1. One web screening (coding). This is the standard sort of leetcode question
2. One Leetcode interview. Wasn't actually hard, just some Leetcode medium question involving divide-and-conquer
3. Two interviews with research scientists. They asked me about my research direction, basically just spoke about my research for a full hour. Asked plenty of questions about my papers.
Interview questions [1]
Question 1
If I have a deep neural network, and I increase the training batch size to be $K$ times larger, how should I scale my hyperparameters (learning rate, momentum)?
I applied online. The process took 4 weeks. I interviewed at Meta (Menlo Park, CA) in Feb 2021
Interview
There were two interview rounds: the first was a one-hour technical interview involving a LeetCode-style coding problem, and the second was a one-hour research discussion centered on my previous publications, academic contributions, and relevance to the role.
Interview questions [1]
Question 1
1st round (Coding): It was Leetcode hard, which was surprising to me as this was an interview for an internship. It was the LC 297 (Serialize and Deserialize Binary Tree). I managed to provide a solution, but I am not sure if it was the correct one, but the interview went well overall.
2nd round (Research): It was a discussion with the hiring manager. During the first half of the interview, we discussed about one of my previous papers. In the second half, we discussed about my workflow when it comes to research and there were a also few behavioural questions.
2 rounds, 1 coding 1 research. they are scheduled consecutively, each 45mins, the research interviewer is also the hiring manager / actual manager during the internship. there could be one more research if another manager is also interested in you.
Interview questions [1]
Question 1
coding: implement k-means from scratch
research: talk through my papers and their project