Embedding Differentiable Sparsity into Deep Neural Network

Yongjin Lee

In this paper, we propose embedding sparsity into the structure of deep neural networks, where model parameters can be exactly zero during training with the stochastic gradient des…