Table 6 Accuracy per label and statistical features of their filters for VGG-16 trained on \(K\) labels from CIFAR-10.

Layer	\({N}_{f}\)	\({F}_{s}\)	\(F{C}_{s}\)	Accuracy	\(n\)	\({N}_{c}\)	\({C}_{s}\)
VGG-16 on CIFAR-3/10
13	512	1 × 1	512	0.988	0.07	1	1.02
10	512	2 × 2	2048	0.988	0.27	1.5	1.02
7	256	4 × 4	4096	0.989	0.67	1.2	1.06
4	128	8 × 8	8192	0.972	1.70	1.1	1.12
2	64	16 × 16	16,384	0.927	1.78	1.1	1.25
VGG-16 on CIFAR-6/10
13	512	1 × 1	512	0.968	0.40	1	1.8
10	512	2 × 2	2048	0.967	1.16	2.4	1.3
7	256	4 × 4	4096	0.957	2.12	1.3	1.4
4	128	8 × 8	8192	0.930	6.69	1.2	1.6
2	64	16 × 16	16,384	0.860	7.59	1.1	1.7
VGG-16 on CIFAR-8/10
13	512	1 × 1	512	0.961	0.63	1	2.2
10	512	2 × 2	2048	0.958	2.17	2.8	1.4
7	256	4 × 4	4096	0.954	4.07	1.2	1.6
4	128	8 × 8	8192	0.890	12.4	1.4	1.8
2	64	16 × 16	16,384	0.783	13.0	1.3	1.8
VGG-16 on CIFAR-10/10
13	512	1 × 1	512	0.94	1.5	1	2.8
10	512	2 × 2	2048	0.94	3.8	3.2	1.6
7	256	4 × 4	4096	0.93	6.4	1.3	1.6
4	128	8 × 8	8192	0.85	18.3	1.4	2.1
2	64	16 × 16	16,384	0.72	19.6	1.3	2.1

The results of VGG-16 trained on \(K=3, 6, 8,\) and \(10\) labels, namely CIFAR-K/10 (Supplementary Information).

Quick links

Search