A disciplined approach to neural network hyper-parameters: Part 1 -- learning rate, batch size, momentum, and weight decay Paper • 1803.09820 • Published Mar 26, 2018 • 1