In this article, we will discuss the strategies to use transfer learning. Here, we’ll see the example of an image classification problem using CNN, but this concept applies to most problem types.
Before moving on to how we can leverage the transfer learning paradigm, we must know that almost all the networks (either pre-trained or your own) have a common structure in them.
This is illustrated below:
Let’s understand the architecture shown above.
That means that even pre-trained models, which we may use, have this same structure. So, we can leverage those pre-trained models using the transfer learning paradigm in ways: