The Overlap Gap Property in Principal Submatrix Recovery

From MaRDI portal
(Redirected from Publication:6324231)



Abstract: We study support recovery for a kimesk principal submatrix with elevated mean lambda/N, hidden in an NimesN symmetric mean zero Gaussian matrix. Here lambda>0 is a universal constant, and we assume k=Nho for some constant hoin(0,1). We establish that {there exists a constant C>0 such that} the MLE recovers a constant proportion of the hidden submatrix if lambdageqCsqrtfrac1hologfrac1ho, {while such recovery is information theoretically impossible if lambda=o(sqrtfrac1hologfrac1ho)}. The MLE is computationally intractable in general, and in fact, for ho>0 sufficiently small, this problem is conjectured to exhibit a emph{statistical-computational gap}. To provide rigorous evidence for this, we study the likelihood landscape for this problem, and establish that for some varepsilon>0 and sqrtfrac1hologfrac1holllambdallfrac1ho1/2+varepsilon, the problem exhibits a variant of the emph{Overlap-Gap-Property (OGP)}. As a direct consequence, we establish that a family of local MCMC based algorithms do not achieve optimal recovery. Finally, we establish that for lambda>1/ho, a simple spectral method recovers a constant proportion of the hidden submatrix.














This page was built for publication: The Overlap Gap Property in Principal Submatrix Recovery

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6324231)