Differentiable Network Pruning via Polarization of Probabilistic Channelwise Soft Masks

Table 1

Setting of hyperparameters and for ResNet56 on CIFAR-10 and ResNet50 on ImageNet.

Datasets	Network		t	FLOPs (%)
CIFAR-10	ResNet56	0.0004	0.2	54.60
			0.8	66.84
			1.2	75.62
ImageNet	ResNet50	0.00005	−0.2	53.07
			0.2	65.91
			0.5	72.43