Leaky echo state network for brainstates classification

Abstract

[en] The usage of neural networks for classification tasks has gained significant attention in recent years due to its potential in various domains, including medicine, finances or even in social media. In this particular project, based on the previous study realised by Xènia in [11], we will take advantage of these computational models in order to investigate the utility of temporal dynamics in electroencephalogram (EEG) signal classification. Also we aim to evaluate the influence of different classifier methods when classifying those EEG signals. The research employs Leaky Echo State Networks (ESNs), a type of recurrent neural network, as the main tool for extracting temporal dynamics from EEG signals. As classifiers, two distinct methods will be used to evaluate their impact on the classification task: Ridge Regression and Logistic Regression classifier. The script starts with a theoretical introduction to neural networks, with a particular focus on Leaky Echo State Networks. Subsequently, a concise overview of the two classification methods employed to construct our network architecture is presented. The final chapter is dedicated to define the aforementioned architecture and revealing the outcomes derived from the application of said network to real EEG data.