A neural network that reads handwriting, written from scratch in Scala.
scriptophile was my first attempt to work with the MNIST dataset of handwritten digits. Inspired by Michael Nielsen’s excellent textbook Neural Networks and Deep Learning, I decided to hack together a Scala implementation of the neural network algorithms described in Nielsen’s book. The result: a simple network that learns through mini-batch stochastic gradient descent & identifies digits from the MNIST dataset with roughly 94% accuracy. It’s written with Breeze for linear algebra. I wrote up a blog post about my thought processes while building it. The code is, of course, all on my GitHub.