A Training Methodology for Neural Networks Noise-Filtering when no Training Sets are available for Supervised Learning

Noise filtering is considered one of the main applications of neural networks due to its importance in a wide range of scientific and technological areas. The standard methodology needs to obtain first an accurate measure of the desired signal, which is a must in supervised learning. Nevertheless, in some areas these data sets are rarely available, nor can be determined noise function although its distribution is usually known. In this paper, we propose a training methodology combining data simulation, modular neural networks and an interval-splitting strategy for noise-filtering where training data sets are not necessary. Method is explained step by step, and finally results are presented and conclusions done