Optimization of the EEG2Code model used for a VEP-based Brain-Computer Interface

Master’s Thesis, assigned to Katharina Deibel

Abstract

A Brain-Computer Interface (BCI) is an interface between the human brain and a computer. BCIs enable physically paralyzed patients, for example, to control a computer via pure brain activity. The EEG2Code BCI developed in Tübingen is currently the fastest non-invasive BCI system. The schematic flow of the method is shown in the figure. Each letter of the virtual keyboard flickers with a random stimulation pattern. This flickering generates so-called visually evoked potentials (VEPs), which can be recorded by electroencephalography (EEG). The EEG is then classified using a previously trained neural network, whereby the output corresponds to the stimulation pattern of the letter the user has been looking at (with a certain accuracy).

Task

The developed BCI already enables fast and asynchronous control, but tests have shown that the currently used computational neural network (CNN) can be heavily optimized to achive better classification performances and to reduce computational costs. Therefore, the aim of this work is to implement different machine learning methods for the EEG2Code model. You will be provided with data from several EEG recordings to test and validate your models.

Requirements

  • Interest in neuroscience, machine learning, and data analysis
  • good programming skills (experience in Python or MATLAB are beneficial)

Contact

Nagel, Sebastian