TU Berlin

Neural Information ProcessingHyper-Ellipsoidal Conjugate Gradient Descent

Neuronale Informationsverarbeitung

Page Content

to Navigation

Hyper-Ellipsoidal Conjugate Gradient Descent

On this page you can find a MATLAB implementation of the Hyper-Ellipsoidal Conjugate Gradient Descent algorithm, that can be used for sparse opitimization of second order kernel methods like kernel-PCA, kernel-SFA (slow feature analysis), or kernel-CCA (canonical correlation analysis). The following two files you may download, use, redistribute, and/or modify under the terms of the GNU General Public License.

hecgd.m - hyper-ellipsoidal conjugate gradient descent algorithm
errsokm.m - error function for sparse second order kernel methods

How to use these files is described here.

If you use this software in publications, please cite:

Vollgraf, R. and Obermayer, K. (2006). Sparse Optimization for Second Order Kernel Methods. IJCNN 2006 Conference Proceedings. IEEE, 145 – 152.


Quick Access

Schnellnavigation zur Seite über Nummerneingabe