DocumentCode :
295811
Title :
Second-order training for recurrent neural networks without teacher-forcing
Author :
Von Zuben, Fernando J. ; De Andrade Netto, Márcio L.
Author_Institution :
Sch. of Electr. Eng., Univ. Estadual de Campinas, Sao Paulo, Brazil
Volume :
2
fYear :
1995
fDate :
Nov/Dec 1995
Firstpage :
801
Abstract :
Neural networks with external recurrences can be successfully applied to nonlinear autoregressive moving average modeling. The process of weight adjustment is presented as a nonlinear optimization problem in the N-dimensional Euclidean space, where N is the number of adjustable weights. The least-squares criterion can be effectively minimized using a version of the conjugate gradient algorithm. Expending about the same amount of computation necessary to obtain the gradient, the required second-order information is calculated exactly. A simulation example confirms the efficacy of the training process when applied to time series prediction. Contrary to the proposed method, teacher-forced learning is shown to be ill-suited for multistep prediction
Keywords :
autoregressive moving average processes; conjugate gradient methods; learning (artificial intelligence); least squares approximations; modelling; optimisation; recurrent neural nets; conjugate gradient algorithm; external recurrences; least-squares criterion; multidimensional Euclidean space; multistep prediction; nonlinear autoregressive moving average modeling; nonlinear optimization; recurrent neural networks; second-order information; second-order training; teacher-forced learning; teacher-forcing; time series prediction; Autoregressive processes; Computational modeling; Delay effects; Dynamic range; Erbium; Multilayer perceptrons; Neural networks; Predictive models; Recurrent neural networks; Steady-state;
fLanguage :
English
Publisher :
ieee
Conference_Titel :
Neural Networks, 1995. Proceedings., IEEE International Conference on
Conference_Location :
Perth, WA
Print_ISBN :
0-7803-2768-3
Type :
conf
DOI :
10.1109/ICNN.1995.487520
Filename :
487520
Link To Document :
بازگشت