Software‎ > ‎

Matlab code for nonlinearly preconditioned NCG and NGMRES optimization

The Matlab files in the zipped folder below contain the implementation of the nonlinearly preconditioned NCG algorithm for CP tensor decomposition from reference [1] below (PNCG), and provide code to compare with other optimization methods, including standard NCG and LBFGS, ALS, and the nonlinearly preconditioned NGMRES method from [2]. (Note that some of the Matlab files in the zipped folder use functions from Sandia's Tensor and Poblano toolboxes, see the file README.txt.)

Matlab code


[1] Hans De Sterck and Manda Winlaw, "Nonlinearly Preconditioned Conjugate Gradient Algorithm for Rank-R Canonical Tensor Approximation", submitted, 2014. [arXiv link]

[2] Hans De Sterck, "A Nonlinear GMRES Optimization Algorithm for Canonical Tensor Decomposition", SIAM J. Sci. Comput. 34, A1351-A1379, 2012. [pdf]