NeurIPS 2019
Sun Dec 8th through Sat the 14th, 2019 at Vancouver Convention Center
Paper ID:2674
Title:Global Convergence of Least Squares EM for Demixing Two Log-Concave Densities

This paper analyzes least square EM for mixture of two log concave distributions (that are rotationally invariant) and proves global convergence. This generalizes previous results on mixture of two Gaussians. Reviewers found the paper to be reasonable, although there is also doubt on the applicability given that the result cannot be generalized to 3 or more components.