Regresar

Active learning using a constructive neural network algorithm

Abstract:

Constructive neural network algorithms suffer severely from overfitting noisy datasets as, in general, they learn the set of available examples until zero error is achieved.We introduce in this work a method for detect and filter noisy examples using a recently proposed constructive neural network algorithm. The new method works by exploiting the fact that noisy examples are in general harder to be learnt than normal examples, needing a larger number of synaptic weight modifications. Different tests are carried out, both with controlled and real benchmark datasets, showing the effectiveness of the approach. Using different classification algorithms, it is observed an improved generalization ability in most cases when the filtered dataset is used instead of the original one. © 2009 Springer-Verlag Berlin Heidelberg.