Download the benchmark dataset MNIST from http://yann.lecun.com/ exdb/mnist/. Implement multiclass logistic regression and try it on MNIST.
Run the regularized multiclass logistic regression on MNIST, using the basic minibatch SGD, and compare its results to those of the basic minibatch SGD with non-regularized loss, in Question #1.
Investigate the basic minibatch SGD with, at least, another regularization method discussed in class here - early stopping, dropout