arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2108.02727math.ATmath.FAstat.OT
收藏

签名、Lipschitz-free 空间与持久性图路径

Signatures, Lipschitz-free spaces, and paths of persistence diagrams

Chad Giusti, Darrick Lee

AI总结:

本文通过将持久性图嵌入Lipschitz-free空间,利用路径签名和矩映射构建可计算特征,用于动态拓扑数据分析,并在群体行为参数估计中验证有效性。

摘要:

持久性图的路径提供了一族单参数度量空间的动态拓扑结构的摘要。这些摘要可用于研究和表征数据的动态形状,例如多智能体系统中的群体行为、神经科学中的时变 fMRI 扫描以及流体动力学中的时间相关标量场。虽然持久性图可以提供强大的数据拓扑摘要,但标准的持久性图空间缺乏许多理论和计算分析所需的足够的代数和解析结构。我们通过将持久性图空间等距嵌入到 Lipschitz-free 空间(一种由通用构造构建的 Banach 空间)中来丰富它。我们利用 Banach 空间结构定义有界变差持久性图路径,这可以使用路径签名(一种对 Banach 空间值路径的重参数化不变表征)来研究。该签名是通用的且具有特征性,这使我们能够在理论上表征路径空间上的测度,并推动其在核方法中的应用。然而,核方法通常需要映射到希尔伯特空间的特征图,因此我们引入了矩映射,一种用于静态持久性图的稳定且单射的特征图,并将其与离散路径签名组合,产生一个可计算的映射到希尔伯特空间的特征图。最后,我们通过将其应用于三维群体行为模型的参数估计问题来证明我们方法的有效性。

英文摘要:

Paths of persistence diagrams provide a summary of the dynamic topological structure of a one-parameter family of metric spaces. These summaries can be used to study and characterize the dynamic shape of data such as swarming behavior in multi-agent systems, time-varying fMRI scans from neuroscience, and time-dependent scalar fields in hydrodynamics. While persistence diagrams can provide a powerful topological summary of data, the standard space of persistence diagrams lacks the sufficient algebraic and analytic structure required for many theoretical and computational analyses. We enrich the space of persistence diagrams by isometrically embedding it into a Lipschitz-free space, a Banach space built from a universal construction. We utilize the Banach space structure to define bounded variation paths of persistence diagrams, which can be studied using the path signature, a reparametrization-invariant characterization of paths valued in a Banach space. The signature is universal and characteristic which allows us to theoretically characterize measures on the space of paths and motivates its use in the context of kernel methods. However, kernel methods often require a feature map into a Hilbert space, so we introduce the moment map, a stable and injective feature map for static persistence diagrams, and compose it with the discrete path signature, producing a computable feature map into a Hilbert space. Finally, we demonstrate the efficacy of our methods by applying this to a parameter estimation problem for a 3D model of swarming behavior.

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。

↑