1) Reinforcement learning optimizes:
A) Expected cumulative reward
2) A model can be accurate and unfair:
A) True
3) Fairness constraints:
A) May reduce accuracy
4) Ethical AI primarily concerns:
A) Fairness, accountability, transparency
5) SGD provides implicit regularization:
A) Under certain conditions
6) Transformers rely primarily on:
A) Self-attention
7) Double descent phenomenon suggests:
A) Test error may decrease again after interpolation
8) A perfectly fair model must:
A) Satisfy a fairness definition
9) Is bias always bad in ML?
A) It depends on context
10) Data leakage occurs when:
A) Test data influences training
11) GAN training is unstable because:
A) Two-player minimax optimization
12) Increasing model capacity always increases variance:
A) It depends on data size