An efficient pruning scheme of deep neural networks for Internet of Things applications

EURASIP Journal on Advances in Signal Processing

Table 2 Results of pruning VGG on CIFAR-10

Method	Baseline (%)	Accuracy (%)	FLOPs pruned (%)	Parameters pruned (%)
PFEC[18]	93.25	↑ 0.15	34.2	64.0
NS[19]	93.66	↑ 0.14	51.0	88.5
CFP[32]	93.49	↓ 0.51	81.9	-
COP[33]	93.56	↓ 0.25	73.5	92.8
GAL[35]	93.96	↓ 0.54	45.2	82.2
PFS[38]	93.44	↑ 0.19	50.0	-
Ours	93.50	↑ 0.25	49.3	83.8
		↓ 0.23	72.6	94.1

We report both baseline and after-pruning accuracy of the state-of-the-art methods referring to their papers. The label “-” in last column indicates that such item is not reported