Table 6.

ResNet50 on ImageNet100.

Pruning iteration	Params (M)	FLOPS (M)	Pruned ratio (%)		Val. acc. (%)
			Params	FLOPS
Baseline	23.71	4116.69	—	—	79.98
1	13.33	2388.97	43.78	41.97	80.35
2	8.64	1569.34	63.56	61.88	80.39
3	6.32	1156.42	73.34	71.91	79.82
4	5.10	895.66	78.49	78.24	79.30
5	4.30	739.39	81.86	82.04	78.08

The bold values mean that the best optimized network is obtained at the 4th iteration.