Abstract
This thesis delves into the application of machine learning techniques, specifically Transformers, for denoising EEG signals. The accurate capture of brain activity through EEG is critically compromised by various sources of noise, and machine learning offers a powerful paradigm for learning complex noise patterns and recovering clean neural signals without extensive manual filtering. The first section introduces the problem, outlines the research questions, and provides an overview of fundamental concepts in machine learning and signal processing. The second section explores the historical context and current state of signal analysis, with a focus on EEG signal analysis and denoising methods, identifying gaps that advanced machine learning models can address. The third section delves into the machine learning techniques employed in this study, detailing the Transformer architecture and its adaptation for temporal EEG data. The fourth section presents the experimental results, highlighting the significant achievements attained by the proposed machine learning framework in reducing noise while preserving neural features of interest. Finally, the last section concludes the thesis, discussing the implications of the findings for clinical and research applications of machine learning in brain-computer interfaces and suggesting avenues for future research in machine learning-driven EEG enhancement.