A0625
Title: Bayesian inference on brain-computer interfaces via GLASS
Authors: Jian Kang - University of Michigan (United States) [presenting]
Abstract: Brain-computer interfaces (BCIs), specifically the P300 BCI, enable direct brain-computer communication. Classifying target vs. non-target stimuli from electroencephalogram (EEG) signals, with their low signal-to-noise ratio and complex correlations, is challenging, particularly for users with severe physical disabilities. The Gaussian latent channel model is proposed with sparse time-varying effects (GLASS) within a Bayesian framework, designed to improve classification in imbalanced data. GLASS mitigates spatial correlations through latent channel decomposition and employs a soft-thresholder Gaussian process for sparse, smooth temporal effects. Demonstrated improvements in ALS participants, GLASS highlights critical EEG channels in parietal and occipital regions, corroborating literature. An efficient gradient-based variational inference algorithm and a user-friendly Python module are also introduced for computational ease.
