Deep learning techniques for brain-computer interface systems

Abstract

Brain-Computer Interfaces (BCI) has been subject of great attention for its potential applications in a myriad of contexts, for example, in assistive, rehabilitation and entertainment technologies. Significant advances, such as data collect with non-invasive methods by means of electroencephalograms (EEG), motivates the study and development of this promising interface. However, the high variability of patterns observed in users of BCI, and its application in increasing sophisticated contexts, makes its use a very challenging problem. In this sense, this research project aims at applying Deep Learning techniques for improvement of BCI systems, making them more efficient and robust. In a first stage, the focus shall be on the treatment of EEG signals through Independent Component Analysis (ICA) and imaging techniques, allowing an efficient feature extraction. In a second stage, deep learning networks and generative adversarial nets shall be used for BCI data classification. Their great potential to deal with high variability data shall be very useful for BCI systems. (AU)