Design a Double Descent Experiment
Company: Anthropic
Role: Machine Learning Engineer
Category: Machine Learning
Difficulty: medium
Interview Round: HR Screen
You are given a take-home assignment for a mechanistic interpretability or machine learning interview.
Design an experiment that demonstrates **sample-to-feature-ratio double descent** in a supervised learning problem. Your submission should be possible to complete in roughly four hours and summarized in a short slide deck.
Your task is to:
1. Choose a concrete learning setup and dataset generation process.
2. Vary the number of training samples relative to the feature dimension so that the test error curve shows double descent, especially near the interpolation threshold.
3. Provide a theoretical explanation for why the double descent behavior appears.
4. Propose at least one method to mitigate the phenomenon and explain why it should help.
5. Present the experimental setup, plots, theory, and mitigation clearly in slides.
A simple high-dimensional linear regression setting is acceptable if it cleanly exhibits the effect.
Quick Answer: This question evaluates understanding of double descent and generalization phenomena in supervised learning, requiring skills in experimental design, statistical learning theory, and empirical analysis.