Table 1 Comparison results of combination of batch normalization (BN) and dropout rate. BN only showed the best accuracy.

Regularization Layer	Kernel size	Activation function	Optimizer	Number of Epochs	Accuracy
BN	2 × 2	ReLU^a	Adam^b	46	93.7%
BN and Dropout (0.1)^c	2 × 2	ReLU	Adam	28	74.6%
BN and Dropout (0.2)	2 × 2	ReLU	Adam	165	92.7%
BN and Dropout (0.3)	2 × 2	ReLU	Adam	48	86.6%
BN and Dropout (0.4)	2 × 2	ReLU	Adam	46	85.0%
BN and Dropout (0.5)	2 × 2	ReLU	Adam	55	73.9%
Dropout (0.1)	2 × 2	ReLU	Adam	59	86.6%
Dropout (0.2)	2 × 2	ReLU	Adam	116	88.2%
Dropout (0.3)	2 × 2	ReLU	Adam	96	87.7%
Dropout (0.4)	2 × 2	ReLU	Adam	112	87.9%
Dropout (0.5)	2 × 2	ReLU	Adam	28	72.1%

Quick links

Search