In the chapters ahead, we will be using some of these popular CNNs and their variants to solve image and video processing tasks. Right now, let's use the LeNet architecture proposed by Yann LeCun to recognize handwritten digits. This architecture was used by the US Postal Service to recognize handwritten ZIP codes on the letters they received (http://yann.lecun.com/exdb/publis/pdf/jackel-95.pdf).
LeNet consists of five layers with two convolutional max pool layers and three fully connected layers. The network also uses dropout feature, that is while training, some of the weights are turned off. This forces the other interconnections to compensate for them, and hence helps in overcoming overfitting: