Hands-On One-shot Learning with Python

Architecture

A Siamese network consists of two identical neural networks that share similar parameters, each head taking one input data point. In the middle layer, we extract similar kinds of features, as weights and biases are the same. The last layers of these networks are fed to a contrastive loss function layer, which calculates the similarity between the two inputs.

One question you might have is why do Siamese networks' layers share parameters? If we are already putting the effort into changing the loss function, won't it help us to train the layers separately?

There are two major reasons why we are not training layers separately:

• For every layer, we have thousands of parameters being added. Therefore, similar to how we do in a convolutional ...