“…Two specific configurations achieved the same top accuracy value of 98.72% and thus outperforming the best CNNs. This is one of the most relevant results of our experiments, suggesting that, by exploiting a sequence of [5,5], [5,5]] 64 [120, 80] [0.5, 0.5] 736,170 96.5142 0.1280 [ [5,5], [5,5]] 32 [160, 80] [0.5, 0.5] 454,386 96.5142 0.1241 [ [5,5], [3,3]] 64 [160, 0] [0.5, 0.5] 1,063,074 96.4418 0.1327 [ [5,5], [3,3] [3,3], [5,5]] 128 50 [0, 0] 2,981,310 98.7225 0.0413 5], [3,3]] 64 250 [0.5, 0.5] 6,690,094 98.7119 0.0541 [ [3,3], [5,5]] 32 250 [0.5, 0.5] 3,477,454 98.7119 0.0503 [ [3,3], [5,5]] 32 50 [0, 0] 676,254 98.6908 0.0438 [ [5,5], [5,5]] 128 250 [0, 0] 11,032,558 98.6908 0.0428 [ [5,5], [3,3]] 32 100 [0.5, 0.5] 1,330,682 98.6908 0.0497 [ [3,3], [5,5]] 128 100 [0, 0] 5,571,610 98.6908 0.0433 [ [3,3], [5,5]...…”