Gaussian mixture models improve fMRI-based image reconstruction

摘要

New computational models have made it possible to reconstruct perceived images from BOLD responses in visual cortex. We expand a linear Gaussian framework for percept decoding with Gaussian mixture models to better represent the prior distribution of images. In our setup, different mixture components correspond to different letter categories. Our framework not only leads to more accurate reconstructions, but also automatically infers semantic categories from low-level visual areas of the human brain.