Mixed-Norm Based Broad Learning System for EEG Classification

How to design a powerful classifier with strong generalization capability is still an active topic in the brain computer interface (BCI) researches. In this paper, we propose a new classifier, which has the same structure of the recently proposed broad learning system (BLS), but the l2 norm based optimization model in BLS is replaced by a mixed-nrom based one. To optimize the proposed model efficiently, the augmented Lagrange multiplier (ALM) method is utilized. The most attractive feature of the proposed classifier is that it has the potential to maintain good performance in various noise environments, by flexibly setting the value of mixed parameter. Thus, compared with the standard BLS, as well as many existing classifiers, the proposed one is expected to be a more reasonable choice for classifying electroencephalography (EEG) signals, which are usually polluted by various artifacts. The experiments on two publicly available data sets are presented to confirm the desirable performance of the new method.