Multi-task deep neural network reveals distinct and hierarchical  pathways for face perception in visual cortex

We developed a multi-task deep neural network (DNN) model that can simultaneously classify facial expressions and identities. The model’s architecture and weights were optimized, and then used as an efficient tool to investigate the neural responses to facial expression and identity perception in an fMRI experiment. Our results revealed distinct visual pathways for facial expression and identity processing in the dorsal and ventral pathways in IT cortex, respectively. We also found hierarchical processing for facial expression and identity within the visual pathways.