Fast Computation of MoorePenrose Inverse Matrices
Abstract
Many neural learning algorithms require to solve large least square systems in order to obtain synaptic weights. MoorePenrose inverse matrices allow for solving such systems, even with rank deficiency, and they provide minimumnorm vectors of synaptic weights, which contribute to the regularization of the inputoutput mapping. It is thus of interest to develop fast and accurate algorithms for computing MoorePenrose inverse matrices. In this paper, an algorithm based on a full rank Cholesky factorization is proposed. The resulting pseudoinverse matrices are similar to those provided by other algorithms. However the computation time is substantially shorter, particularly for large systems.
 Publication:

arXiv eprints
 Pub Date:
 April 2008
 arXiv:
 arXiv:0804.4809
 Bibcode:
 2008arXiv0804.4809C
 Keywords:

 Computer Science  Neural and Evolutionary Computing
 EPrint:
 Number of pages: 5. Typo page 26 line 3: one must read W=G^+F (instead of W=G^+W, which does not make sense!)