Leveraging transfer learning
This idea of reusing knowledge provided by others is not only important in computer science. The development of human technology over the millennia is the result of our ability to transfer knowledge from one generation to another, and from one domain to another. Many researchers believe that applying this guidance to machine learning could be one of the keys to develop more proficient systems that will be able to solve new tasks without having to relearn everything from scratch.
Therefore, this section will present what transfer learning means for artificial neural networks, and how it can be applied to our models.
Overview
We will first introduce what transfer learning is and how it is performed in deep learning, depending on the use cases.
Definition
In the first part of this chapter, we presented several well-known CNNs, developed for the ImageNet classification challenge. We mentioned that these models are commonly repurposed for a broader range of applications...
