A neural network that recognizes handwritten digits.
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
NeuralNet01
unity/Prog4
.gitignore
README.md

README.md

Neural-net

A neural network that recognizes handwritten digits that I've been working on. I started by gathering information, watching some lecturers and getting an understanding of how neural networks work. Then I followed a few tutorials (http://www.cs.bham.ac.uk/~jxb/INC/nn.html and the tutorial by http://www.millermattson.com/dave/)and started implementing some basic training data to read in black and white images. I started looking for some good data sets to train my network and stumbled upon the MNIST handwritten digits dataset, so I set up a new trainingdata class, and with the help of the mnist parser by Henry Tan I managed to get the network to train itself to recognize digits fairly accurately with as little neurons as possible so that my laptop can still handle it in a decent amount of time. This also includes a small unity app with a stripped down version of the net that reads in weights and visually demonstrates the net's functionality.