Dimensionality Reduction and Frequency Bin Optimization To Improve a Riemannian-based Classification Pipeline
Abstract
Subject variability and BCI illiteracy pose a challenge to the BCI domain. To tackle this, the field has taken a turn towards Riemannian geometry, which has been shown to reach state-of-the-art performance. This promising mathematical concept is applicable in this field as the covariance matrices lay on a Riemannian manifold. This idea is exploited by combining Riemannian geometry with functional connectivity in a 'Functional Connectivity Ensemble' (FUCONE) in [1] with the aim to improve the performance of motor imagery (MI) BCI. FUCONE showed promising results, but this multi-step classification process also holds room for improvement. In this work, we focus on two aspects of the pipeline: the choice of frequency band and dimensionality reduction of the feature space.
Origin : Files produced by the author(s)
