arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2510.08409stat.MLcs.LG
收藏

潜在扩散模型中的最优停止

Optimal Stopping in Latent Diffusion Models

Google DeepMind(谷歌DeepMind);INRIA, ENS-PSL, Paris(INRIA、巴黎大学-PSL)

Yu-Han Wu, Quentin Berthet, Gérard Biau, Claire Boyer, Romuald Elie, Pierre Marion

AI总结:

研究揭示了潜在扩散模型中潜在维度与停止时间对样本质量的影响,提出通过调整停止时间优化生成质量的方法。

摘要:

我们识别并分析了潜在扩散模型(LDMs)中一个令人惊讶的现象,即扩散过程的最后几步会降低样本质量。与传统观点认为早期停止是为了数值稳定性不同,这种现象本质上源于LDMs中的降维。我们通过分析潜在维度与停止时间的相互作用,提供了一个系统性的解释。在高斯框架下使用线性自编码器,我们刻画了在生成分布与目标分布之间距离最小化时,早期停止所需的条件。更具体地说,我们表明低维表示受益于更早的终止,而高维潜在空间则需要更晚的停止时间。我们进一步建立了潜在维度与其他问题超参数(如分数匹配参数的约束)之间的相互作用。在合成和真实数据集上的实验验证了这些性质,强调早期停止可以提高生成质量。总的来说,我们的结果为理解潜在维度如何影响样本质量提供了理论基础,并突显了停止时间作为LDMs中的关键超参数。

英文摘要:

We identify and analyze a surprising phenomenon of Latent Diffusion Models (LDMs) where the final steps of the diffusion can degrade sample quality. In contrast to conventional arguments that justify early stopping for numerical stability, this phenomenon is intrinsic to the dimensionality reduction in LDMs. We provide a principled explanation by analyzing the interaction between latent dimension and stopping time. Under a Gaussian framework with linear autoencoders, we characterize the conditions under which early stopping is needed to minimize the distance between generated and target distributions. More precisely, we show that lower-dimensional representations benefit from earlier termination, whereas higher-dimensional latent spaces require later stopping time. We further establish that the latent dimension interplays with other hyperparameters of the problem such as constraints in the parameters of score matching. Experiments on synthetic and real datasets illustrate these properties, underlining that early stopping can improve generative quality. Together, our results offer a theoretical foundation for understanding how the latent dimension influences the sample quality, and highlight stopping time as a key hyperparameter in LDMs.

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。

↑