Free AWS ML Engineer Associate practice test

A. × Incorrect: Amazon S3 provides storage for data assets but does not manage or catalog them.

B. ✓ Correct: AWS Glue includes a Data Catalog that helps manage metadata and discover datasets.

C. × Incorrect: Amazon SageMaker focuses on model building and deployment, not data management.

D. × Incorrect: AWS DataBrew is used for data preparation tasks but does not provide the cataloging features of AWS Glue.

A. ✓ Correct: Amazon SageMaker AutoPilot automates the end-to-end process of building and deploying machine learning models, including feature engineering and model tuning. It simplifies the workflow for users who want to quickly deploy models without extensive coding.

B. × Incorrect: Amazon SageMaker Training Compiler optimizes performance during training jobs but does not handle the full lifecycle of model deployment.

C. × Incorrect: Amazon SageMaker Model Monitor focuses on continuous monitoring and detecting drift in model quality metrics, rather than automating the end-to-end process of building and deploying models.

D. × Incorrect: Amazon SageMaker JumpStart simplifies model deployment by providing pre-built templates and managed infrastructure but does not automate the entire lifecycle from building to deployment.

A. ✓ Correct: Model quality monitoring helps detect changes in performance metrics like accuracy, precision, recall, etc., which are crucial for maintaining operational health after deployment.

B. × Incorrect: Monitoring focuses on detecting changes or anomalies in the input data rather than the model's performance metrics.

C. × Incorrect: Feature attribution drift monitors how feature importance changes over time but does not directly measure overall model performance.

D. × Incorrect: Bias drift monitoring tracks changes in prediction bias, which can affect fairness and equity but does not cover general performance metrics.

A. ✓ Correct: Continuous monitoring with a real-time endpoint because it ensures ongoing surveillance of model performance without delays.

B. × Incorrect: Scheduled batch job execution is incorrect as it involves periodic assessments rather than continuous oversight.

C. × Incorrect: Manual data validation processes are incorrect since Model Monitor aims to automate quality checks to reduce manual effort.

D. × Incorrect: Automated retraining models while useful, this does not directly address the need for real-time monitoring.

A. ✓ Correct: Amazon SageMaker Serverless Endpoints provide fully managed infrastructure with automatic scaling to handle varying workloads efficiently, ensuring high throughput and low-latency predictions.

B. × Incorrect: Amazon SageMaker Batch Transform is designed for batch processing of large datasets and does not support real-time or low-latency predictions.

C. × Incorrect: AWS Lambda can be used for serverless computing but lacks the specialized features needed for ML model deployment, such as automatic scaling based on prediction volume.

D. × Incorrect: Amazon EC2 requires manual management and configuration to ensure efficient handling of varying workloads, which makes it less suitable compared to managed services like Serverless Endpoints.

A. × Incorrect: Fairness focuses on how outcomes affect different groups.

B. × Incorrect: Explainability is about understanding and evaluating outputs.

C. × Incorrect: Privacy and security is about handling and protecting data and models appropriately.

D. ✓ Correct: Safety is the AWS Responsible AI dimension focused on preventing harmful outcomes and misuse.

A. × Incorrect: Model deployment follows data preparation and does not ensure suitability of data.

B. ✓ Correct: B is correct as it directly addresses ensuring that data meets requirements before proceeding with other steps.

C. × Incorrect: Feature selection occurs after initial data preparation to refine the dataset for modeling.

D. × Incorrect: Algorithm tuning happens later in the process, after models are initially trained.

A. × Incorrect: Fairness pertains to ensuring equitable impacts across different groups but does not cover understanding system outputs.

B. ✓ Correct: Explainability directly addresses the need for comprehending and assessing how AI systems generate their results.

C. × Incorrect: Privacy and security involves appropriate handling of data and models but does not relate to evaluating system outputs.

D. × Incorrect: Safety focuses on preventing harm from AI systems but does not cover understanding system outputs.

A. × Incorrect: Continuous monitoring with a real-time endpoint is correct but not specific to bias detection.

B. × Incorrect: On-schedule monitoring for asynchronous batch transform jobs is correct but not specific to bias detection.

C. ✓ Correct: Bias drift for models in production because it specifically monitors changes in the model's predictions' bias over time, ensuring fairness and unbiasedness.

D. × Incorrect: Model quality because while it covers various aspects of model performance, it does not specifically address bias.

A. ✓ Correct: Feature engineering involves selecting, transforming, and creating features that align with business objectives.

B. × Incorrect: Data preprocessing while important, it focuses on cleaning and formatting data rather than aligning with specific business goals.

C. × Incorrect: Algorithm selection choosing an algorithm does not directly address the alignment of model features with business needs.

D. × Incorrect: Hyperparameter tuning optimizes existing models but does not ensure that features represent business objectives.