SPERA: Spherical Prior EEG Foundation Model with Geometry- and Frequency-Aware Latent Prediction
SPERA is a new EEG foundation model that uses a joint-embedding predictive architecture to learn representations from massive EEG data. It integrates spatial prior encoding based on Legendre polynomials, hybrid attention mechanisms, and spectral regularization to capture complex brain signals across different datasets. This results in state-of-the-art accuracy and strong performance in few-shot learning.