Hangman BCI: An unsupervised adaptive self-paced Brain-Computer Interface for playing games

This paper presents a novel user interface suitable for adaptive Brain Computer Interface (BCI) system. A customized self-paced BCI architecture is introduced where the system combines onset detection system along with an adaptive classifier working in parallel. An unsupervised adaptive method based on sequential expectation maximization for Gaussian mixture model is employed with new timing scheme and an additional averaging step to avoid over-fitting. Sigmoid function based post-processing approach is proposed to enhance the classifiers' output. The adaptive system is compared to a non-adaptive one and tested on five subjects who used the BCI to play the hangman game. The results show significant improvement of the True-False difference for all the classes and a reduction in the number of steps required to solve the problem.