ERPs for the rare and frequent events were calculated on each dataset by averaging the artifact-corrected EEG signals with respect to their corresponding triggers (100 ms pre-stimulus and 600 ms post-stimulus time). The group-level ERPs for rare and frequent events over the scalp are obtained by computing a grand average (GA) from the individual ERPs [Goldstein et al., 2002]. Next, we used ICA on the GA, to extract the most relevant features of the event-related response, thereby increasing the contribution of time-locked brain activity and simultaneously reducing the contribution of ongoing brain activity and of noise [Makeig et al., 1997]. Accordingly, the group-level ERPs for the rare and frequent events are concatenated and subjected to temporal ICA, assuming for them different event-related temporal responses but the same source locations [Makeig et al., 1999]. The resulting ERP-ICs were classified, to exclude artifacts and noise, on the basis of the signal-to-noise ratio (SNR) and the explained variance. The SNR, defined as the ratio between the maximum amplitude in the post-stimulus interval and the root mean square (RMS) in the pre-stimulus interval, was
