Problem 1
0.2.
Faculty use
Low-Rank Approximation and Eckart–Young
Reward correct mathematics, explicit dimensions/hypotheses, interpretation, and verification. Equivalent valid reasoning should receive credit.
Problem 2
It stores a small number of singular vectors and values instead of every matrix entry.
Problem 3
Definitions should identify the vector/matrix/space/transformation, dimensions or domain/codomain, and the hypotheses behind retaining dominant singular components with optimal error. Use the formulas A_k=Σ_{i=1}^k σᵢuᵢvᵢᵀ; ‖A−A_k‖₂=σ_{k+1}.
Problem 4
A strong error analysis identifies the first step that violates a definition, dimension, theorem hypothesis, or matrix-operation rule; then repairs that step and re-verifies the final result.
Problem 5
Application responses vary. Require defined variables, a compatible matrix/vector model connected to data science, image compression, inverse problems, recommendation systems, and noise control, a complete computation, interpretation, and limitations.
Problem 6
Accept two independent checks appropriate to the task: reconstruction, residual, RREF comparison, determinant/rank test, orthogonality, substitution, eigenpair check Av=λv, factorization product, or a second method.
Scoring guidance
Six problems × 10 points: 5 mathematical accuracy, 2 structure/justification, 2 verification/interpretation, 1 notation and communication.