Introduction

The previous chapter explained the different building blocks of traditional CNNs, as well as some techniques for improving their performance and reducing training time. The architecture explained there, although typical, is not set in stone, and a proliferation of CNN architectures have emerged to solve different data problems, most commonly in the field of computer vision.

These architectures vary in configuration as well as learning tasks. A very popular one nowadays is the Visual Geometry Group (VGG) architecture created by Karen Simonyan and Andrew Zisserman of Oxford's Robotic Institute. It was developed for object recognition and achieved state-of-the-art performance thanks to the massive number of parameters that the network relies on. One of the main reasons for its popularity among data scientists is the availability of the parameters (weights and biases) of the trained model, which allows researchers to use it without training, as well as the outstanding...