[edit]
Rates of Convergence of Spectral Methods for Graphon Estimation
Proceedings of the 35th International Conference on Machine Learning, PMLR 80:5433-5442, 2018.
Abstract
This paper studies the problem of estimating the graphon function – a generative mechanism for a class of random graphs that are useful approximations to real networks. Specifically, a graph of n vertices is generated such that each pair of two vertices i and j are connected independently with probability ρn×f(xi,xj), where xi is the unknown d-dimensional label of vertex i, f is an unknown symmetric function, and ρn, assumed to be Ω(logn/n), is a scaling parameter characterizing the graph sparsity. The task is to estimate graphon f given the graph. Recent studies have identified the minimax optimal estimation error rate for d=1. However, there exists a wide gap between the known error rates of polynomial-time estimators and the minimax optimal error rate. We improve on the previously known error rates of polynomial-time estimators, by analyzing a spectral method, namely universal singular value thresholding (USVT) algorithm. When f belongs to either Hölder or Sobolev space with smoothness index α, we show the error rates of USVT are at most (nρ)−2α/(2α+d). These error rates approach the minimax optimal error rate log(nρ)/(nρ) proved in prior work for d=1, as α increases, i.e., f becomes smoother. Furthermore, when f is analytic with infinitely many times differentiability, we show the error rate of USVT is at most logd(nρ)/(nρ). When f is a step function which corresponds to the stochastic block model with k blocks for some k, the error rate of USVT is at most k/(nρ), which is larger than the minimax optimal error rate by at most a multiplicative factor k/logk. This coincides with the computational gap observed in community detection. A key ingredient of our analysis is to derive the eigenvalue decaying rate of the edge probability matrix using piecewise polynomial approximations of the graphon function f.