UDL Notebook Atlas

Chapter 3 (Shallow Neural Networks): Adjust weights $\omega_i$, biases $\beta_i$, and output slopes $\phi_i$ to see how 1D piecewise linear function folding emerges.

Chapter 12 (Transformers): Scaled dot-product attention softmax $\text{Softmax}(QK^T/\sqrt{d_k})$ heatmap across 5 tokens.