Convolutional Neural Networks

In the 2012 ImageNet Large Scale Visual Recognition Challenge, a convolutional network called AlexNet sorted photographs into 1,000 classes with a top-5 error of 15.3%. The second-best entry scored 26.2%. AlexNet had 60 million parameters in five convolutional and three fully connected layers, and the idea it scaled up was more than twenty years old: instead of giving every output its own weight for every pixel, slide one small set of weights across the whole image and reuse it at every position.