Title :
On adaptively trained neural networks
Author :
Pedreira, C.E. ; Roehl, N.M.
Author_Institution :
Catholic Univ., Rio de Janeiro, Brazil
Abstract :
In this paper a new procedure to adaptively adjust weights in a layered neural network is proposed. Nonlinear programming techniques are used in order to properly calculate the new weight set. This methodology can be used for time varying models with no necessity of retraining One of the main features of our approach concerns the designer flexibility to control a trade off problem between fitting new incoming data and causing minimum damage to the information related to the original data set. We analyze the solution existence.
Keywords :
learning (artificial intelligence); multilayer perceptrons; nonlinear programming; adaptive weight adjustment; adaptively trained neural networks; layered neural network; multilayer neural network; nonlinear programming; time-varying models; trade-off problem; Joining processes; Neural networks; Neurons; Nonlinear systems; Taylor series; Time varying systems; Training data;
Conference_Titel :
Neural Networks, 1993. IJCNN '93-Nagoya. Proceedings of 1993 International Joint Conference on
Print_ISBN :
0-7803-1421-2
DOI :
10.1109/IJCNN.1993.713978
