19-01-29 Machine Learning Interview Recombination

Category: Idea Lists (Upon Request)

Read the original document

<!-- gdoc-inlined -->

Interview questions are about the whats, whys, hows and sometimes whens. Less the whos. I have topics at Comprehensive Machine Learning Topics. These topics can be turned into questions in many ways, which I’d like to outline in general (in the abstract) in an algorithm that can be run on any topic.

1. What is x?
2. How do you implement x?
   1. Why implement x with y rather than z?
   2. What do you do to implement x? (Same, phrased as a what)
   3. Write down some code that implements x.
      1. Ideas for improving the runtime / accuracy / training speed of x?
3. For each category, what is the category? Name elements of the category. Explain the element you know best.
   1. Ex., what is matrix factorization? Name a few types (SVD, LU decomposition, …) explain how to implement the type you know best.
4. When is x useful?
   1. When is x1 more useful than x2?
5. What is the difference between x and y?
   1. How is x different than y? (Same, phrased as how rather than what - is there some deep similarity between what and how?)
6. What are the tradeoffs involved in using x?
   1. What are the downsides to x?
   2. What are the downsides to using x with y?
   3. What are the limitations to x?
   4. How would we ameliorate those tradeoffs?
7. Discuss / Describe the concept of x.
8. Is this property of x true?
9. Solve this problem. (The problem requires the use of x)
10. What problem does x solve?
11. Compute x.

Source: Original Google Doc