MML-16-04: Weight Initialization: Glorot/Xavier, He/Kaiming & Variance Preservation Derivations – Practice Quiz

Master the probability theory and variance propagation derivations behind deep neural network weight initialization: symmetry breaking failure of constant/zero weights, forward activation variance $\operatorname{Var}(z) = n_{\text{in}}\operatorname{Var}(w)\operatorname{Var}(x)$, backward gradient variance scaling, Glorot/Xavier normal and uniform bounds for symmetric activations, why Xavier collapses by $2^{-L}$ on ReLU, and He/Kaiming initialization derivations for ReLU and Leaky ReLU. Aligns with Prerequisites: QS 16.3, Part 08; Cognitive Level: Tier 2–4.