KH-SGD iteratively reorders datapoints during stochastic gradient descent training to provably accelerate convergence. - View it on GitHub