Google ML Engineer Model Serving and Prediction

1. Which Google Cloud tool is used to containerize ML models for serving?

Cloud Build and Cloud Run

BigQuery only

Dataprep

Cloud Composer

2. In model serving, what is the benefit of using gRPC over REST?

Lower latency and better performance for high-throughput scenarios

Easier to debug

Simpler code syntax

Better for batch processing

3. What does serving infrastructure need to handle for reliable production predictions?

Scalability, fault tolerance, and low latency

Only data storage

Only model training

Only feature engineering

4. Which approach optimizes inference speed by converting models to lower precision formats?

Quantization

Augmentation

Validation

Normalization

5. What is the primary purpose of A/B testing in production ML systems?

Compare model versions to evaluate performance impact

Reduce training dataset size

Eliminate feature preprocessing

Automate hyperparameter selection

6. Which metric measures the percentage of predictions that match ground truth labels?

Accuracy

Latency

Throughput

Cost

7. In Vertex AI, what is the purpose of the prediction request-response format?

Standardizes input features and output predictions

Encrypts sensitive data only

Increases training speed

Reduces storage requirements

8. What does knowledge distillation accomplish in model optimization?

Transfers knowledge from large to smaller, faster models

Increases model complexity

Removes all regularization

Eliminates feature scaling needs

9. Which prediction strategy uses multiple model versions simultaneously to reduce risk?

Canary deployment

Batch processing

Feature engineering

Hyperparameter tuning

10. What is the main advantage of using a feature store in production ML systems?

Consistent feature computation for training and serving

Eliminates the need for data validation

Automatically tunes model hyperparameters

Removes requirement for model monitoring

11. Which Google Cloud service provides managed model serving and supports both online and batch predictions?

Vertex AI Predictions

Cloud Storage

BigQuery ML

Dataflow

12. In online prediction, what is the typical acceptable latency range for real-time applications?

Milliseconds to a few seconds

Minutes

Hours

Days

13. What does containerization enable for model serving?

Consistent deployment across environments

Automatic feature scaling

Real-time model retraining

Improved hyperparameter tuning

14. Which monitoring metric is critical for detecting model drift in predictions?

Prediction output distribution changes

Training loss values

Feature importance scores

Cross-validation accuracy

15. What is the purpose of model versioning in production environments?

Track model changes and enable rollback if needed

Increase training dataset size

Reduce computational cost

Simplify data preprocessing

16. Which technique reduces model size by removing redundant weights?

Pruning

Augmentation

Regularization

Normalization

17. What does batch prediction allow ML engineers to do?

Process multiple samples asynchronously in bulk

Train models faster on GPUs

Reduce feature engineering complexity

Improve model interpretability

18. Which format is most efficient for serving TensorFlow models on Vertex AI?

SavedModel

H5

Pickle

JSON

19. In model serving, what does SLA stand for?

Service Level Agreement

System Load Analysis

Serialized Learning Algorithm

Scalable Learning Architecture

20. What is the primary benefit of using quantization in model serving?

Increases model accuracy

Reduces model size and inference latency

Improves data preprocessing speed

Enables distributed training

Google ML Engineer Model Serving and Prediction Quiz