Multi-layer Perceptron from scratch with numpy
The MLP in model.py contains backpropagation, regularization, and activation functions like ReLU and softmax.
It does not use any frameworks but numpy.
The performance on MNIST is around 95% accurate.
Enjoy the codes.