I want to interview a group of candidates for an ML role. I'm looking for interesting questions, beyond the usuals. Do you have any in mind that you like, you and you got in an interview and thought it was a good question?
If this is for a scientist role, you'll want to check for the breadth and then depth in area of experience. If it is for an ML engineer, you'll want to check for depth in the projects done so far.
You can get them to implement a simple ml algorithm like a tree or naive... Get get them to so some math like derive a maximum likelihood estimate or a gradient. Honestly, I still think the best thing to do is to step through an ML problem and see how they would design it end to end
Implement a tree in a interview? What is the Gini / C4.5 Splitting criteria d00d? Write down, how you are going to implement stopping criteria? How are you going to prune your trees? Write all this up without googling!
I think the only ml algorithms implementable in an in intervirew are knn and k-means.
Regression. It has a closed solution, can calculate variance of the coefficients, and can explain exactly what its doing. What is the loss function, how to change to add regularization, what are the assumptions math-ified. When to use regression vs other algo’s. I wouldn’t expect ppl to implement any other algo from scratch, maybe a NN with gradient descent but thats pretty DL specific.
Do regression but for a short fat matrix. See if they get the closed form solution, then ask how they might improve it if they had some reasonable assumptions for a model of x.
ML interviews are pretty standard. Ask him to design an ML system (recommendations, detection ...) and walk through the various parts: collecting data, evaluation, training, serving