deep residual networks (resnets)

These networks, designed for image classification tasks, have layers that feed into layers much further forward in the network. This 'shortcut' or 'skip connection' helps to address the vanishing gradient problem in deep neural networks.