Improving homology estimates with random walks

Abstract

This experimental paper makes the case for a new approach to the use of persistent homology in the study of shape and feature in datasets. By introducing ideas from diffusion geometry and random walks, we discover that homological features can be enhanced and more effectively extracted from spaces that are sampled densely and evenly, and with a small amount of noise. This study paves the way for a more theoretical analysis of how random walk metrics affect persistence diagrams, and provides evidence that combining topological data analysis with techniques inspired by diffusion geometry holds great promise for new analyses of a wide variety of datasets.