Rapid Fire MLE Knowledge Questions - Core ML Concepts Interview

Question

## Round 1 - ML Knowledge

## Problem

This is a rapid-fire conceptual round. Expect 8-12 short questions covering ML fundamentals. Representative questions:

**Q1:** What is the bias-variance tradeoff? Give an example of a high-bias and a high-variance model.

**Q2:** You train a model with 99% training accuracy and 60% validation accuracy. What is happening and what are three remedies?

**Q3:** Explain the vanishing gradient problem. Which architectures mitigate it and how?

**Q4:** Compare L1 and L2 regularization. When would you prefer L1?

**Q5:** What is the difference between bagging and boosting? Name one algorithm for each.

**Q6:** Your classification model has 95% accuracy on an imbalanced dataset (5% positive class). Is 95% accuracy meaningful? What metric would you use instead?

**Q7:** Explain how attention in Transformers works at a high level. What problem does it solve that RNNs struggle with?

**Q8:** You need to deploy a model that makes predictions in under 10ms. What techniques reduce inference latency?

## Follow-ups

1. Walk through the forward and backward pass of a single-layer neural network with one hidden unit.
2. What is data leakage and how does it cause deceptively high validation scores?
3. Describe how you would debug a model that has good offline metrics but poor A/B test results.
4. When would you choose a tree-based model over a neural network for a tabular dataset?

Rapid Fire MLE Knowledge Questions - Core ML Concepts Interview

Question Details

Full Details

Topics

More from Pinterest