Arrange the following steps in the process of implementing identity mapping in deep neural networks to optimize training: A) Add identity mapping layers, B) Train the network with residual connections, C) Initialize weights for deeper layers, D) Evaluate performance on training data.

To set up an identity‑mapping deep network you first add the identity‑mapping layers (A) so that each layer can pass its input unchanged if needed. Next you initialize the weights for the deeper layers (C), usually with small random values that keep the identity mapping stable. Then you train the network using residual connections (B), letting the optimizer learn how much to adjust each layer while keeping the identity path intact. Finally you evaluate performance on the training data (D) to confirm that the identity mapping helps the model learn without vanishing gradients. For example, a 20‑layer ResNet starts with identity shortcuts, initializes each block with He weights, trains with back‑propagation, and checks training loss after each epoch.