Gradual Capacity Growth for Sparse Network Discovery

Qihang Yao, Constantine Dovrolis

Sparse neural network methods typically assume that the target sparsity (or density) is fixed in advance, even though the relationship between network capacity and performance is g…