Attention modulates neural representation to render reconstructions according to subjective appearance

Abstract

Horikawa and Kamitani investigate the effects of attention on visual image reconstructions from fMRI activity measured while human participants focus on one of two superimposed images. By reconstructing images from deep neural network features decoded from the brain, they show that top-down attention counters stimulus-induced responses, modulating neural representations to render reconstructions in accordance with subjective appearance.