Self-Supervised Learning using Latent Representations of EEG Signals for Automatic Sleep Staging

• Grant number: 23/06715-3
• Support Opportunities: Scholarships in Brazil - Scientific Initiation
• Start date: August 01, 2023
• End date: July 31, 2024
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Principal Investigator: Raphael Yokoingawa de Camargo
• Grantee: Gustavo Henrique Santos Rodrigues
• Host Institution: Centro de Matemática, Computação e Cognição (CMCC). Universidade Federal do ABC (UFABC). Ministério da Educação (Brasil). Santo André , SP, Brazil

Abstract

Automated classification of sleep stages is a complex task, arising from the high-dimensionality and low signal-to-noise ratio of electroencephalogram (EEG) signals, in addition to the high variability among individuals. Although deep neural networks, including convolutional architectures and Transformers, have shown promising results for automated classification, training these models still faces the challenge of requiring a large number of labeled examples.In this work, we will use Data2Vec 2.0, a Transformer-based architecture that utilizes self-supervised learning to encode speech, text, and video signals into low-dimensional latent representations. We will adapt the model to the context of EEG signals and train the weights of the encoder and the Transformer using sleep data. For the classification of sleep stages, we will use the output of the Transformer, which has the ability to capture contextual information, which is essential for the classification of sleep stage sequences.