AI Anomaly Detection Quiz

Answers and explanations appear as you take the quiz.

1. Why does anomaly detection often work without labeled examples of anomalies?
   • Anomalies are rare and unpredictable, so models learn 'normal' instead
   • Labels are illegal to use
   • Anomalies never happen
   • It only uses images
2. A 'contextual anomaly' is a data point that is:
   • Always abnormal everywhere
   • Normal in one context but abnormal in another
   • A typo in a database
   • A duplicate record
3. What does the cited WACV autoencoder study use as an anomaly score?
   • A legal fraud label assigned to every record
   • A guarantee that no future fault will occur
   • Reconstruction error from an autoencoder trained to model the data
   • The number of random trees in an Isolation Forest
4. How does the original Isolation Forest paper isolate candidate anomalies?
   • It requires an expert to label every anomaly in advance
   • It measures reconstruction error from a neural decoder
   • It uses a fixed threshold that proves an item is defective
   • Random partitioning tends to isolate anomalous instances in fewer splits
5. What is the main danger of setting an anomaly threshold too sensitively?
   • Analysts get flooded with false alarms
   • The model runs faster
   • Data gets encrypted
   • Anomalies become impossible
6. What can an unusual machine-vibration score reasonably trigger?
   • An automatic shutdown because failure is guaranteed
   • A maintenance investigation or additional measurement, without predicting failure with certainty
   • A confirmed diagnosis of the failed component
   • A claim that the machine will fail within a fixed number of hours
7. Why is an anomaly score not automatically a fraud or defect label?
   • It is always a verified label
   • Anomaly scores cannot be compared
   • It indicates unusualness under a model or baseline, not the cause or consequence
   • The threshold proves intent
8. What should be documented when setting an anomaly threshold?
   • Only the model’s name
   • A claim that false positives are impossible
   • The number of GPUs alone
   • The reference data, score definition, threshold rationale, and review process