CNNs are one of the most influential classes of networks in the history of deep learning. Invented by Yann LeCun (now head of Facebook Artificial Intelligence Research), CNNs really came into their own in 2012, with the introduction of deep Convolutional Neural Networks by Alex Krizhevsky.
Plain old neural networks don't scale well to images; CNNs adapt regular old feedforward neural networks by adding one or more convolutional layers as the input layer to the network. These convolutions are specifically designed to take in two-dimensional input, such as images or even sound, as illustrated in the following diagram:
As you can see, CNNs add these layers of convolutions together with something called, appropriately pooling layers in order to insert an image. The second part of the network is nothing more than the standard feedforward network that we&apos...