From Brainwaves to Actions: Evaluating Uncertainty in CNNs and Riemannian Geometry Models for BCI

Suurmeijer, Joris (2024) From Brainwaves to Actions: Evaluating Uncertainty in CNNs and Riemannian Geometry Models for BCI. Bachelor's Thesis, Artificial Intelligence.

Preview	Text bAI2024SuurmeijerJ.pdf Download (3MB) | Preview
	Text toestemming.pdf Restricted to Registered users only Download (402kB)

Abstract

Brain-computer interfaces (BCIs) capture brain signals and transform them into functionally useful output. In the field of Motor Imagery, a Machine Learning model can learn from the captured data to predict the imagined movement of a user. This study aims to guide researchers in their choice for such a Machine Learning model, by comparing the capabilities of Deep Learning and non-Deep Learning models in terms of their classification performance and Uncertainty Quantification (UQ). Convolutional Neural Networks (CNNs) are used as the Deep Learning model and a Minimum Distance to Riemannian Mean (MDRM) model is used as the non-Deep Learning model. The UQ methods used are Deep Ensembles and DUQ for CNNs, and two distance-based uncertainty methods for MDRM. Results show that Deep Ensembles have better accuracy than the other methods, whereas MDRM models offer better UQ despite lower classification performance. DUQ performs worse than Deep Ensembles and MDRM in both areas.

Item Type:	Thesis (Bachelor's Thesis)
Supervisor name:	Jong, I.P. de
Degree programme:	Artificial Intelligence
Thesis type:	Bachelor's Thesis
Language:	English
Date Deposited:	30 Jul 2024 09:38
Last Modified:	30 Jul 2024 09:38
URI:	https://fse.studenttheses.ub.rug.nl/id/eprint/33606

Actions (login required)

View Item