subject-adaptation

Proposes Stacked LoRA, a framework that decouples subject-invariant and subject-specific knowledge for adapting EEG foundation models to motor imagery decoding, achieving improved accuracy across multiple benchmarks.