English

Multi-class Graph Clustering via Approximated Effective $p$-Resistance

Machine Learning 2023-07-20 v2 Machine Learning

Abstract

This paper develops an approximation to the (effective) pp-resistance and applies it to multi-class clustering. Spectral methods based on the graph Laplacian and its generalization to the graph pp-Laplacian have been a backbone of non-euclidean clustering techniques. The advantage of the pp-Laplacian is that the parameter pp induces a controllable bias on cluster structure. The drawback of pp-Laplacian eigenvector based methods is that the third and higher eigenvectors are difficult to compute. Thus, instead, we are motivated to use the pp-resistance induced by the pp-Laplacian for clustering. For pp-resistance, small pp biases towards clusters with high internal connectivity while large pp biases towards clusters of small "extent," that is a preference for smaller shortest-path distances between vertices in the cluster. However, the pp-resistance is expensive to compute. We overcome this by developing an approximation to the pp-resistance. We prove upper and lower bounds on this approximation and observe that it is exact when the graph is a tree. We also provide theoretical justification for the use of pp-resistance for clustering. Finally, we provide experiments comparing our approximated pp-resistance clustering to other pp-Laplacian based methods.

Keywords

Cite

@article{arxiv.2306.08617,
  title  = {Multi-class Graph Clustering via Approximated Effective $p$-Resistance},
  author = {Shota Saito and Mark Herbster},
  journal= {arXiv preprint arXiv:2306.08617},
  year   = {2023}
}

Comments

Accepted to ICML2023

R2 v1 2026-06-28T11:05:13.428Z