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 2503.01006cs.LGstat.ML
收藏

欠阻尼扩散桥及其在采样中的应用

Underdamped Diffusion Bridges with Applications to Sampling

Karlsruhe Institute of Technology(卡尔斯鲁厄理工学院);NVIDIA(NVIDIA公司);Zuse Institute Berlin(柏林泽尼克研究所);dida Datenschmiede GmbH(dida数据安全公司);FZI Research Center for Information Technology(信息技术研究中心)

Denis Blessing, Julius Berner, Lorenz Richter, Gerhard Neumann

AI总结:

本文提出了一个学习扩散桥的通用框架,包含具有退化扩散矩阵的欠阻尼版本,并严格证明了欠阻尼情况下 score matching 等价于最大化似然下界,应用于无目标样本的非归一化密度采样时取得了更优性能且无需调参。

摘要:

我们提供了一个学习扩散桥的通用框架,该框架将先验分布传输到目标分布。它包含了用于生成式建模的现有扩散模型,但也包含了具有退化扩散矩阵的欠阻尼版本,其中噪声仅作用于某些维度。扩展了先前的发现,我们的框架允许严格证明在欠阻尼情况下,score matching 确实等价于最大化似然的下界。受欠阻尼随机过程卓越的收敛性质以及与复杂数值积分方案的兼容性所驱动,我们提出了 underdamped diffusion bridges,其中学习的是一般的密度演化,而不是由固定的加噪过程所规定。我们将我们的方法应用于从非归一化密度中采样的挑战性任务,且无法获取目标分布的样本。在多种采样问题中,我们的方法展现了 state-of-the-art 的性能,显著优于其他替代方法,同时需要更少的离散化步骤且无需超参数调优。

英文摘要:

We provide a general framework for learning diffusion bridges that transport prior to target distributions. It includes existing diffusion models for generative modeling, but also underdamped versions with degenerate diffusion matrices, where the noise only acts in certain dimensions. Extending previous findings, our framework allows to rigorously show that score matching in the underdamped case is indeed equivalent to maximizing a lower bound on the likelihood. Motivated by superior convergence properties and compatibility with sophisticated numerical integration schemes of underdamped stochastic processes, we propose \emph{underdamped diffusion bridges}, where a general density evolution is learned rather than prescribed by a fixed noising process. We apply our method to the challenging task of sampling from unnormalized densities without access to samples from the target distribution. Across a diverse range of sampling problems, our approach demonstrates state-of-the-art performance, notably outperforming alternative methods, while requiring significantly fewer discretization steps and no hyperparameter tuning.

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。

↑