Created a 2-layer neural network that can classify digits in the MNIST dataset with 98.1% accuracy with the pretrained model (data/pretrained.model).
The model was trained with 800 hidden layers with a 5% learning rate and 20 epochs.
Potential future additions would be SIMD/GPU acceleration as the model does take some time to train, ~30 min for the model. Smaller models can be trained with ~97% in less than a minute however.
The dataset can be downloaded on Kaggle.