Mixed ML Q&A - Set 2 20 Core Questions
Interview Prep

Mixed Machine Learning Concepts: Q&A (Set 2)

Short mixed-topic questions on interpretability, fairness, experimentation and ML system design.

Fairness Experimentation Explainability Systems
1 What is model interpretability and why does it matter? âš¡ Beginner
Answer: Interpretability is the ability to understand and explain why a model made a prediction, which is critical for trust, debugging and regulation.
2 Name some techniques for explaining ML model predictions. 📊 Intermediate
Answer: Techniques include feature importance, partial dependence plots, SHAP, LIME and counterfactual explanations.
3 What is SHAP in simple terms? 🔥 Advanced
Answer: SHAP uses ideas from Shapley values in game theory to attribute how much each feature contributed to an individual prediction.
4 What does fairness mean in ML systems? 🔥 Advanced
Answer: Fairness roughly means similar individuals or groups are treated similarly; metrics often focus on error or decision parity across protected groups.
5 Name two common fairness metrics. 🔥 Advanced
Answer: Examples: demographic parity, equal opportunity, equalized odds.
6 Why can optimizing purely for accuracy be risky in production ML? 📊 Intermediate
Answer: Accuracy may hide minority errors, fairness issues, calibration problems or misalignment with business outcomes.
7 What is an A/B test in the context of ML systems? âš¡ Beginner
Answer: An A/B test compares two versions of a system (A vs B) on real users to see which performs better on chosen metrics.
8 What is conceptually different between offline validation and online A/B testing? 🔥 Advanced
Answer: Offline validation uses historical data and proxy metrics; A/B tests measure live impact on business KPIs with real users.
9 What is MLOps and why is it important? 📊 Intermediate
Answer: MLOps applies DevOps principles to ML, focusing on reliable training, deployment, monitoring and iteration of models in production.
10 What is a feature store and when would you use one? 🔥 Advanced
Answer: A feature store centralizes feature definitions, storage and serving for both training and inference, improving consistency and reuse.
11 How do you handle model versioning in an ML project? 📊 Intermediate
Answer: Use tools or conventions to track code, data, hyperparameters and artifacts per version (e.g., Git + MLflow/DVC).
12 Why is data quality often more critical than model choice? âš¡ Beginner
Answer: No model can fix systematic errors, label noise, or missing coverage; clean, representative data sets the ceiling on performance.
13 What is active learning in ML? 🔥 Advanced
Answer: Active learning lets the model choose which unlabeled examples to label next, aiming to improve accuracy with fewer labels.
14 What is transfer learning and when is it helpful? 📊 Intermediate
Answer: Transfer learning reuses a model trained on a related task or large dataset, then fine-tunes it on a smaller, specific dataset.
15 Why is feature importance not the same as causality? 🔥 Advanced
Answer: High importance shows a predictive association, not necessarily that changing the feature will change the outcome.
16 What are some common sources of label noise in real datasets? 📊 Intermediate
Answer: Label noise comes from manual annotation errors, ambiguous cases, outdated labels, or automated heuristics.
17 How do you decide whether to invest in a more complex model or better data? 🔥 Advanced
Answer: If simple models already hit a data-quality ceiling, better data is key; if metrics improve with complexity, model work may pay off, but always compare to baselines.
18 What is a data pipeline in ML systems? âš¡ Beginner
Answer: The data pipeline covers extracting, cleaning, transforming and delivering data consistently to both training and inference environments.
19 Why is it important to align ML metrics with business goals? âš¡ Beginner
Answer: A model can optimize a technical metric but still hurt revenue, user experience or risk if metrics aren’t aligned with real objectives.
20 What is the main takeaway from this advanced mixed set? âš¡ Beginner
Answer: Modern ML practice combines solid modeling with careful design of experiments, fairness, monitoring and system engineering.

Quick Recap: Mixed ML Concepts 2

Thinking beyond algorithms—about users, systems and long-term behavior—is what makes ML solutions reliable and responsible in the real world.

Back: Mixed ML Q&A - Set 1 Back to ML Tutorial