MML-16-03: Universal Approximation Theorems & Expressive Capacity – Practice Quiz

Explore the theoretical foundations of neural network expressive capacity: Cybenko (1989) Universal Approximation Theorem for sigmoidal networks, Hornik (1991) and Leshno-Lin-Pinkus-Schocken (1993) non-polynomial characterization, constructive bump/hat function approximations, Barron spectral bounds, width-bounded deep ReLU universality, and Telgarsky depth separation proofs demonstrating exponential width requirements for shallow networks. Aligns with Prerequisites: QS 16.2; Cognitive Level: Tier 3–4.