An Experiment on the Use of Genetic Algorithms for Topology Selection in Deep Learning

Table 4

Correlation between model hyperparameters and test accuracy.

Hyperparameters	Correlation coefficient ()
	MNIST	CIFAR-10
# convolutional layers	−0.08	−0.25
# fully connected layers	−0.23	−0.21
# convolutional filters	0.12	0.13
Kernel size	0.13	0.11
# fully connected neurons	0.24	0.20
Dropout rate	0.02	0.00