Detecting Hidden Communities by Power Iterations with Connections to Vanilla Spectral Algorithms

TL;DR

This paper introduces simple power-iteration and spectral algorithms for community detection in stochastic block models, achieving near-optimal performance and addressing challenges with many communities and small clusters.

Contribution

It presents new vanilla algorithms based on power iterations and spectral methods, solving open problems for large community numbers and small cluster recovery.

Findings

Algorithms perform optimally up to logarithmic factors in dense graphs

Robust recovery of large clusters despite small cluster barriers

Connection established between powered adjacency matrices and eigenvector-based spectral methods

Abstract

Community detection in the stochastic block model is one of the central problems of graph clustering. Since its introduction, many subsequent papers have made great strides in solving and understanding this model. In this setup, spectral algorithms have been one of the most widely used frameworks. However, despite the long history of study, there are still unsolved challenges. One of the main open problems is the design and analysis of "simple"(vanilla) spectral algorithms, especially when the number of communities is large. In this paper, we provide two algorithms. The first one is based on the power-iteration method. It is a simple algorithm which only compares the rows of the powered adjacency matrix. Our algorithm performs optimally (up to logarithmic factors) compared to the best known bounds in the dense graph regime by Van Vu (Combinatorics Probability and Computing, 2018). Furthermore, our algorithm is also robust to the "small cluster barrier", recovering large clusters in the presence of an arbitrary number of small clusters. Then based on a connection between the powered adjacency matrix and eigenvectors, we provide a vanilla spectral algorithm for large number of communities in the balanced case. This answers an open question by Van Vu (Combinatorics Probability and Computing, 2018) in the balanced case. Our methods also partially solve technical barriers discussed by Abbe, Fan, Wang and Zhong (Annals of Statistics, 2020). In the technical side, we introduce a random partition method to analyze each entry of a powered random matrix. This method can be viewed as an eigenvector version of Wigner's trace method. Recall that Wigner's trace method links the trace of powered matrix to eigenvalues. Our method links the whole powered matrix to the span of eigenvectors. We expect our method to have more applications in random matrix theory.