Deep and Transfer Learning with Riemannian Geometry for EEG Signal Analysis

• Grant number: 23/13300-4
• Support Opportunities: Scholarships in Brazil - Doctorate
• Effective date (Start): May 01, 2024
• Effective date (End): April 30, 2027
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computer Systems
• Principal Investigator: Denis Gustavo Fantinato
• Grantee: João Guilherme Prado Barbon
• Host Institution: Faculdade de Engenharia Elétrica e de Computação (FEEC). Universidade Estadual de Campinas (UNICAMP). Campinas , SP, Brazil
• Host Company: Universidade Estadual de Campinas (UNICAMP). Faculdade de Engenharia Elétrica e de Computação (FEEC)
• Associated research grant: 20/09838-0 - BI0S - Brazilian Institute of Data Science, AP.PCPE

Abstract

Electroencephalographic (EEG) data are records of brain electrical activity, with potential application in a wide range of contexts, ranging from medical diagnosis, assistive technologies and rehabilitation to entertainment devices. In that sense, with the consolidation of signal processing and machine learning techniques, the direct information transmission system between the brain and the computer, called the Brain-Computer Interface (BCI), can be highlighted. However, the high variability of patterns observed in EEG data from BCI users and the application of BCI in increasingly sophisticated contexts, make its use a very challenging problem. In this sense, this research project aims at applying Deep Learning and Signal Processing techniques for improvement of BCI systems, making them more efficient and robust. In the first stage, the focus shall be on the improvement of classifier models for EEG-based BCI systems, considering deeper architectures for the SPDNet and EE(G)-SPDNet models. In a second stage, deep Riemannian networks shall be used in transfer learning techniques, considering generative adversarial networks (GANs) and Autoencoders.