A Nonlinear GMRES Optimization Algorithm for Canonical Tensor Decomposition

TL;DR

This paper introduces a nonlinear GMRES-based algorithm for canonical tensor decomposition that accelerates convergence and outperforms traditional methods like ALS and nonlinear conjugate gradient in challenging scenarios.

Contribution

The paper proposes a novel nonlinear GMRES optimization algorithm that enhances tensor decomposition by effectively recombining iterates, improving convergence speed and accuracy.

Findings

N-GMRES significantly outperforms ALS and nonlinear conjugate gradient methods.

The algorithm effectively handles dense and sparse tensor problems.

It achieves higher accuracy in stationary points for difficult problems.

Abstract

A new algorithm is presented for computing a canonical rank-R tensor approximation that has minimal distance to a given tensor in the Frobenius norm, where the canonical rank-R tensor consists of the sum of R rank-one components. Each iteration of the method consists of three steps. In the first step, a tentative new iterate is generated by a stand-alone one-step process, for which we use alternating least squares (ALS). In the second step, an accelerated iterate is generated by a nonlinear generalized minimal residual (GMRES) approach, recombining previous iterates in an optimal way, and essentially using the stand-alone one-step process as a preconditioner. In particular, the nonlinear extension of GMRES is used that was proposed by Washio and Oosterlee in [ETNA Vol. 15 (2003), pp. 165-185] for nonlinear partial differential equation problems. In the third step, a line search is performed for globalization. The resulting nonlinear GMRES (N-GMRES) optimization algorithm is applied to dense and sparse tensor decomposition test problems. The numerical tests show that ALS accelerated by N-GMRES may significantly outperform both stand-alone ALS and a standard nonlinear conjugate gradient optimization method, especially when highly accurate stationary points are desired for difficult problems. The proposed N-GMRES optimization algorithm is based on general concepts and may be applied to other nonlinear optimization problems.