Fig.2 shows the relative between
iterations and loss function at different rate.
The method of adaptive rate can quickly reduce the loss function value under 1.0 after 20 iterations.
Provided much more samples
and deeper layers, the ADCNN could have larger advantage than other models