Differentiable Mask Pruning for Neural Networks