Network Clustering for Latent State and Changepoint Detection

TL;DR

This paper introduces a convex clustering method for networks that automatically discovers hierarchical groupings without predefining the number of clusters, effectively capturing evolving structures in complex data.

Contribution

It proposes a novel convex fusion penalty approach for network clustering that produces a hierarchical structure without needing to specify cluster count beforehand.

Findings

Effective on synthetic data examples

Produces a smooth, tree-like cluster hierarchy

Eliminates need for pre-specifying number of clusters

Abstract

Network models provide a powerful and flexible framework for analyzing a wide range of structured data sources. In many situations of interest, however, multiple networks can be constructed to capture different aspects of an underlying phenomenon or to capture changing behavior over time. In such settings, it is often useful to cluster together related networks in attempt to identify patterns of common structure. In this paper, we propose a convex approach for the task of network clustering. Our approach uses a convex fusion penalty to induce a smoothly-varying tree-like cluster structure, eliminating the need to select the number of clusters a priori. We provide an efficient algorithm for convex network clustering and demonstrate its effectiveness on synthetic examples.