Digit Recognizer Neural Network

A Machine Learning application trained to identify handwritten digits from 0-9.

This digit recognition application is able to detect and identify single numerical digits (0-9). It served as a great experiment that tested the skills I've gained after completing Andrew Ng’s Machine Learning course on Coursera. The 3 layered neural network was trained on 5000 training examples (that I have modified) from the MNIST database of handwritten digits on Octave. Then, I transferred over the test pre-trained weights over to Unity3D, where I have just implemented forward propagation using these weights.

In the about section within the app, I have briefly outlined the how the recognizer works while explaining the inner workings of Feedforward Neural Networks if you're curious.

Upon thorough testing (5 minutes), I have come to the conclusion that the digit recognizer I have built is unable to recognize digits at a state of the art performance. I believe this is due to a lack of training data. Considering this project was created from an experimental perspective, I have decided to settle for now. I may decide to add more data from the MNIST data later in my life.

...It may be a little rough around the edges, but I will always love my firstborn.

If you would like to check out the source code:

https://github.com/edweenie123/Digit-Recognition-App