Date of Defense
30-4-2025 11:00 AM
Location
E1-2025
Document Type
Dissertation Defense
Degree Name
Doctor of Philosophy in Informatics and Computing
College
CIT
Department
Computer Science and Software Engineering
First Advisor
Prof. Abderrahmane Lakas
Keywords
Brain-computer interface (BCI), brain-machine interface (BMI), electroencephalography/electroencephalogram (EEG), adaptive decoding algorithms, transfer learning.
Abstract
Brain Computer Interface (BCI), Also known as brain-machine interface (BMI) is a mean of controlling machines without the need to activate peripheral nerves or muscles. It has received the attention of research for decades. Motor imagery-based BCI is a paradigm that is characterized by its user friendliness where users can generate control commands at their freewill, without waiting for a que from the BCI module. Motor imagery brain–computer interface (MI–BCI) has considerable potential in increasing the quality of the lives for people with mobility impairment and the healthy ones as well. Though, its diffusion in application still has many pitfalls due to different limitations. The decoding of the signal, typically EEG, requires frequent calibration to maintain an acceptable accuracy threshold. Those limitations stem from many factors related to the nature of the signal, availability of training examples, and user related aspects. BCI-illiteracy is one of the open challenges that have been in literature for decades. This dissertation addresses the motor imager BCI-illiteracy by investigating the effect of Riemannian adaptive decoding and transfer learning on the performance of the classification accuracy. It introduces a variety of methods including supervised, unsupervised, and rebiased decoding of EEG signals. Also, it uses statistical methods to investigate the relationship between the motor execution and motor imagery in an endeavor to reframe the BCI-illiteracy. Finally, it transfers the domain knowledge between motor execution and motor imagery reducing the need for model calibration on motor imagery examples. The transfer of knowledge used resembles a straight forward and simple transfer approach where the weights of the class prototype presented a noticeable improvement with zero calibration. The methods presented in this dissertation contribute to the literature by reframing the BCI-illiteracy from an unprecedented approach by comparing the accuracy of motor imagery and motor execution as well. This dissertation paves the road towards having more reliable user-centered BCIs and cater for better understating and overcoming of BCI-illiteracy.
OVERCOMING MOTOR IMAGERY BCI ILLITERACY: ADAPTIVE DECODING AND KNOWLEDGE TRANSFER IN EEG-BASED BRAIN-COMPUTER INTERFACES
E1-2025
Brain Computer Interface (BCI), Also known as brain-machine interface (BMI) is a mean of controlling machines without the need to activate peripheral nerves or muscles. It has received the attention of research for decades. Motor imagery-based BCI is a paradigm that is characterized by its user friendliness where users can generate control commands at their freewill, without waiting for a que from the BCI module. Motor imagery brain–computer interface (MI–BCI) has considerable potential in increasing the quality of the lives for people with mobility impairment and the healthy ones as well. Though, its diffusion in application still has many pitfalls due to different limitations. The decoding of the signal, typically EEG, requires frequent calibration to maintain an acceptable accuracy threshold. Those limitations stem from many factors related to the nature of the signal, availability of training examples, and user related aspects. BCI-illiteracy is one of the open challenges that have been in literature for decades. This dissertation addresses the motor imager BCI-illiteracy by investigating the effect of Riemannian adaptive decoding and transfer learning on the performance of the classification accuracy. It introduces a variety of methods including supervised, unsupervised, and rebiased decoding of EEG signals. Also, it uses statistical methods to investigate the relationship between the motor execution and motor imagery in an endeavor to reframe the BCI-illiteracy. Finally, it transfers the domain knowledge between motor execution and motor imagery reducing the need for model calibration on motor imagery examples. The transfer of knowledge used resembles a straight forward and simple transfer approach where the weights of the class prototype presented a noticeable improvement with zero calibration. The methods presented in this dissertation contribute to the literature by reframing the BCI-illiteracy from an unprecedented approach by comparing the accuracy of motor imagery and motor execution as well. This dissertation paves the road towards having more reliable user-centered BCIs and cater for better understating and overcoming of BCI-illiteracy.
