A new first-order algorithmic framework for optimization problems with orthogonality constraints

Bin Gao, Xin Liu, Xiaojun Chen, Ya Xiang Yuan

Research output: Journal article publicationJournal articleAcademic researchpeer-review

13 Citations (Scopus)

Abstract

In this paper, we consider a class of optimization problems with orthogonality constraints, the feasible region of which is called the Stiefel manifold. Our new framework combines a function value reduction step with a correction step. Different from the existing approaches, the function value reduction step of our algorithmic framework searches along the standard Euclidean descent directions instead of the vectors in the tangent space of the Stiefel manifold, and the correction step further reduces the function value and guarantees a symmetric dual variable at the same time. We construct two types of algorithms based on this new framework. The First type is based on gradient reduction including the gradient reection (GR) and the gradient projection (GP) algorithms. The other one adopts a columnwise block coordinate descent (CBCD) scheme with a novel idea for solving the corresponding CBCD subproblem inexactly. We prove that both GR/GP with a fixed step size and CBCD belong to our algorithmic framework, and any clustering point of the iterates generated by the proposed framework is a First-order stationary point. Preliminary experiments illustrate that our new framework is of great potential.

Original languageEnglish
Pages (from-to)302-332
Number of pages31
JournalSIAM Journal on Optimization
Volume28
Issue number1
DOIs
Publication statusPublished - 1 Feb 2018

Keywords

  • Block coordinate descent
  • Gradient projection
  • Householder transformation
  • Orthogonality constraint
  • Stiefel manifold

ASJC Scopus subject areas

  • Software
  • Theoretical Computer Science

Cite this