Overfitting, Dropout and Normalization From Training to Deployment
Problem statement
How would you fix a neural network that fits its training data but gets worse on new data? The model predicts whether users will report content. The team is considering regularization and has questions about dropout and serving with batch normalization. Diagnose overfitting and review the dropout and batch normalization implementations.
Show hint
Where do the training and validation curves part ways, and what does that say about bias versus variance?
Relevant roles: ML Engineer, Deep Learning Engineer, Applied Scientist and Data Scientist.