NeurIPS 2020

Robustness of Community Detection to Random Geometric Perturbations

Meta Review

The authors consider the problem of community detection in a two class SBM whose edges are perturbed by a latent random geometric graph. The theoretical results show that the second eigenvector of adjacency matrix is highly correlated to the latent community membership vector under a broad parameter regime. The reviewers mostly feel that this is interesting work and I will recommend accepting this paper. However the authors should find a way to move part of the simulation study to the main paper, and address other typographic issues pointed out by the reviewers. While reviewer 4 had the most negative view of this paper, he/she does not object to acceptance. In line with his/her remark, I will strongly urge the authors to cite “Robustness of Spectral Methods for Community Detection” since it is extremely relevant to this work.