Using Deep Learning Techniques to Analyze and Classify EEG Recordings

Deep learning is a recently emerged field within machine learning which is gaining more and more attention as it significantly outperforms other approaches in an increas-ing number of applications – from speech detection to recognizing objects in images.1 The term “deep learning ” originally referred to artificial neural networks (ANNs) with more than one hidden layer but can be generalized to architectures composed of mul-tiple non-linear transformations. One major advantage of deep learning compared to traditional approaches is that they can work directly on raw data and do not require any tuning or hand-crafting of features. Instead, they are able to learn their own feature representations (see, e.g., [1]). Up to now, applications of deep learning techniques within neuroscience have been very limited. They have been successfully applied to detect anomalies related to epilepsy in electroencephalography (EEG) recordings [6] and to classify sleep stages from EEG as well as recordings of eye movements and skeletal muscle activity [3]. Very recently, the potential of deep learning techniques for neuroimaging has also been demonstrate