Universality of Computational Lower Bounds for Submatrix Detection

arXiv1902.06916MaRDI QIDQ6314261FDOQ6314261

Authors: Matthew D. Brennan, Guy Bresler, Wasim Huleihel 

Publication date: 19 February 2019

Abstract: In the general submatrix detection problem, the task is to detect the presence of a small $kimesk$ submatrix with entries sampled from a distribution $mathcalP$ in an $nimesn$ matrix of samples from $mathcalQ$. This formulation includes a number of well-studied problems, such as biclustering when $mathcalP$ and $mathcalQ$ are Gaussians and the planted dense subgraph formulation of community detection when the submatrix is a principal minor and $mathcalP$ and $mathcalQ$ are Bernoulli random variables. These problems all seem to exhibit a universal phenomenon: there is a statistical-computational gap depending on $mathcalP$ and $mathcalQ$ between the minimum $k$ at which this task can be solved and the minimum $k$ at which it can be solved in polynomial time. Our main result is to tightly characterize this computational barrier as a tradeoff between $k$ and the KL divergences between $mathcalP$ and $mathcalQ$ through average-case reductions from the planted clique conjecture. These computational lower bounds hold given mild assumptions on $mathcalP$ and $mathcalQ$ arising naturally from classical binary hypothesis testing. Our results recover and generalize the planted clique lower bounds for Gaussian biclustering in Ma-Wu (2015) and Brennan et al. (2018) and for the sparse and general regimes of planted dense subgraph in Hajek et al. (2015) and Brennan et al. (2018). This yields the first universality principle for computational lower bounds obtained through average-case reductions.

This page was built for publication: Universality of Computational Lower Bounds for Submatrix Detection

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6314261)