They asked a lot of behavioral questions. Regarding the technical questions, they asked computer vision (6 DoF pose estimation methods, YOLO, image segmentation), statistics (covariance, p-value, distributions), classic machine learning algorithms (SVM, clustering, linear regression), deep learning, regularization methods. The coding question a typical leetcode question (easy level).
Explain Dropout, how it works and why?
Tell me about network you used. Why did you choose them ?
Signed NDA. Unable to disclose
How would design a system for xxx (e.g. ASR, dialogue, object recognition) from scratch ?
The first interviewer spend maximum time on random forest algorithm and went into great detail. from bagging and boosting to gradient boosting techniques. He also spent quite some time with multiple linear regression problems. In the end there was one algorithm question. The second interviewer asked about classification for imbalanced data sets and response rate of models. In the end there were few SQL questions. Both interviewers were super helpful and walked through their questions. From my experience i think they really want the candidate to succeed.
remove string duplicate
What is the Bert algorithm's advantage?
Implement a dice loss and multi class dice loss, implement a graph algorithm, some theoritical ML questions.
ML questions cover topics such as data imbalance, collinearity, feature selection, linear regression, logistic regression, L1/L2 norm regularization; ML application question -- how would you design a recommendation system that recommend books to users;
Viewing 1031 - 1040 interview questions