Abstract: One promising approach to mitigating the computational complexity of deep neural networks is to leverage the sparsity of input activations that results from the application of the ReLU ...