Questions: Active Learning

5 questions to test your understanding

Score: 0 / 5
Question 1 Multiple Choice

A model trained with uncertainty sampling consistently queries borderline examples between two classes but never improves on a third class that exists in a distant cluster. What is the root cause and the fix?

AThe model needs a larger architecture to handle three classes simultaneously
BUncertainty sampling is myopic — it ignores distant, unvisited regions; diversity sampling addresses this by querying examples far from any already-labeled point
CThe labeling budget is too small; with more labels the model will eventually reach the third cluster through uncertainty sampling
DThe model should switch to unsupervised learning to discover the third cluster
Question 2 Multiple Choice

What is the essential difference between active learning and standard supervised learning?

AActive learning uses unlabeled data at test time to improve predictions
BActive learning replaces gradient-based optimization with reinforcement signals from a reward function
CThe model selects which examples to request labels for, rather than passively receiving a fixed pre-labeled dataset
DActive learning requires more labeled data to achieve the same accuracy as standard supervised learning
Question 3 True / False

In uncertainty sampling, the model should preferentially label examples where it is most confident about the correct class label.

TTrue
FFalse
Question 4 True / False

Active learning requires a small initial labeled seed set to begin — it cannot start from a completely unlabeled pool.

TTrue
FFalse
Question 5 Short Answer

Explain why pure uncertainty sampling can fail in practice and what additional criterion addresses this limitation.

Think about your answer, then reveal below.