Handwritten Number Recognition: A Machine Learning Demo

A browser-based handwriting recognizer using deep learning and TensorFlow.js.

The handwriting recognizer uses a basic convolutional neural network (CNN) model trained on the well-known MNIST dataset to recognize single digit numeric input. This model has an approximately 0.89% error rate on the test set, meaning it should get most handwriting input correct. Note that since the training data is focused specifically on handwriting, it may have a higher error rate for those of you drawing with a mouse.

For more information, check out my article on how I made this.