222afbe0d68c61de60374b96f1d86715-AuthorFeedback.pdf

Some other differences7 are: 1) CHL performs approximate gradient-descent. Our network performs exact gradient descent-ascent. Neurons are organized with a stride of 2 and multiple neurons (or features) per site (NpS), got feedforward input from pixels within radius 4, and lateral input from neurons within radius dl. Increasing NpS anddl increased the performanceoftheclassifier. Similarity matching as a deep learning objective function:This is10 a very fair criticism and we should have been very clear about it11 in our paper.