We focus on empirically understanding the behavior of deep neural networks, developing deep learning algorithms and architectures that learn better, faster, and cheaper.