Computation and Visualization of Multi-Parameter Topological Invariants of Data

点击翻译按钮获取中文摘要

Topological data analysis (TDA) is a relatively new branch of statistics whose goal is to apply topology to develop tools for studying the coarse-scale, global, non-linear, geometric features of data. Persistent homology, the most widely studied tool for TDA, has been applied to many areas of science and engineering, including image processing, time series data in biological systems, and sensor networks. Persistent homology yields invariants of data, called barcodes, by associating to the data a sequence of nested topological spaces called a filtration, and then applying standard topological and algebraic constructions. However, for many data sets of interest, such as point cloud data with noise or non-uniformities in density, a single filtration is not rich enough to encode the structure of interest in the data. This motivates the consideration of multidimensional persistent homology, which associates to the data a topological space simultaneously equipped with two or more filtrations. Multi-D persistent homology yields algebraic invariants of data far more complex than in the 1-D setting. New methodology is thus required for working with these invariants in practice. The goal of this project is to introduce such methodology in the 2-D setting.Specifically, this project is to develop algorithms and design practical software tools that extend the usual persistent homology methodology for exploratory data analysis to the 2-D setting. The proposed tools provide an interactive visualization of the barcodes of the restriction of a 2-D persistence module to affine 1-D lines. At the heart of the computational approach is a novel data structure, based on planar line arrangements, on which one can perform fast queries for these barcodes. The tools also provide a visualization of the multi-graded Betti numbers of a 2-D persistence module. It is proposed to apply the tools to the study of scientific data - especially data arising from biological systems - in much the same way that ordinary persistent homology has been applied to the study of data over the last ten to fifteen years. This project will intend to establish statistical foundations for the corresponding methodology.