Why do residual connections help?

Question: Why do residual connections help very deep neural networks train better? A) They provide shortcut paths that improve gradient flow through the network B) They remove all nonlinear activations C) They force every layer to learn the same representation D) They make the model non-differentiable Correct: A Explanation: Residual connections allow gradients to flow through shortcut paths. Instead of forcing each block to learn a complete transformation, the block can learn a residual correction, which makes optimization easier in deep networks. Topic: advanced ML / deep learning / ResNet