Machine Learning in Elixir

Understanding Transfer Learning

Transfer learning is a technique for repurposing a pre-trained model for use on a related task. In the previous section, you saw transfer learning in the context of a deep learning problem—specifically a computer vision problem. It is possible to make use of transfer learning with other machine learning techniques and in other domains, but transfer learning is incredibly common in computer vision problems.

The power of transfer learning should make some intuitive sense. Rather than start from scratch for every problem, you inject a model with some past knowledge of the problem to speed up training. In Chapter 6, ​Go Deep with Axon​, you learned that the power of deep learning is all about learning useful representations ...