Title :
Sobolev Gradients and Neural Networks
Author :
Bastian, Michael R. ; Gunther, Jacob H. ; Moon, Todd K.
Author_Institution :
Dept. of Electr. & Comput. Eng., Utah State Univ., Logan, UT
fDate :
March 31 2008-April 4 2008
Abstract :
By using a formulation similar to a Sobolev gradient for the natural gradient a new algorithm has been developed that converges faster, uses fewer additional parameters and has smaller storage requirements all while not overtraining to a training set. Simulation results show the improvements for an applicable problem.
Keywords :
Newton method; error analysis; gradient methods; learning (artificial intelligence); matrix algebra; multilayer perceptrons; Newton method; Sobolev gradient formulation; block-diagonal matrix; error function; multilayer perceptron; neural network training; Backpropagation algorithms; Computational modeling; Feedforward neural networks; Jacobian matrices; Moon; Multilayer perceptrons; Neural networks; Newton method; Partial differential equations; Signal processing algorithms; Algorithms; Feedforward Neural Networks; Newton Methods;
Conference_Titel :
Acoustics, Speech and Signal Processing, 2008. ICASSP 2008. IEEE International Conference on
Conference_Location :
Las Vegas, NV
Print_ISBN :
978-1-4244-1483-3
Electronic_ISBN :
1520-6149
DOI :
10.1109/ICASSP.2008.4518052
