Senior Data/Applied Scientist onsite: What questions?
Have an onsite with the Bing team for Senior DS/AS position with MSFT next week. Besides leetcode type questions, what can I expect and prep for? DS questions are super broad and any help to narrow down and focus would be much appreciated. At other companies I've seen everything from: 1) SQL questions 2) Prob questions (e.g. coin toss, card counting) 3) Stats questions (e.g. A/B testing) 4) Classical ML questions 5) DL questions 6) Specialized DL (e.g. computer vision, NLP) questions Which one do I focus my prep on?
1. Explain bagging and boosting. How are they different? 2. What are the assumptions of OLS? What happens if the assumptions are violated? 3. What is cointegration? 4. What are spurious correlations between 2 time series? 5. What is the null space of a matrix? How do I compute vectors spanning the null space? 6. What is a conjugate prior? 7. What are MAP and MLE in Bayesian estimation? 8. What does the Viterbi algorithm do? 9. What does the EM algorithm do? 10. How does matrix factorization work in a recommender system? 11. How is ANOVA related to linear regression? 12. What is the loss function used in logistic regression? Why?
Feynman told me not to memorize stuff I can look up though. I feel like a more open ended scenario based set of questions would be better. You could get a feel for their intuition and the extent of their k knowledge base. Treat like the Oregon trail of ML interviews. Although most of your questions are the fundamentals.
As a senior DS, I hope I’d know what bagging is without looking it up. Note that when I asked about the Viterbi algorithm, I wasn’t asking for pseudo code. I was just asking what it does.
May I ask your background and/or YOE? Also PhD?
Yes PhD with 4 yoe post PhD. Does it matter? Do they change the questions based on my background?
It might. Study ML system design if you have time. Big data architectures, etc
Focus on the coin toss.
From my experience lots of classical two tail stats tests, A/B testing with irregular datasets, and experiment design.
Not for applied scientist. It will be more around ML and theory and deep learning maybe if the role requires it, which it does.
Oh I completely agree. The bar for AS is going to be at a different level than DS.