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 2402.04516stat.MLcs.LG
收藏

图上概率测度的广义Sobolev传输

Generalized Sobolev Transport for Probability Measures on a Graph

The Institute of Statistical Mathematics (ISM)(统计数学研究所);RIKEN AIP(理化学研究所人工智能项目);The University of Akron(阿克伦大学)

Tam Le, Truyen Nguyen, Kenji Fukumizu

AI总结:

本文提出广义Sobolev传输(GST),通过利用Orlicz几何结构克服了ST仅限于$L^p$结构的缺陷,仅需单变量优化即可高效计算,在文档分类等任务上展现出显著优势。

摘要:

我们研究了支撑在图度量空间上的测度的最优传输(OT)问题。最近,Le等人(2022)利用图结构提出了OT的一个变体,即Sobolev传输(ST),它产生了一个闭式表达式以进行快速计算。然而,ST在其定义中本质上与$L^p$几何结构耦合,这使得将ST用于其他先验结构变得不平凡。相比之下,经典的OT具有通过修改底层成本函数来适应各种几何结构的灵活性。一个重要的实例是Orlicz-Wasserstein(OW),它通过利用Orlicz几何结构超越了$L^p$结构。与使用标准的$p$阶Wasserstein相比,OW显著有助于推进某些机器学习方法。然而,由于其双层优化公式,OW在计算上带来了新的挑战。在这项工作中,我们利用Orlicz结构的一类特定凸函数,提出了广义Sobolev传输(GST)。GST将ST作为其特例包含在内,并可用于$L^p$几何之外的先验结构。与OW相联系,我们表明只需简单求解一个单变量优化问题即可计算GST,而不像OW中复杂的双层优化问题。我们从经验上说明GST比OW快几个数量级。此外,我们为GST在文档分类和拓扑数据分析中的若干任务上的优势提供了初步证据。

英文摘要:

We study the optimal transport (OT) problem for measures supported on a graph metric space. Recently, Le et al. (2022) leverage the graph structure and propose a variant of OT, namely Sobolev transport (ST), which yields a closed-form expression for a fast computation. However, ST is essentially coupled with the $L^p$ geometric structure within its definition which makes it nontrivial to utilize ST for other prior structures. In contrast, the classic OT has the flexibility to adapt to various geometric structures by modifying the underlying cost function. An important instance is the Orlicz-Wasserstein (OW) which moves beyond the $L^p$ structure by leveraging the \emph{Orlicz geometric structure}. Comparing to the usage of standard $p$-order Wasserstein, OW remarkably helps to advance certain machine learning approaches. Nevertheless, OW brings up a new challenge on its computation due to its two-level optimization formulation. In this work, we leverage a specific class of convex functions for Orlicz structure to propose the generalized Sobolev transport (GST). GST encompasses the ST as its special case, and can be utilized for prior structures beyond the $L^p$ geometry. In connection with the OW, we show that one only needs to simply solve a univariate optimization problem to compute the GST, unlike the complex two-level optimization problem in OW. We empirically illustrate that GST is several-order faster than the OW. Moreover, we provide preliminary evidences on the advantages of GST for document classification and for several tasks in topological data analysis.

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。

↑