Convolutional Neural Networks

In Chapter 2, Neural Networks, you learned about traditional neural networks, such as perceptrons, that are composed of fully connected layers (also called dense layers). Each layer is composed of neurons that perform matrix multiplication, followed by a non-linear transformation with an activation function.

CNNs are actually very similar to traditional neural networks, but instead of using fully connected layers, they use convolutional layers. Each convolution layer will have a defined number of filters (or kernels) that will apply the convolution operation with a given stride on an input image with or without padding and can be followed by an activation function.

CNNs are widely used for image classification, where the network will have to predict the right class for a given input. This is exactly the same as classification problems for traditional machine learning algorithms. If the output can only be from two different classes, it will be a binary...