We defne a Cny wodel as fCNN$($X$)=$Softmax$($FC$($Conv$1($MP$($Relu$($Conv$2($X$)))))),$ X is $28*28*3$; Conv$1$ is $106*6*3$ filters,stride $2.$ padding $1$: Relui indictes the first Relulayer; MP is a $2$ x $2$ max pooling layer, stride $2$: the second convolutional layer Conv$2$ includes $505$ x $5$ x $10$ filters, stride$=$l$,$ padding $0$; FC indicates the fully connected layer, where there are $10$ output neurons:Softmax denotes the Softmax activation function.The ground truth label of X is denoted as t$,$ and theloss function used for training this CINN model is denoted as Ly t$).$