In this chapter, we covered a lot of ground. We began by introducing a new kind of neural network, the convnet. We explored the theory and architecture of a convnet in the most ubiquitous form and also by discussing some state-of the-art network design principles that have been developing as recently as mid-2015 in organizations such as Google and Baidu. We built an understanding of the topology and also of how the network operates.

Following this, we began to work with the convnet itself, applying it to the CIFAR-10 dataset. We used modular convnet code to create a functional architecture that reached a reasonable level of accuracy in classifying 10-class image data. While we're definitely still at some distance from human levels of accuracy, we're gradually closing the gap! Chapter 8, Ensemble Methods will pick up from what you learned here, taking these techniques and their application to the next level.