Whenever you're ready, walk me through how you'd approach this.
Walk me through the mathematical intuition of backpropagation. How do vanishing gradients occur, and how do modern architectures mitigate them?
Explain the chain rule through multiple layers, relate gradients to parameter updates, and distinguish vanishing from exploding gradients. Discuss the effects of activation functions, initialization, normalization, gating, residual connections, and attention. Use equations or a small implementation to make the explanation concrete.