Multiactivation Pooling Method in Convolutional Neural Networks for Image Recognition

Table 11

Different CIFAR-10 classification results with different combinations of k and σ based on VGG-11. The left two column results are based on the modelusing 4×4 MAP method. The right two column results are based on the model using 8×8 MAP method.

(k, σ)	result (%)	(k, σ)	result (%)
(1, 1)	8.24	(1,1)	7.84
(3, 0.5)	8.07	(4, 0.5)	6.94
(3, 0.75)	8.21	(4, 0.75)	7.14
(3, 0.33)	8.16	(4, 0.25)	7.07
(4, 0.5)	8.25	(6, 0.5)	7.15
(4, 0.25)	8.31	(6, 0.17)	7.33
(6, 0.5)	8.40	(8, 0.5)	7.57
(8, 0.5)	8.51	(8, 0.125)	7.63