Bridge: Weight Sharing, Convolution and Attention

How a network computes, why gradients vanish, and what makes depth trainable

You now own the dense layer completely. Forward it computes Wx + b, backward it returns Wᵀδ to the layer below and δxᵀ for its own weights, and a whole stack of them trains by dl.15. Almost all of modern machine learning is two constrained versions of that one layer.

🔒 This is a Pro lesson — the interactive figure, worked examples, quiz and practice open with Pro access.

▶ Bridge: Weight Sharing, Convolution and Attention
← Why Depth Works: Folding Space and Reusing Features