I'm working on a book on machine learning interviews so I've been spending the last few months talking to companies about their hiring process for ML roles. This thread is a summary of what I've learned. It will be updated as the book progresses. (1/n)
The average interviewer gets very little training. You start your full-time job. You shadow a few interviews. Then you're on your own. As a result, interviews are wildly different even within the same company.
I ask interviewers to give me their most frequently asked questions, and show them questions that other interviewers ask. I notice the pattern that if an interviewer doesn't know the answer to a question, they immediately flag it down as bad.
"Explain Adam optimizer" is the most frequently asked question for machine learning roles.
Some companies still ask candidates for research scientist roles to write linked list or binary search tree. smh
Given how much of our life is spent at work and how much impact machine learning research/productization can have, it surprised me that no company asks candidates about values. All interviews are about whether you can do something, not whether or not you should do something.
As of now, @AndrewYNg's Machine learning course on Coursera has had 2.4M students. # of people enrolled in all online ML courses must be tens of millions. If 5% of them complete, there'd be 100,000's of people with ML knowledge. Why is it still so hard to hire for ML roles?
Previous employers matter a lot. Having Google, Facebook, etc. on resume almost always guarantees an interview. College names, surprisingly, don't matter as much. Some hiring managers actually told me they're wary of recent grads from elite schools because of their "entitlement".
I've been helping a lot of people (mostly friends + students for now) prepare for ML interviews and in return, they share with me their interviewing experiences. DM me if you want me to prep you.
Most candidates told me the hardest questions for them are the machine learning system design questions. They don't know what a good answer to these questions looks like. Interviewers: any tips?
"Machine learning interviews as a black box". Few companies have done any research on the correlation b/w interview techniques/signals and quality of hires. Challenges: people who interview aren't the same people who evaluate performance, no established metrics, no control group.
Woah just spent the last 3 hours responding to 50+ interview prep messages. Only 4 of them are from women. Excited to start talking to more candidates about the process, but don't think I'd be able to take on more requests for now, unless it's from women!
Proposed interview question: how to build a system to predict someone's day-to-day performance based on their interview performance.
And what exactly is "culture fit"? It seems to me that this term is sometimes used as a proxy for exclusivity: to hire only people who share your background/interest/worldview as you. As long as you're not a jerk, shouldn't you be able to fit in anywhere?
And companies complain why it's so hard to hire for machine learning roles. 
Because some of you asked, I put up a mailing list for those who want to be notified when the book is out.
I also wrote this post to explain what the book is about.
Feedback is always welcome 🙏
huyenchip.com/2019/07/21/mac…
I also wrote this post to explain what the book is about.
Feedback is always welcome 🙏
huyenchip.com/2019/07/21/mac…
Machine learning people have very different idea of what is important. A ML engineer asked me if convex optimization is 'a Stanford thing' because he's never used it in his work. How important is the basic knowledge of convex optimization when working with ML models?
Alternative SWE interview format: code comprehension. Give candidates a piece of code and ask them to explain/improve it. Programming languages are languages. Why shouldn’t we evaluate someone's ability to code the same way we evaluate their ability to use languages?
