Filtration Surfaces for Dynamic Graph Classification

TL;DR

This paper introduces filtration surfaces, a scalable and flexible method for dynamic graph classification that effectively incorporates edge weight information, outperforming existing baselines with minimal parameters and high stability.

Contribution

We propose filtration surfaces, a novel, scalable, and parameter-efficient approach for dynamic graph classification that handles changing node sets and edge weights.

Findings

Outperforms previous state-of-the-art baselines on weighted dynamic graph datasets.

Is scalable and flexible, handling changing node sets and edge weights.

Achieves the lowest standard deviation among comparable methods.

Abstract

Existing approaches for classifying dynamic graphs either lift graph kernels to the temporal domain, or use graph neural networks (GNNs). However, current baselines have scalability issues, cannot handle a changing node set, or do not take edge weight information into account. We propose filtration surfaces, a novel method that is scalable and flexible, to alleviate said restrictions. We experimentally validate the efficacy of our model and show that filtration surfaces outperform previous state-of-the-art baselines on datasets that rely on edge weight information. Our method does so while being either completely parameter-free or having at most one parameter, and yielding the lowest overall standard deviation among similarly scalable methods.