A piece of code that creates a NN, trains it and tests it in a set of examples by Zalando Research as found on kaggle.com
The accuracy with 60,000 images on 10,000 test images is 53%, rising to 76% when trained multiple (10) times on the same set.
gcc nn_bp.c -o nn_bp -lm -O3
- Make the number of hidden layers flexible
- Include multiple activation functions to choose from