MLOps and Deployment

Question 1

A company needs to deploy an ML model for real-time product recommendations. Traffic is highly variable, with unpredictable spikes and idle periods. Which SageMaker deployment option is most cost-effective?

Accepted Answer

SageMaker Serverless Inference

Answer

SageMaker Real-time Inference

Answer

SageMaker Batch Transform

Answer

SageMaker Asynchronous Inference

Question 2

Which AWS SageMaker tool is used to automatically detect 'data drift' and 'model drift' in production?

Accepted Answer

SageMaker Model Monitor

Answer

SageMaker Debugger

Answer

SageMaker Clarify

Answer

SageMaker Pipelines

Question 3

A company needs to incorporate human review for its model predictions when confidence is below 80%. Which service facilitates this 'Human-in-the-loop' workflow?

Accepted Answer

Amazon Augmented AI (A2I)

Answer

Amazon Mechanical Turk

Answer

SageMaker Ground Truth

Answer

SageMaker Model Registry

Question 4

Which AWS service is used to create, automate, and manage end-to-end machine learning workflows (CI/CD for ML)?

Accepted Answer

SageMaker Pipelines

Answer

AWS CodePipeline

Answer

AWS Step Functions

Answer

SageMaker Experiments

Question 5

An ML team wants to test a new model version in production by sending it 10% of traffic, while the remaining 90% goes to the old model. Which deployment strategy should they use?

Accepted Answer

A/B Testing or Canary Deployment

Answer

Shadow Testing

Answer

Blue/Green Deployment

Answer

Re-training

Question 6

Which service is used to manage, catalog, and version ML models, allowing tracking of their approval status (e.g., 'Pending', 'Approved')?

Accepted Answer

SageMaker Model Registry

Answer

SageMaker Model Monitor

Answer

SageMaker Feature Store

Answer

Amazon ECR

Question 7

Which EC2 instance type is most suitable for distributed training of large and complex Deep Learning models?

Accepted Answer

Accelerated computing instances (P4, P5 with GPU)

Answer

Compute optimized instances (C5)

Answer

General purpose instances (M5)

Answer

Memory optimized instances (R5)

Question 8

A company wants to optimize inference costs. They have multiple small models that are used infrequently. Which SageMaker feature allows hosting multiple models on a single endpoint to share resources?

Accepted Answer

SageMaker Multi-Model Endpoints (MME)

Answer

SageMaker Serial Inference Pipelines

Answer

SageMaker Inference Recommender

Answer

SageMaker Model Monitor

Question 9

Which service is used to run predictions on a large dataset (e.g., 1 TB of files) all at once, without needing a real-time endpoint?

Accepted Answer

SageMaker Batch Transform

Answer

SageMaker Real-time Inference

Answer

SageMaker Canvas

Answer

AWS Lambda

Question 10

Which security feature allows connecting SageMaker Studio to data resources within your VPC without traffic traversing the public Internet?

Accepted Answer

VPC Endpoints (AWS PrivateLink)

Answer

AWS Internet Gateway

Answer

AWS Direct Connect

Answer

NAT Gateway

MLOps and Deployment

A company needs to deploy an ML model for real-time product recommendations. Traffic is highly variable, with unpredictable spikes and idle periods. Which SageMaker deployment option is most cost-effective?

Explanation

Inference Options