Optimisation of Spectral Wavelets for Persistence-based Graph Classification

TL;DR

This paper introduces a framework to optimize spectral wavelets for graph classification, enhancing the ability of persistence diagrams to capture relevant graph features without prior node attributes.

Contribution

It develops a method to optimize wavelet choices based on datasets, extending differentiability results to extended persistent homology for theoretical support.

Findings

Competitive performance with existing persistence-based methods

Framework encodes geometric properties of graphs

Applicable to graphs without node attributes

Abstract

A graph's spectral wavelet signature determines a filtration, and consequently an associated set of extended persistence diagrams. We propose a framework that optimises the choice of wavelet for a dataset of graphs, such that their associated persistence diagrams capture features of the graphs that are best suited to a given data science problem. Since the spectral wavelet signature of a graph is derived from its Laplacian, our framework encodes geometric properties of graphs in their associated persistence diagrams and can be applied to graphs without a priori node attributes. We apply our framework to graph classification problems and obtain performances competitive with other persistence-based architectures. To provide the underlying theoretical foundations, we extend the differentiability result for ordinary persistent homology to extended persistent homology.