Can you compare and contrast different CNN architectures like AlexNet, VGG, and ResNet? How do their approaches to image classification differ?