Take the CNN example for MNIST:

$$ Change the optimizer to Adam with L$2$ regularization $($Lambda$=0\mathrm{.}1).$

$$ Use $3$ convolutional layers with the third layer having $32$ channels and $($kernel size of $3,$ stride is $2,$ no zero$-$padding$).$

$$ Use max$-$pooling after the third convolutional layer with kernel$-$size of $2$ and stride of $2.$

$$ Plot the confusion matrix