Clustering simulated Event-Related Potentials based on similarity of centroids

Event-Related Potentials (ERPs) are electrical signals produced by the brain in response to external stimuli. Due to the enormous noise affecting them, traditionally, the analysis of ERPs has relied of averaging the signals recorded in many repetitions of an experiment. However, while averaging helps to improve the signal to noise ratio of an ERP, it does so at the expense of blurring it, thereby making it possible to observe only large scale effects. It has recently been shown that binning ERPs based on user response-times and then averaging can mitigate this problem. However, this technique relies the availability of a physical manifestation of a user's mental processes (e.g., a key press), and could not be used in experiments where this is not desirable or possible. In our programme of research, we attempt to go beyond this limitation by using forms of unsupervised clustering of ERPs prior to averaging. In this paper, we test clustering based on similarity of centroids with simulated ERPs.