Study of reconstruction ICA for feature extraction in images and signals

Abstract

[en] During the last years, neural networks have become a vehicular discipline in the field of machine learning. At the same time, classical machine learning methods have become easier to use due to the availability of higher computational power. The goal of this project is to reconstruct a classical machine learning algorithm used for feature and source extraction (ICA) using neural networks. This reconstruction could bypass some of the drawbacks presents when using ICA. We have studied how the reconstruction operates under different conditions and performed a comparison with the classical algorithm that we reconstructed.