EEG-PRIME: Multi-Level Conditioning for Cross-Dataset EEG Decoding

EEG-PRIME is a two-stage foundation model using masked pretraining and prototype-aligned instruction tuning to address domain shifts in EEG data. It employs Layer-wise Query Modulation and frozen text embeddings as prototypes to enable subject-invariant decoding across heterogeneous datasets.