Round1: [45 Minutes]

First, the interviewer introduced himself and asked me to tell him about myself, domains I’ve worked on, projects, etc.

Coming to the interview, there were 3 case studies:

1. Given a set of features and actual label as an ordinal feature(0-4), we pass it to the binary black-box classifier that will provide the probability of y=1, i.e., P(yi=1). The task is to predict ordinal value(0-4) using the probability of the above black-box classifier.
2. Given paraphrased sentences, the task is to find whether sentences are paraphrases or not? How do you come up with features like feature engineering techniques? Which deep learning model would you prefer in this case and why?
3. Given a set of unlabeled images stored somewhere, the user will provide queries in the form of colors like if a user writes red, need to retrieve corresponding images? Colors can be treated as a categorical feature(at most 25 colors for this problem). How to scale it or optimize the search engine to retrieve images efficiently?

Tips: Keep interacting with the interviewer so that he can know in which direction you’re thinking.