Question from the Deep learning - Fundamentals test

How to handle missing data in a given training data set ?