Title :
Gradient based adaptive regularization
Author :
Eigenmann, Robert ; Nossek, Josef A.
Author_Institution :
Inst. for Network Theory & Circuit Design, Munchen Univ., Germany
Abstract :
A technique to optimize regularization parameters for a given supervised training problem is presented. A training database is applied to minimize a regularized cost function, and a validation database is used to estimate and optimize generalization properties by means of a modification of regularization. The performance is validated for a vowel classification task and compared to other approaches
Keywords :
feedforward neural nets; learning (artificial intelligence); multilayer perceptrons; pattern classification; speech recognition; generalization properties; gradient based adaptive regularization; regularization parameters; regularized cost function; supervised training problem; training database; validation database; vowel classification task; Circuit synthesis; Cost function; Data structures; Databases; Error correction; Training data;
Conference_Titel :
Neural Networks for Signal Processing IX, 1999. Proceedings of the 1999 IEEE Signal Processing Society Workshop.
Conference_Location :
Madison, WI
Print_ISBN :
0-7803-5673-X
DOI :
10.1109/NNSP.1999.788126
