Interview Practice Questions

What else do you know about Batch Normalization and Layer Normalization?

What is ROC-AUC?

What would you use - MSE or MAE - when there are outliers?

Can you tell about precision, recall, accuracy?

What will happen to ROC-AUC if positive examples are duplicated 8 times and negative examples 11 times?

Should we trust ROC-AUC 0.99 with 1% positive class rate?

Tell about bias-variance decomposition

What methods can influence bias and variance?

Tell about ensemble methods

How does Stacking work?

What's the difference between bagging and Random Forest?

By how many times can variance theoretically be reduced with bagging?

How does learning work in gradient boosting?

What happens at each iteration of boosting?

Why is regularization needed?

What regularization methods exist for different model types?

Is normalization in neural networks a regularization?

What will happen to quality if we remove the 12th tree from Random Forest?

What if we remove the 12th tree from gradient boosting?

What happens if training continues after reaching global minimum?