Learning Representations from EEG with Deep Recurrent-Convolutional\n Neural Networks

One of the challenges in modeling cognitive events from electroencephalogram\n(EEG) data is finding representations that are invariant to inter- and\nintra-subject differences, as well as to inherent noise associated with such\ndata. Herein, we propose a novel approach for learning such representations\nfrom multi-channel EEG time-series, and demonstrate its advantages in the\ncontext of mental load classification task. First, we transform EEG activities\ninto a sequence of topology-preserving multi-spectral images, as opposed to\nstandard EEG analysis techniques that ignore such spatial information. Next, we\ntrain a deep recurrent-convolutional network inspired by state-of-the-art video\nclassification to learn robust representations from the sequence of images. The\nproposed approach is designed to preserve the spatial, spectral, and temporal\nstructure of EEG which leads to finding features that are less sensitive to\nvariations and distortions within each dimension. Empirical evaluation on the\ncognitive load classification task demonstrated significant improvements in\nclassification accuracy over current state-of-the-art approaches in this field.\n

Hali tarjima qilinmagan