I work on applied ML at a well know public company but don’t have a grad degree. Recruiter severely discouraged me from taking ML interviews because the bar is extremely high and she has seen many PhDs with 10+ yoe rejected. She said I could still match for ML teams if I go through generalist interviews, and a few people she helped with in the past month had succeeded in this route. I saw someone say it’s difficult to match for ML teams going through the generalist process, so is what my recruiter said true? Which process do you recommend if you have been there or are familiar with it? I’m unlikely to join any non ML team
If your perf is solid, you can switch easily to ML teams in year.
Solid = CME or above? How difficult is it to match for an ML team before joining?
Not sure how your theory is but it finds its way into many ML interviews. Experience is cool but since ML requires constantly updating knowledge since it's so new, white paper type questions are common.
Ignore the recruiter, it's not that high.
Can you share what type of questions are asked at ML interviews?
"How will you design a system that identifies users who should be shown x type (eg sports) of ads on YouTube" You are expected to give a 1 hour lecture on all ML aspects of this problem.
I think usually what happens is the recruiter doesnt cover the ML team and thus if you eventually go there they wont get a bonus / smaller bonus (im just guessing). I applied to Goldman and asked for the QIS team, and the recruiter also told me the bar is extremely high lol...
Don’t undersell. Also depends are you willing to let go Google coz you want to only do ML. I would not join Google at the risk of not doing ML
I wouldn’t consider Google if I can’t join an ML team, unless they uplevel me or give me extremely high TC (which will never happen lol). Have you interviewed for generalist or ML? Would you recommend shooting for ML directly?
I’m also interviewing for ML at FB, what questions should I expect for the ML system design interview?
I may soon face a similar question, as a generalist SWE who currently does ML. Curious to see what people say here. What is the best preparation for ML interviews? There doesn't seem to be a LeetCode equivalent for ML interviews.
Google ML interviews are not very hard. Mostly design and few basic ML questions. Fb ML is also pretty standard with 2 coding rounds 1-2 ML design with standard classification or ranking problems. Amazon, Apple and Microsoft do far more deep dives on ML core areas, but most likely bcoz the interviews are team specific and good ML teams will try to test on the areas they are interested in.
Thanks for the tips. When you say ML design, do you mean implementing a classification algorithm (e.g. MLP neural net, linear SVM or something else) from scratch? Or just describing a general approach to taking such an algorithm to prod (e.g. First I featurized my data. Then I trained a random forest with scikit learn and deployed it.)
Seems like most ML work at my company consists of training lots of different models to see which performs best, and then deploying it. Mostly people just use scikit learn. It's very rare for anyone to implement ML algorithms from scratch
A recruiter called me sometime back and she said the same to me. I haven't responded but felt a bit disappointed.