How do you handle label noise in training data?
hardAnswer
- (1) Confident Learning (cleanlab): identifies mislabeled examples via out-of-fold predictions.
- (2) Symmetric losses (MAE, generalized cross-entropy) more robust than CE.
- (3) Bootstrap / co-teaching: two networks vote.
- (4) Label smoothing (softens hard labels).
- (5) Active re-labeling: send uncertain / high-loss examples for human review.
- (6) Weighting by annotator reliability if multi-annotator.
- Empirically, models tolerate ~10-20% label noise; more requires cleaning.
Check yourself — multiple choice
- Random
- Cleanlab (confident learning) + symmetric losses (MAE, GCE) + co-teaching + label smoothing + active re-labeling + annotator weighting; tolerate ~10-20% before cleaning
- Ignore
- Not real
Label noise: cleanlab + symmetric loss + co-teach + smoothing + active.
#data-quality
Practise MLOps & Data Quality
215 interview questions in this topic.
Related questions
- What is data drift and how do you detect it?
- How is concept drift different from data drift?
- What does 'data imbalance' mean and why is it a problem?
- What are common sources of data leakage in an ML pipeline?
- How do you validate data quality in an ML pipeline?
- What is schema drift and how do you detect it?