In this chapter, we learned how, when, and why to use transfer learning. This is considered to be a very powerful tool, because it allows us to generalize well with less data using features learned from other domains. We looked at some examples, and it should now be clear how to implement transfer learning in your own tasks.

In the next chapter, we will see how to organize our data and how to scale the CNN architectures in order to build accurate and practical machine learning systems.