SchoolWhool
Quizzes and saved progress, freeSign in freeSign in
Maths · Module 4Module material

Calculus and Optimisation interview questions

15 questions interviewers ask on this module’s topics, from first principles to the follow-ups. Each comes with a model answer.

  1. What is a derivative, in one sentence, and how would you compute one numerically?
  2. Why not make the finite-difference step as small as possible?
  3. State the chain rule and use it on the sigmoid.
  4. You set the derivative to zero and found a point. What else do you need to check?
  5. Which single number minimises squared error, and which minimises absolute error?
  6. What is the gradient, and why does gradient descent step against it?
  7. What are the gradients of aᵀx, xᵀx and xᵀAx?
  8. How do you choose a learning rate for gradient descent on a quadratic loss?
  9. What is Newton's method, and why is it not used to train neural networks?
  10. What does the AUC measure?
  11. Why does convexity matter, and is logistic regression's loss convex?
  12. What is a saddle point, and why do they matter in deep learning?
  13. Explain backpropagation.
  14. Derive the gradient of logistic regression's log loss.
  15. What does a Lagrange multiplier mean, and how does it relate to ridge and lasso?