A. AB is m×p.B. BA is always defined.C. Matrix multiplication is generally not commutative.D. Computing AB by the standard algorithm uses O(mnp) scalar operations.
A. The learning rate controls step size.B. A very large learning rate can cause divergence.C. Every non-convex objective has one unique global minimum.D. Feature scaling can improve convergence.
Correct Answer: A|B|D
Explanation:
Non-convex objectives may have many local features.