Average Sensitivity of Spectral Clustering

TL;DR

This paper analyzes the stability of spectral clustering under edge perturbations by deriving theoretical bounds based on eigenvalues and confirming them through experiments, highlighting its robustness when clear cluster structures exist.

Contribution

It provides the first theoretical bounds on the average sensitivity of spectral clustering to edge removals, linking stability to eigenvalues of the Laplacian.

Findings

Spectral clustering's stability is proportional to λ2/λ3^2.

Theoretical bounds are confirmed empirically on synthetic and real data.

Clustering stability improves with clearer cluster structures.

Abstract

Spectral clustering is one of the most popular clustering methods for finding clusters in a graph, which has found many applications in data mining. However, the input graph in those applications may have many missing edges due to error in measurement, withholding for a privacy reason, or arbitrariness in data conversion. To make reliable and efficient decisions based on spectral clustering, we assess the stability of spectral clustering against edge perturbations in the input graph using the notion of average sensitivity, which is the expected size of the symmetric difference of the output clusters before and after we randomly remove edges. We first prove that the average sensitivity of spectral clustering is proportional to $\lambda_2/\lambda_3^2$, where $\lambda_i$ is the $i$-th smallest eigenvalue of the (normalized) Laplacian. We also prove an analogous bound for $k$-way spectral clustering, which partitions the graph into $k$ clusters. Then, we empirically confirm our theoretical bounds by conducting experiments on synthetic and real networks. Our results suggest that spectral clustering is stable against edge perturbations when there is a cluster structure in the input graph.