Multi-Scale Recursive Feature Interaction for Auditory Attention Detection Using EEG Signals

Decoding auditory attention from electroencephalography (EEG) signals in cocktail party scenarios is crucial in the advancement of brain-computer interface applications, particularly in the development of neuro-guided hearing aids. Despite the potential benefits, effectively implementing auditory attention detection (AAD) with low latency and limited resources remains a challenging task. In this study, we propose a lightweight multi-scale recursive feature interactive network, named mRFInet, which incorporates long-range spatial feature extraction and higher-order feature interactions through recursive design to effectively extract feature representations from EEG for AAD. Through extensive experiments on two publicly available datasets, mRFInet demonstrates competitive performance, showcasing its potential for practical implementation in neuro-steering hearing aids.