Constrained-learning in artificial neural networks
R. Parra-Hernandez · 2004
The capacity to generalize is the most important characteristic in neural networks. However, the generalization capacity is lost when over-fitting occurs during the neural network training process; i.e., although the error after the training process is very small, when new data is presented to the neural network the error is large. An approach aiming to improve the neural network generalization capacity is presented in this work.