ML Pipeline Orchestration on Google Cloud

1. In Vertex AI Pipelines, how are pipeline artifacts typically stored?

In local machine memory

In Cloud Storage buckets

In a local database

In a Kubernetes pod

2. What is the primary goal of A/B testing in a production ML pipeline?

To reduce pipeline costs

To compare model versions and validate improvements before full deployment

To train multiple models simultaneously

To eliminate data preprocessing

3. In ML pipeline design, what is the advantage of using feature stores?

Centralized feature management, consistency, and reusability across pipelines

Automatic model training

Reduces the number of data scientists needed

Eliminates the need for preprocessing

4. Which tool would you use to schedule recurring ML pipeline runs on a daily basis?

Cloud Storage

Cloud Scheduler with Vertex AI Pipelines

Cloud CDN

Cloud Armor

5. What is data drift in the context of ML pipeline monitoring?

The process of cleaning training data

A shift in input data distribution causing model performance degradation

The speed of data transfer

The storage of historical data

6. Which approach allows multiple pipeline tasks to run simultaneously?

Sequential execution

Parallel execution

Batch processing only

Asynchronous scheduling

7. In a production pipeline, what is the role of model registry or model versioning?

Tracks model versions, metadata, and deployment history for reproducibility

Stores raw training data

Encrypts model files

Reduces model file size

8. What does hyperparameter tuning accomplish in an ML pipeline?

Reduces storage requirements

Optimizes model performance by finding best parameter values

Eliminates the need for training data

Speeds up data ingestion

9. Which Google Cloud service integrates with Vertex AI for real-time model serving?

Cloud Pub/Sub

Vertex AI Endpoints

Cloud Dataflow

BigQuery ML

10. What is a common strategy for handling model retraining in production pipelines?

Manual retraining once per year

Scheduled retraining based on data drift or performance degradation

Retraining only when the model fails

Never retraining after deployment

11. What is the primary purpose of ML pipeline orchestration in production environments?

To automate repetitive ML tasks and ensure consistent workflow execution

To manually train models multiple times per day

To eliminate the need for data preprocessing

To replace data scientists with automated systems

12. What is the purpose of data validation steps in an ML pipeline?

To ensure data quality and detect anomalies before training

To encrypt sensitive information

To increase pipeline execution speed

To reduce model accuracy

13. Which metric is most important for monitoring model performance in a production pipeline?

Pipeline execution time only

Model accuracy, latency, and data drift

Storage capacity used

Number of pipeline runs

14. What does CI/CD stand for in the context of ML pipeline deployment?

Cloud Infrastructure and Configuration Data

Continuous Integration and Continuous Deployment

Containerized Integration and Cloud Development

Compute Integration and Code Distribution

15. In a production ML pipeline, what is the primary benefit of containerizing components?

Faster data transfer speeds

Consistency, portability, and reproducibility across environments

Reduced storage costs

Automatic model retraining

16. Which feature allows you to run pipeline steps conditionally based on upstream results?

Branching logic

Parallel processing

Data validation

Model versioning

17. What is a directed acyclic graph (DAG) in the context of ML pipeline orchestration?

A data storage format

A representation of pipeline tasks and their dependencies with no cycles

A type of neural network architecture

A database query language

18. Cloud Composer is built on which open-source orchestration framework?

Kubernetes

Apache Airflow

Spark

TensorFlow

19. In Vertex AI Pipelines, what does a component represent?

A reusable containerized task or step in a pipeline

A machine learning model only

A data storage location

A user interface dashboard

20. Which Google Cloud service is designed specifically for orchestrating ML workflows?