Asynchronous Parallel Algorithms for Nonconvex Optimization

TL;DR

This paper introduces an asynchronous parallel algorithmic framework for nonconvex optimization problems that effectively models modern computational architectures, ensuring convergence and near-linear speedup.

Contribution

It presents a unified, flexible framework for asynchronous parallel nonconvex optimization that handles various constraints and architectures with proven convergence.

Findings

Almost sure convergence to stationary points.

Theoretical results show near-linear speedup with multiple workers.

Framework covers multiple existing methods in a unified way.

Abstract

We propose a new asynchronous parallel block-descent algorithmic framework for the minimization of the sum of a smooth nonconvex function and a nonsmooth convex one, subject to both convex and nonconvex constraints. The proposed framework hinges on successive convex approximation techniques and a novel probabilistic model that captures key elements of modern computational architectures and asynchronous implementations in a more faithful way than current state-of-the-art models. Other key features of the framework are: i) it covers in a unified way several specific solution methods; ii) it accommodates a variety of possible parallel computing architectures; and iii) it can deal with nonconvex constraints. Almost sure convergence to stationary solutions is proved, and theoretical complexity results are provided, showing nearly ideal linear speedup when the number of workers is not too large.