## Part II: Questions

This part contains over 200 questions that have more or less deterministic answers. This type of question is to test your understanding of machine learning concepts. In an hour-long interview, you can cover 10 - 15 of those questions. I rank the questions by three levels of difficulty:

• [E]asy - fundamental, everyone should know.
• [M]edium - people who understand a concept beyond just the definition should know.
• [H]ard - people with experience should know.

Some of the knowledge questions are considered bad interview questions, especially those about definitions that can be easily looked up. For example, asking someone to explain PCA is good for evaluating their memorization of PCA, not their understanding of PCA. However, some bad interview questions can still make good questions when practicing for interviews, so I include some definition questions to remind readers that certain concepts are important.

Techniques go in and out of fashion, but fundamental challenges stay the same. Instead of asking candidates to write out complex equations for certain techniques, this book focuses on the challenges that gave rise to those techniques in the first place. Most of the questions in this section are about why something matters and how it works.

🌳 Tip: Strategy for definition questions 🌳
When asked to give the definition of or explain a technique, always start with the motivation for that technique. For example, if asked to explain LSTM for recurrent neural networks, you should first bring up the problems that arise in normal RNNs and how LSTMs address those problems.