Machine Learning for Neuroimaging Talk, by Sanne Schoenmakers and Tom Heskes, DCC/iCIS, Radboud University Nijmegen, Friday April 25, 10:00-11:00, at DTU Compute, Building 324, Room 240
Abstract
New computational models make it possible to reconstruct perceived images from BOLD responses in human visual cortex. We propose a Bayesian framework, consisting of an encoding model, which linearly maps images to brain activity, and a Gaussian mixture model for the prior distribution of images. We show that such an analytical framework automatically infers semantic categories from low-level visual areas and provides accurate reconstructions of individual instances of handwritten characters.