Comparison of deep neural networks to spatio-temporal cortical dynamics of human visual object recognition reveals hierarchical correspondence

• DOI: 10.1038/srep27755
• Publication Date: 2016-06-10T09:23:57Z
• Authors: Radoslaw Martin C...; Aditya Khosla; Dimitrios Pantazis; Antonio Torralba; Aude Oliva
• Keywords: Magnetoencephalography; Visual Objects; Visual processing

Abstract The complex multi-stage architecture of cortical visual pathways provides the neural basis for efficient object recognition in humans. However, stage-wise computations therein remain poorly understood. Here, we compared temporal (magnetoencephalography) and spatial (functional MRI) brain representations with an artificial deep network (DNN) tuned to statistics real-world recognition. We showed that DNN captured stages human processing both time space from early areas towards dorsal ventral streams. Further investigation crucial parameters revealed while model was important, training on categorization necessary enforce spatio-temporal hierarchical relationships brain. Together our results provide algorithmically informed view dynamics