arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2408.16765cs.LGcs.AImath.PRmath.STstat.MLstat.TH
收藏

一种基于分数的密度公式及其在扩散生成模型中的应用

A Score-Based Density Formula, with Applications in Diffusion Generative Models

The Chinese University of Hong Kong(香港中文大学);University of Wisconsin-Madison(威斯康星大学麦迪逊分校)

Gen Li, Yuling Yan

AI总结:

本文通过建立连续时间扩散过程的密度公式,揭示了目标密度与分数函数的关系,并证明ELBO优化目标与真实目标的最小化点几乎一致,为DDPM等扩散生成模型的训练提供了理论基础。

摘要:

基于分数的生成模型(SGMs)已经彻底改变了生成建模领域,在生成逼真且多样化的内容方面取得了前所未有的成功。尽管在经验上取得了诸多进展,但关于为何在对数似然上优化证据下界(ELBO)对于训练扩散生成模型(如DDPMs)是有效的这一问题的理论基础,在很大程度上仍未得到探索。在本文中,我们通过为连续时间扩散过程建立一个密度公式来回答这一问题,该公式可被视为SGM中前向过程的连续时间极限。该公式揭示了目标密度与前向过程每一步所关联的分数函数之间的联系。在此基础上,我们证明了训练DDPMs的优化目标的最小化点几乎与真实目标的最小化点一致,从而为使用ELBO优化DDPMs提供了理论基础。此外,我们还为训练GANs中分数匹配正则化的作用、扩散分类器中ELBO的使用以及最近提出的扩散损失提供了新的见解。

英文摘要:

Score-based generative models (SGMs) have revolutionized the field of generative modeling, achieving unprecedented success in generating realistic and diverse content. Despite empirical advances, the theoretical basis for why optimizing the evidence lower bound (ELBO) on the log-likelihood is effective for training diffusion generative models, such as DDPMs, remains largely unexplored. In this paper, we address this question by establishing a density formula for a continuous-time diffusion process, which can be viewed as the continuous-time limit of the forward process in an SGM. This formula reveals the connection between the target density and the score function associated with each step of the forward process. Building on this, we demonstrate that the minimizer of the optimization objective for training DDPMs nearly coincides with that of the true objective, providing a theoretical foundation for optimizing DDPMs using the ELBO. Furthermore, we offer new insights into the role of score-matching regularization in training GANs, the use of ELBO in diffusion classifiers, and the recently proposed diffusion loss.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑