Table 3.

Comparison of floating-point operations(FLOPs) and weights between the convolutional network LeNet-5 and the pruned network LeNet-300-100 with 8% density on the MNIST dataset.

Network	Weights	FLOPs	FLOPs%	Error
LeNet-5 unpruned	431 K	4586 K	100%	0.8%
LeNet-300-100 unpruned	266 K	532 K	11.6%	1.82%
LeNet-300-100 block pruned (Ours)	23 K	46 K	1%	2.15%