Convolutional deep belief networks for feature extraction of EEG signal

In recent years, deep learning approaches have been successfully used to learn hierarchical representations of image data, audio data etc. However, to our knowledge, these deep learning approaches have not been extensively studied for electroencephalographic (EEG) data. Considering the properties of EEG data, high-dimensional and multichannel, we applied convolutional deep belief networks to the feature learning of EEG data and evaluated it on the datasets from previous BCI competitions. Compared with other state-of-the-art feature extraction methods, the learned features using convolutional deep belief network have better performance.