This is a recreation of the Handwritten Digit recognition demo by Martin Görner using tensorflow and deep neural networks. MNIST dataset was used for this demo.

The ipython notebook explains the steps and progression of building the network to finally obtain >99% accuracy. It briefly covers fully connected networks and convolutional networks and also talks about techniques to optimize and improve performance.

The code is written in python3 and tensorflow is used for creating the network while matplotlib is used to plot the results.