Table 5.

The McNemar exact test p-values for the tested “Network in Network” (NiN) models trained on CIFAR-100.

	Baseline	Finetuning	Deep supervision	Hints	RDL
Baseline	—	0.00←	0.00↑	0.00↑	0.00↑
Finetuning	0.00↑	—	0.00↑	0.00↑	0.00↑
Deep supervision	0.00←	0.00←	—	0.86	0.08
Hints	0.00←	0.00←	0.86	—	0.05
RDL	0.00←	0.00←	0.08	0.05	—

Arrows indicate a significant difference (p < 0.05,uncorr.) and point to the better model.