This notebook builds an SNN to determine similarity scores between MNIST digits using a triplet loss function. The use of class prototypes at inference time is also explored.
This notebook is based heavily on the approach described in this Coursera course, which in turn is based on the FaceNet paper. Any uses of open-source code are linked throughout where utilised.
For an in-depth guide to understand this code, and the theory behind it, please see my article for Towards Data Science.