The State Of Sparsity In Deep Neural Networks