Adding more layers should give a neural network more capacity, but very deep networks can become harder to optimize. Residual connections, popularized by ResNet, solve an important part of this problem and allow hundreds of layers to be trained effectively. The degradation problem A deeper model does not always achieve lower training error than a…