Clustering Degree-Corrected Stochastic Block Model with Outliers
This addresses robust community detection in networks with outliers, which is important for applications like social network analysis, but appears incremental as it builds on existing degree-corrected stochastic block model frameworks.
The paper tackles the problem of clustering networks with adversarial outliers in degree-corrected stochastic block models, developing a convex-optimization-based algorithm that achieves exact recovery under mild conditions. Experimental results show it performs well on heterogeneous networks like those with Pareto degree distributions and achieves significantly lower error rates than existing methods.
For the degree corrected stochastic block model in the presence of arbitrary or even adversarial outliers, we develop a convex-optimization-based clustering algorithm that includes a penalization term depending on the positive deviation of a node from the expected number of edges to other inliers. We prove that under mild conditions, this method achieves exact recovery of the underlying clusters. Our synthetic experiments show that our algorithm performs well on heterogeneous networks, and in particular those with Pareto degree distributions, for which outliers have a broad range of possible degrees that may enhance their adversarial power. We also demonstrate that our method allows for recovery with significantly lower error rates compared to existing algorithms.