EasyDeepLearn

How do you detect anomalous predictions in production?

medium

Answer

  • (1) Statistical: z-score of prediction / feature vs recent history.
  • (2) Isolation Forest / LOF on request features.
  • (3) Autoencoder reconstruction error on inputs.
  • (4) Prediction confidence: sudden increase in low-confidence predictions.
  • (5) Feature range violations (values outside training distribution).
  • (6) Model-specific: attention weights unusual for LLMs. Aggregate to hourly / minute counts + alert on spike.
  • Route anomalies to sample store for review.
Check yourself — multiple choice
  • Random
  • Statistical z-score + Isolation Forest / LOF + autoencoder reconstruction + prediction confidence + feature range + attention anomalies for LLMs; aggregate + alert spike + sample for review
  • Not possible
  • Just count

Anomaly monitoring: z-score / IF / AE / confidence / range / attention; aggregate + review.

#monitoring

Practise MLOps & Data Quality

215 interview questions in this topic.

Related questions