RECTOR: Masked Region-Channel-Temporal Modeling for Affective and Cognitive Representation Learning
Researchers have introduced RECTOR, a novel self-supervised framework designed for learning representations from EEG/sEEG data to aid in the diagnosis of affective and cognitive disorders. This framework employs a hierarchical, block-sparse self-attention mechanism that adapts region structures dynamically. RECTOR's self-supervision is driven by masked topology and representation learning, optimizing predictive modeling, topological structure, and cross-view consistency. The model demonstrates state-of-the-art performance in EEG emotion recognition and sEEG task-engagement classification, showing significant robustness to missing data and generalization across different montages. AI