This was an applied scientist internship for (explicitly) Computer Vision, Machine Learning, Speech Technologies, and Robotics. I state the position I applied for as it becomes more relevant later in the process.
My recruiter first messaged me stating that someone from the ‘Vesta’ team was wanting to interview me. I did not know about the ‘Vesta’ team, so I tried googling. I found very little, other than articles from 2018/2019 about a home robot. Since I was trying to prepare for this interview, I asked the recruiter what I can expect during the interviews and what topics I may be asked. I was told the first interview was on ‘Domain Expertise’ and the second was ‘Problem Solving’. The recruiter said that the interviewers generate their own questions for the interviews. I may be asked ‘machine learning-related’ questions during the ‘Domain Expertise’ as well as the ‘Problem Solving’ interview. The recruiter also said there may be small coding questions in either interview. I personally have fair amount experience in Machine Learning and Computer Vision, so I was not terribly concerned by this.
The first interview went in my opinion, really well. I was asked machine learning and behavioral questions based on my previous experience.
The second interview should have its named changed from ‘Problem Solving’ to “Reinforcement Learning Expertise”. First of all, I do not have a lot of experience in Reinforcement Learning. I have one project on my resume which is based on reinforcement learning and the remaining five are Computer Vision and Machine Learning related. Second, I applied for “Applied Science Intern - Speech Technologies, Computer Vision, Machine Learning, Robotics”. You could easily say the Robotics part covers Reinforcement Learning, but I was not expecting to be chosen for that. The limited information I received from the recruiter explicitly stated, ‘Machine Learning’, and it was mentioned multiple times. Needless to say, I was not prepared for a fully ‘Reinforcement Learning’ interview.
I feel mislead by my recruiter because I was provided very limited information about the internship. Literally, what I wrote above, is all I was given…And the recruiter told me multiple times it was a machine learning internship.
- ( I recorded my responses, and I am deriving the questions from my response )
- Now that I wrote out my questions… I feel like I was asked a lot of questions…
Interview 1 Questions:
• What is Bagging and how is it implemented?
• What is an example of a Bagging algorithm?
• What is Boosting and how is it implemented?
• What is an example of a Boosting algorithm?
• What is the difference between Bagging and Boosting?
• Have you heard of a Decision Tree?
• What are some ways to split the data at the nodes in the tree?
• What are some disadvantages of Decision Trees?
• What is Overfitting?
• What are some ways to reducing a Decision Trees ability to overfit?
• What is Bias and Variance?
• What is the Bias-Variance Trade off?
• What is an Ensemble and how is it implemented?
• What is an example of an Ensemble?
• Given a Balanced classification data set, what are ways you could measure the performance of your model?
• What is Precision?
• What is Recall?
• What is F1-Score?
• What if the data set is unbalanced, what are some measurements you could use?
• What is ROC curve?
• How do you interpret the ROC curve?
• What is AOC?
• What are some ways to reduce overfitting?
• What is Regularization?
• What is the difference between L1 and L2?
• What are some ways of reducing the complexity of a model?
• Specifically, WRT Deep Learning, what are some techniques of regularization?
• How does an Autoencoder work?
• What is the Curse of Dimensionality?
• Ways to counteract the Curse of Dimensionality?
• What is PCA and ICA?
• How could I use a Random Forest for Dimensionality Reduction?
Interview 2 Questions:
• Tell me about yourself
• What is a Random Forest (asked in both interviews)?
• How does a Random Forest work?
• What is a reason for using a Random Forest?
• Methods of splitting on the nodes?
• What are some ways to reduce overfitting in Random Forest with unlimited depth trees?
• If I wanted to convert a pruned decision tree to a regression tree, how would I do it?
• What is Feature Importance, WRT decision trees?
• What are some dimensionality reduction techniques?
• What is a MDP (Markov Decision Process)?
• What is a Markov Chain?
• What is the difference between an MDP and Markov Chain?
• What is a Transition Matrix?
• How do you get a Transition Matrix?
• What is the formula for the Bellman Equation?
• What does the Bellman Equation do?
• What is a Policy?
• What are ways to get/make/create a Policy?
• What is Value-Iteration?
• What is Policy-Iteration?
• What is Q-Learning?
• Does Q-Learning use Policy-Iteration or Value-Iteration?
• What is a Q-Table?
• How is a Q-Table populated?
... Not Enough room, adding remaining as follow-up
