Table 2.

Ablation evaluation of eXnet.

Model Parameter	Accuracy
Effect of depth reduction on the accuracy of the eXnet
eXnet after removal of initial two CBR blocks	67
eXnet after removal of last two CBR blocks	69
Effect of pooling layers on the accuracy of the eXnet
eXnet having convolutions with larger strides	58
eXnet having convolutions with smaller strides	65
eXnet having pooling layers	>71
Effect of dropout layer on the accuracy of the eXnet
eXnet without dropout layer	69
eXnet with dropout between fully-connected layers	71
Effect of kernel size on the accuracy of the eXnet
eXnet with kernel size 32	48
eXnet with kernel size 16	57
eXnet with kernel size 8	68
Effect of $ParaFeat$ blocks on the accuracy of the eXnet
eXnet after skipping first $ParaFeat$	68
eXnet after skipping second $ParaFeat$	67
eXnet without using both $ParaFeat$	62
Effect of fully connected layer on the accuracy of the eXnet
eXnet without any fully connected layer	60
eXnet without fully connected layer 1	65
eXnet without fully connected layer 2	68
Effect of changing optimizer on the accuracy of the eXnet
eXnet with Adaptive moment estimation (Adam) optimizer	70
eXnet with SGD optimizer	71.67
Effect of learning rate on the accuracy of the eXnet
eXnet with fix learning rate	69
eXnet cyclical learning rate	71.67