MI Classification

Motor imagery classification uses EEG to distinguish imagined actions, such as left- versus right-hand movement, so people can control devices or rehabilitation systems without overt movement. Approaches range from common spatial patterns and Riemannian covariance methods to CNNs, recurrent networks, Transformers, and state-space models that learn temporal and spatial features. Because EEG varies substantially across people and sessions, current work explores domain adaptation, subject-specific calibration, online test-time adaptation, and compact ensembles, often evaluating on public motor-imagery benchmarks. A central tension is that complex pretrained models do not consistently transfer better than task-specific decoders, while reliable personalization, fair held-out-subject evaluation, and low-latency deployment remain challenging.