Table 2.

Hyperparameters tuned in both the Features Extraction (FE) and Fine-Tuning (FT) experiments. Algorithm DE indicates the classification layers added to the Deep Embeddings models considered in our experiments.

Task	Algorithm	Parameter	Values
FE	SVM	Regularization	[$10^{-3},\dots ,10^3$]
		Kernel	[rbf, poly, sigmoid]
		Kernel coefficient	[$10^{-3},\dots ,10$]
		Degree of poly kernel	[$2,\dots ,5$]
	AB	Estimators	[10, 20, 50, 100]
		Learning rate	[1, .5, .1, .05, .01, .001]
	LR	Penalty	[l1, l2]
		Regularization	[$10^{-3},\dots ,10^3$]
	RF	Estimators	[10, 20, 50, 100]
		Min samples split	[2, 8, 10, 12]
		Max depth	[10, 30, 50]
		Split criterion	[entropy, gini]
	PCA	Explained variance	[.6, .65, .7, .75, .8, .85, .9. .95, .99]
FT	DE	Hidden layers	[$1,\dots ,5$]
		Hidden units	[128, 512, 1024, 2048, 6144]
		Dropout rate	[0, .1, .2, .3, .4]