How do skip connections improve gradient flow in deep Keras networks?

Skip connections allow gradients to bypass layers, reducing vanishing gradient issues.

They create identity shortcuts that preserve gradient magnitude during backpropagation, enabling very deep networks like ResNet to train effectively.