Fully Kernected Neural Networks

Table 6

Training time of CNN, KNN, and our FKNN on CIFAR100.

Methods	Network-depth	Time per epoch (s)	Convergence time (h)
CNN	ResNet-18	62	2.38
KNN	ResNet-18	64	2.31
FKNN (ours)	ResNet-18	61	1.81
CNN	DenseNet-40	72	3.76
KNN	DenseNet-40	78	3.03
FKNN (ours)	DenseNet-40	71	2.56

We mark the best results in bold values.