MML-14-09: Multiclass Softmax Regression & Information Geometry – Practice Quiz

35 worked single-select and multiselect questions on Multiclass Softmax Regression & Information Geometry. Cognitive level: Tier 2–3. Prerequisites: QS 14.8, Part 11.

Learning objectives: Normalize multiclass logits stably; Interpret pairwise odds and reference-class identifiability; Compute cross-entropy and class-weight gradients; Evaluate the softmax Jacobian and curvature; Connect categorical covariance with simplex geometry; Diagnose temperature, smoothing and exclusive-class modeling.

Each item is self-contained, with calculations, model diagnostics and explanations of every distractor. Study references: Stanford CS229 notes.