A Discriminative and Robust Feature Learning Approach for EEG-Based Motor Imagery Decoding (Student Abstract)

Convolutional neural networks (CNNs) have been commonly applied in the area of Electroencephalography (EEG)-based Motor Imagery (MI) classification, significantly pushing boundary state-of-the-art. In order to simultaneously decode discriminative features and eliminate negative effects non-Gaussian noise outliers motor imagery data, this abstract, we propose a novel robust supervision signal, called Correntropy based Center Loss (CCL), for CNN training, which utilizes correntropy induced distance as objective measure. It is encouraging see that model trained by combination softmax loss CCL outperforms state-of-the-art models on two public datasets.