Table 2.

Optimal values for the hyperparameters for the custom WRN and pretrained CNNs obtained through randomized grid search M: Momentum, ILR: Initial learning rate, and L2: L2-weight decay).

Models	Optimal values
	M	ILR	L2
Custom	0.90	1e-3	1e-5
Pretrained	0.95	1e-3	1e-6