6aed000af86a084f9cb0264161e29dd3-AuthorFeedback.pdf

Herewe19 estimate a differentw[r]for each ROI to account for the regional hemodynamic response variability.φFt Second,ourmodel36 can be easily applied to data not having a trial-based structure (e.g., resting-state data). We evaluated the model on37 task based EEG-fMRI data rather than resting state data because we originally wanted the model tobetested using38 ahypothesis-drivenapproach andourtaskEEG-fMRI datahadbeenanalyzed thoroughly byourgroup. Inaddition,39 for the resting-state case, there is no task activation so we potentially lose the ability to highlight the importance of40 thespatialspecificity fromfMRI. Since variational inference algorithms are generally sensitive to the initialization, we chose a more informative48 initialization procedure whichusessomefMRI-constrained techniques proposedin[12-14] toachieveabetterinitial49 solution for the EEG source localization.