Multi-class motor imagery EEG classification using collaborative representation-based semi-supervised extreme learning machine.

摘要: Both labeled and unlabeled data have been widely used in electroencephalographic (EEG)-based brain-computer interface (BCI). However, EEG samples are generally scarce expensive to collect, while considered be abundant real applications. Although the semi-supervised learning (SSL) allows us utilize both improve classification performance as against supervised algorithms, it has reported that occasionally undermine of SSL some cases. To overcome this challenge, we propose a collaborative representation-based extreme machine (CR-SSELM) algorithm evaluate risk by new safety-control mechanism. Specifically, ELM model is firstly predict then representation (CR) approach employed reconstruct according obtained prediction results, from which degree sample defined. A risk-based regularization term constructed accordingly embedded into objective function SS-ELM. Experiments conducted on benchmark datasets demonstrate proposed method outperforms SS-ELM algorithm. Moreover, CR-SSELM even offers best yields worse compared with its counterpart (ELM).