Previous | Next --- Slide 15 of 41
Back to Lecture Thumbnails

The middle column is a typical visualization of the weights of the first conv layer of a convnet. This specific image is taken from alexnet. There has been some hypothesis that convolution neural network in vision tasks, albeit not biologically inspired, bares some similarity with our visual cortex. Specifically, these stripes are similar to gabor filter, which is a common model for the receptive fields of primary visual cortex (V1). Therefore, there has been some claim that our visual cortex has been through the similar optimization process as the network to develop its functionalities.