Digit recognizer

Digit recognition is a classic application of machine learning. The idea is to interpret image data as one of the digits 0-9. The image might be scanned from actual handwriting or drawn via a mouse or touchpad.

You can try it out by drawing a number in the box below.

These days, perhaps this example seems quaint. Kaggle calls it the defacto “hello world” of computer vision. Nonetheless, it’s easy to see why this is such an important example as a first step to more general optical character recognition and, more generally, text recognition.

This demo was taken from Adam Smith’s work on DenseInL2 and modified only slightly. It uses a convolutional neural network (or CNN) with six layers that’s a pretty standard example using Tensorflow by now. He built it from the ground up, though, and the result is pretty nice.