On Calibration Algorithms for Real-Time Brain-Computer Interfaces

A Brain-Computer Interface (BCI) is a system that, in real-time, translates the user's brain activity into commands that can be used to control applications, such as moving a cursor on the screen. The translation is made possible by machine learning methods and other algorithms. The thesis focuses on EEG-based BCIs which are the most common type of BCIs due to EEG measurements being non-invasive, having good temporal resolution, and being suitable for many applications. As of today, one of the biggest challenges for BCIs is the so-called calibration, which is necessary for the BCI to translate the user's brain activity correctly. The need for calibration comes from the variability of the brain signals over time and between users.


This thesis presents an extensive review of the state-of-the-art algorithms for BCIs, focusing on the calibration problem. Amongst the presented algorithms are methods for processing the EEG data, machine learning algorithms, and a brief introduction to transfer learning and Riemannian geometry. A more in-depth exploration of the so-called multi-armed bandits and Markov decision processes as possible methods to streamline the calibration procedure is presented, as well as a real-time framework for gathering and testing algorithms. Such a framework is crucial for testing new approaches for efficient calibration.