Summary

In this chapter, you learned two very important concepts: transfer learning and fine-tuning. Both help deep learning practitioners to leverage existing pre-trained models and adapt them to their own projects and datasets.

Transfer learning is the re-use of models that have been trained on large datasets such as ImageNet (which contains more than 14 million images). TensorFlow provides a list of such pre-trained models in its core API. You can also access other models from renowned publishers such as Google and NVIDIA through TensorFlow Hub.

Finally, you got some hands-on practice fine-tuning a pre-trained model. You learned how to freeze the early layers of a model and only train the last layers according to the specificities of the input dataset.

These two techniques were a major breakthrough for the community as they facilitated access to state-of-the-art models for anyone interested in applying deep learning models.

In the next chapter, you will look at another...