Local metric learning for EEG-based personal identification

There has been an increasing attention on Electroencephalograph (EEG) based personal identification over the last decade. Most existing methods address this problem by Euclidean metric based Nearest Neighbor (NN) search. However, under various recording conditions, simple Euclidean distance cannot model the similarity relations between EEG signals precisely. To overcome this drawback, a local metric learning based on Large Margin Nearest Neighbor (L-LMNN) for EEG based personal identification is proposed in this paper. For each EEG sample, a separate local metric is learned, making the distance between intra-class EEG samples minimized and simultaneously those of inter-class EEG samples maximized. To balance the locality and computational efficiency, the local metrics are approximated by weighted linear combinations of a small set of anchor samples. Experimental results demonstrate that the proposed approach obtains competitive performance compared with state-of-the-art methods. It improves the identification accuracy overall, especially at shorter EEG durations, which is important for improving the practicability of EEG-based personal identification system.