About this document

Global Convergence in Low-Rank Matrix Recovery by johan is a document available to read on EtoBox.

This document summarizes a research paper about low-rank matrix recovery from noisy linear measurements. The paper shows that, under certain incoherence conditions on the measurement operator, the non-convex factorized formulation has no spurious local minima. It also shows that all saddle points have a direction with negative curvature. These results guarantee that stochastic gradient descent from random initialization will converge to the global optimum in polynomial time. The paper extends these results

Author

johan

Language

EN