arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2501.17049math.APcs.LGmath.OCstat.ML
收藏

Hellinger-Kantorovich 梯度流:熵泛函的全局指数衰减

Hellinger-Kantorovich Gradient Flows: Global Exponential Decay of Entropy Functionals

Humboldt University of Berlin(柏林洪堡大学);Weierstrass Institute for Applied Analysis and Stochastics(魏尔斯特拉斯应用分析与随机学研究所)

Alexander Mielke, Jia-Jie Zhu

AI总结:

研究 Hellinger-Kantorovich 几何下正测度与概率测度梯度流,通过形状-质量分解和泛函不等式完整刻画了熵泛函的全局指数衰减行为,为统计推断、优化与机器学习提供了统一理论框架。

摘要:

我们研究了正测度与概率测度的一族梯度流,重点关注 Hellinger-Kantorovich (HK) 几何,它统一了 Otto-Wasserstein 的输运机制与 Hellinger(或 Fisher-Rao)的生灭机制。核心贡献是完整刻画了在 Otto-Wasserstein 和 Hellinger 型梯度流下熵泛函(如 KL、$χ^2$)的全局指数衰减行为。特别地,对于正测度上更具挑战性的 HK 梯度流分析——此处典型的 log-Sobolev 论证失效——我们开发了一种专门的形状-质量分解,从而得出了新的分析结果。该方法还利用了 (Polyak-)Łojasiewicz 型泛函不等式以及对经典耗散估计的仔细扩展。这些发现为梯度流提供了统一且完整的理论框架,并支撑了统计推断、优化和机器学习中计算算法的应用。

英文摘要:

We investigate a family of gradient flows of positive and probability measures, focusing on the Hellinger-Kantorovich (HK) geometry, which unifies transport mechanism of Otto-Wasserstein, and the birth-death mechanism of Hellinger (or Fisher-Rao). A central contribution is a complete characterization of global exponential decay behaviors of entropy functionals (e.g. KL, $χ^2$) under Otto-Wasserstein and Hellinger-type gradient flows. In particular, for the more challenging analysis of HK gradient flows on positive measures -- where the typical log-Sobolev arguments fail -- we develop a specialized shape-mass decomposition that enables new analysis results. Our approach also leverages the (Polyak-)Łojasiewicz-type functional inequalities and a careful extension of classical dissipation estimates. These findings provide a unified and complete theoretical framework for gradient flows and underpin applications in computational algorithms for statistical inference, optimization, and machine learning.

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。

↑