiently and effectively extending the powerful continuous diffusion processes to discrete modeling. Previous approaches have suffered from the discrepancy between discrete data and continuous modeling. Our study reveals that the absence of guidance from discrete boundaries in learning probability contours is one of the main reasons. To address this issue, we propose a two-step forward process that first estimates the boundary as a prior distribution and then rescales the forward trajectory to construct a boundary conditional diffusion model. The reverse process is proportionally adjusted to guarantee that the learned contours yield more precise discrete data. Experimental results indicate that our approach achieves strong performance in both language modeling and discrete image generation tasks. In language modeling, our approach surpasses previous state-of-the-art continuous diffusion language models in three translation tasks and a summarization task, while also demonstrating competitive performance compared to auto-regressive transformers. Moreover, our method achieves comparable results to continuous diffusion models when using discrete ordinal pixels and establishes a new state-of-the-art for categorical image generation on the Cifar-10 dataset.
title Discrete Modeling via Boundary Conditional Diffusion Processes
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2410.22380