Katkat
Back

Answer practical ML foundations questions

LinkedInMachine Learning Engineer · Onsite
MediumMachine Learning EngineerOnsiteMachine Learning

Last updated

This question evaluates an interviewee's practical mastery of ML foundations, covering probability calibration and its evaluation, feature selection methods including neural-network approaches, tokenizer types and trade-offs in NLP, and optimizer behavior such as Adam's maintained statistics.

In an ML interview, you are asked a series of practical ML foundation questions:

  1. Model outputs probabilities. When do you need probability calibration, and how would you calibrate? How do you evaluate calibration quality?
  2. Feature selection: why do it, and what practical methods do you use (filter/wrapper/embedded)? How could you perform feature selection using neural networks?
  3. Tokenizers: what is a tokenizer, what are common types (BPE/WordPiece/unigram), and what practical trade-offs matter?
  4. Optimizers: explain Adam at a high level (what statistics it maintains and why), and when it can fail or need tuning.

Provide clear, practical answers with examples.