Amazon·Machine Learning Engineer·Technical Phone Screen·Senior

SeniorPrefer not to say

May 2026

Summary

Amazon ML Engineer interview, one technical screen focused entirely on ML fundamentals. The bias-variance question was the main event and it went... okay, I think.

Questions Asked (1)

Can you explain the bias-variance trade-off? Walk through what bias and variance mean, how error decomposes into those components plus irreducible noise, how model complexity plays into it, and how you'd actually apply this when picking or tuning a model.

Technical Trade-offsAlgorithms & Data Structures

Author's notes

I started with the textbook decomposition and I think that was fine, but I fumbled a bit when they pushed on the practical side.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by defining bias and variance clearly, then explain the decomposition of expected error into bias^2, variance, and irreducible noise. Discuss how model complexity affects each component, and finish with practical strategies for diagnosing and addressing bias/variance issues when selecting and tuning models.

Pro tip: Emphasize that the goal is not to minimize bias or variance individually but to find the right balance that minimizes total error, and mention that in practice, you often use learning curves and validation curves to diagnose whether a model is suffering from high bias or high variance.

1. Define bias and variance

Explain bias as the error from erroneous assumptions in the learning algorithm (underfitting) and variance as the error from sensitivity to small fluctuations in the training set (overfitting).

2. Explain error decomposition

Describe how expected prediction error decomposes into bias^2 + variance + irreducible error (noise), and clarify that irreducible error is the noise inherent in the data that no model can eliminate.

3. Relate to model complexity

Discuss how increasing model complexity typically decreases bias but increases variance, and vice versa, leading to the trade-off. Mention that the optimal complexity minimizes total error.

4. Apply to model selection and tuning

Explain practical steps: use learning curves to diagnose bias/variance, then apply techniques like regularization, cross-validation, ensembling, or collecting more data to address the dominant issue.

Key Points to Mention

Bias-variance decomposition formula: E[(y - f̂(x))^2] = Bias[f̂(x)]^2 + Var[f̂(x)] + σ^2
Underfitting corresponds to high bias, overfitting to high variance
Model complexity: simple models (e.g., linear regression) have high bias, low variance; complex models (e.g., deep neural nets) have low bias, high variance
Irreducible error (noise) is the lower bound on expected error and cannot be reduced by any model
Techniques to reduce bias: add features, increase model complexity, reduce regularization
Techniques to reduce variance: add more data, regularization (L1/L2), dropout, bagging, early stopping

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.