Predicting the neural response of primary visual cortex (v1) using deep learning approach

One of the most important parts of the brain is the visual cortex, which receives data from the visual system as input and, by a hierarchical processing, leads to our understanding of the scene. Despite the efforts in recent years, predicting the neural response to the natural stimuli by the best models proposed for the primary visual cortex has not performed well to date. However, the results obtained from machine learning show that deep neural networks are able to learn nonlinear functions to process visual information. In this study, a new approach to modeling V1 neural activity is presented which is based on the VGG-19 deep network. In this approach, inspired by the function of the visual cortex of the brain, a structure is introduced that by adding a convolutional network causes the model to pay attention to important parts of the input. In this structure, a mask is created using the receptive field of the brain neurons and is added to the middle layers of the deep network. The use of this mask makes the network to be more influenced by the areas of the image to which brain neurons are more sensitive. The proposed deep network training shows that the results obtained from predicting the neural response to natural stimuli are faster and more accurate than previous models.