Overview of Deep Learning Architectures for Classifying Brain Signals

One of the challenges in modeling cognitive events from electroencephalogram (EEG) data is finding representations that are invariant to interand intra-subject differences, as well as the inherent noise associated with EEG data collection. Herein, we explore the capabilities of the recent deep neural architectures for modeling cognitive events from EEG data. In this paper, we present recent achievements applying deep learning for EEG signal classification. We investigate the use of feed forward, convolutional, recurrent neural nets, as well as deep belief networks, echo-state networks, reservoir computing, and denoising auto encoder models. We present the application of these architectures for classifying user intent generated through different motor imagery; BCI to control wheelchair and robotic arm; mental load classification; discriminating emotional state; feature dimensionality reduction for EEG data. Many of the models prove to be more accurate and more efficient than current state-of-the-art models.