L1 vs L2 Normalization for Embeddings and Count Profiles

EasyClassical ML

Reported in interviews at Netflix

Problem statement

How do L1 and L2 normalization differ, and when would you use each? A music recommender needs preprocessing for track embeddings and users’ genre play counts. Here normalization means dividing each sample vector by its norm, so explain what that does to the vector.

Compare normalization choices and explain their uses and edge cases.

Show hint

For each method, ask which quantity equals exactly 1 after you divide.

Relevant roles: ML Engineer, Data Scientist and Applied Scientist.

Related free problems