9.5 Results 327 linear discriminator trained with an N-ms training window, the discriminator output y has dimension N × T , where T is the number of trials. We average across all training samples and compute: y j = 1 N N y ij i=1 (9.6) where j is the trial number index. We use y j to vary the amplitude of our fMRI regressor. The onset and duration of the events making up each regressor are determined by the temporal characteristics of the discriminating EEG discrim- inating components as identified by the single-trial EEG analysis. Figure 9.6 illustrates how two EEG components, identified for different time windows, can be used to construct fMRI regressors specific to each EEG discriminating component. To identify the unique contribution of each component, we must orthogonalize regressors with respect to one another.