Given the unique requirements and constraints faced by facial recognition systems, it is clear that the paradigm of training a CNN for classification using a huge dataset (known as batch learning classification) is unsuitable for the facial recognition problem. Instead, our objective is to create a neural network that can learn to recognize any face using just a single training sample. This form of neural network training is known as one-shot learning.

One-shot learning brings about a new and interesting paradigm in machine learning problems. Thus far, we have thought of machine learning problems as mostly classification problems. In Chapter 2, Predicting Diabetes, with Multilayer Perceptrons, we used an MLP to classify patients at risk of diabetes. In Chapter 4, Cats Versus Dogs – Image Classification Using CNNs, we used a CNN to classify images of cats...