Conditional gradient methods via stochastic path-integrated differential estimator

Conditional gradient methods via stochastic path-integrated differential estimator

We propose a class of novel variance-reduced stochastic conditional gradient methods. By adopting the recent stochastic path-integrated differential estimator technique (SPIDER) of Fang ct al. (2018) for the classical Frank-Wolfe (FW) method, we introduce SPIDER-FW for finite-sum minimization as wel...

Full description

Bibliographic Details
Main Author: Sra, Suvrit (Author)
Other Authors: Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science (Contributor)
Format: Article
Language:English
Published: International Machine Learning Society, 2021-04-27T17:11:23Z.
Subjects:
Article
Online Access:Get fulltext
Description
Summary:We propose a class of novel variance-reduced stochastic conditional gradient methods. By adopting the recent stochastic path-integrated differential estimator technique (SPIDER) of Fang ct al. (2018) for the classical Frank-Wolfe (FW) method, we introduce SPIDER-FW for finite-sum minimization as well as the more general expectation minimization problems. SPIDER-FW enjoys superior complexity guarantees in the non-convex setting, while matching the best known FW variants in the convex case. We also extend our framework à la conditional gradient sliding (CGS) of Lan & Zhou (2016), and propose SPIDER-CGS.
National Science Foundation (U.S.) (Grant 200021178865/1)
National Science Foundation (U.S.). Career (Grant 1846088)

Similar Items