Working with the MNIST DATABASE of handwritten digits.
Designing and training an algorithm with the libraries Keras with Tensorflow as backend for it to learn to distinguish between handwritten numbers using completely connected layers, convolutional layers and taking into account the overlearning error dropping random neurons.
Currently obtaining a success rate of 99,9017% using the test set (~85 errors on 10,000 images) with the following layer configuration.
make installmake
git checkout vX.0, 1<=x<=3make
Implement in the project the appropriate functions for saving and loading the model once it is compiled basing on the following guides.