Transfer learning is a machine learning technique in which a model that was developed for an initial task serves now as the starting point for a model on a second duty. It is quite useful in Deep Learning since compute and time resources are limited, so you a pre-trained model can be used as an input for a computer vision or natural language processing task.

Let's demonstrate how Transfer Learning works in ML.NET by exploring the following scenario:

- Firstly, an Azure Custom Vision image classification model that uses the Open Images Dataset is trained, published and exported to TensorFlow.

- Then, transfer learning is applied to this model using the ML.NET Image Classification API in order to create a new, custom deep learning model to identify specific image categories. All the knowledge gained when solving the initial classification problem is useful for shortcutting another training process and solve a second classification.