Microsoft Azure AI Fundamentals Machine Learning 4 — Questions and Answers
Question 1: Which approach does Azure Responsible AI use to detect and mitigate bias in ML models?
- Deleting sensitive columns before training
- Using the Fairness dashboard and Responsible AI tools to assess and mitigate bias (Correct answer)
- Applying equal sample weights to all groups
- Restricting model access to certain users
Correct answer: Using the Fairness dashboard and Responsible AI tools to assess and mitigate bias
Azure's Responsible AI toolkit, including the Fairness dashboard, helps assess model disparity across demographic groups and guides mitigation strategies.
Question 2: What is a 'confusion matrix' used for in machine learning evaluation?
- Visualizing neural network architecture
- Summarizing correct and incorrect predictions across classes for classification models (Correct answer)
- Plotting feature importance scores
- Comparing training vs. validation loss curves
Correct answer: Summarizing correct and incorrect predictions across classes for classification models
A confusion matrix shows true positives, true negatives, false positives, and false negatives, providing a complete picture of a classifier's performance per class.
Question 3: In Azure ML, what is the role of an 'experiment'?
- A deployed endpoint serving predictions
- A logical container that groups and tracks related training runs (Correct answer)
- A dataset stored in Azure Blob Storage
- A GPU compute cluster configuration
Correct answer: A logical container that groups and tracks related training runs
An Azure ML experiment is a named grouping of runs, allowing you to organize, compare, and track the results of multiple model training iterations.
Question 4: Which data split ratio is most commonly used for dividing data into training and test sets in Azure AutoML's default behavior?
- 50% training / 50% test
- 70–80% training / 20–30% test (Correct answer)
- 90% training / 10% test
- 60% training / 40% test
Correct answer: 70–80% training / 20–30% test
A 70–80% training and 20–30% test split is the industry standard and Azure AutoML's default approach to balance learning opportunity with evaluation robustness.
Question 5: What does RMSE (Root Mean Squared Error) measure in a regression model?
- The percentage of predictions within one unit of the true value
- The average magnitude of prediction errors, giving more weight to large errors (Correct answer)
- The correlation between predicted and actual values
- The ratio of variance explained by the model
Correct answer: The average magnitude of prediction errors, giving more weight to large errors
RMSE measures the square root of the average squared differences between predicted and actual values, penalizing larger errors more heavily than MAE.
Question 6: Which Azure ML feature allows you to track, version, and reproduce ML experiments across your team?
- Azure DevOps Pipelines
- MLflow integration in Azure Machine Learning (Correct answer)
- Azure Logic Apps
- Azure Monitor
Correct answer: MLflow integration in Azure Machine Learning
Azure Machine Learning integrates natively with MLflow for experiment tracking, model versioning, and reproducibility across teams and environments.
Question 7: What is 'data drift' in the context of deployed ML models?
- Data loss during model training
- Changes in the statistical properties of input data over time that degrade model accuracy (Correct answer)
- Gradual increase in model file size
- Mismatch between training labels and features
Correct answer: Changes in the statistical properties of input data over time that degrade model accuracy
Data drift occurs when the distribution of real-world input data changes after deployment, causing the model's predictions to become less accurate over time.
Microsoft Azure AI Fundamentals Machine Learning 4
Which approach does Azure Responsible AI use to detect and mitigate bias in ML models?