202
v1v2v3 (latest)

A proximal Newton framework for composite minimization: Graph learning without Cholesky decompositions and matrix inversions

International Conference on Machine Learning (ICML), 2013
Anastasios Kyrillidis
Abstract

We propose an algorithmic framework for convex minimization problems of a composite function with two terms: a self-concordant function and a possibly nonsmooth regularization term. Our method is a new proximal Newton algorithm that features a local quadratic convergence rate. As a specific instance of our framework, we consider the sparse inverse covariance matrix estimation in graph learning problems. Via a careful dual formulation and a novel analytic step-size selection procedure, our approach for graph learning avoids Cholesky decompositions and matrix inversions in its iteration making it attractive for parallel and distributed implementations.

View on arXiv
Comments on this paper