AWS AI/ML Landscape and Workflow: 10 practice questions

Question 1 of 10

Which description best defines cloud computing in the AWS overview?

1. ADesktop software installed locally with no remotely managed infrastructure
2. BOn-demand IT resources delivered over the internet with pay-as-you-go pricing
3. CA private data center maintained by the customer for unlimited capacity
4. DPhysical servers purchased in advance with fixed-capacity annual pricing

Question 2 of 10

What does Amazon Bedrock provide at the introductory level?

1. AA fixed tabular dataset that replaces model selection and evaluation
2. BA batch-only hosting system for traditional regression model endpoints
3. CManaged access to foundation models for generative AI applications
4. DCustomer-owned GPU hardware for local applications

Question 3 of 10

What begins the typical machine learning workflow described for SageMaker AI?

1. AMonitor production drift before any model has been trained
2. BDeploy an unevaluated model into the production application
3. CGenerate and prepare example data for the problem
4. DDelete the training data before selecting an algorithm

Question 4 of 10

Which AWS cloud characteristic addresses a workload whose required capacity changes over time?

1. APurchase peak hardware capacity and keep the cost fixed permanently
2. BMove the workload on premises
3. CReserve one physical data center and prevent capacity adjustments
4. DScale capacity up or down and pay for the resources used

Question 5 of 10

A training dataset contains the same country recorded as both 'United States' and 'US'. Which workflow action addresses this issue?

1. AReplace evaluation with a larger production endpoint immediately
2. BClean the example data to make the values consistent
3. CIncrease inference traffic before inspecting the training examples
4. DDeploy without cleaning the values

Question 6 of 10

Which input is required to train a model in the documented workflow?

1. AA completed monitoring report created before the first training run
2. BAn application deployment with no example data or compute resources
3. CAn algorithm or a pretrained base model suited to the problem
4. DA production endpoint that has already detected future model drift

Question 7 of 10

What decision does evaluation support immediately after model training?

1. AWhether example data can be deleted before the algorithm learns from it
2. BWhether the accuracy of the model's inferences is acceptable
3. CWhether to buy physical data center hardware
4. DWhether production monitoring can be permanently removed from the cycle

Question 8 of 10

A team must build a generative AI application, wants access to foundation models from several providers, and does not want to operate the model-access infrastructure. Which starting service role fits?

1. AUse model monitoring before selecting any model or service
2. BUse a customer-owned data center
3. CUse Amazon Bedrock for managed foundation-model access
4. DUse data cleaning as a substitute for foundation-model access

Question 9 of 10

A deployed model is producing less acceptable inferences, and the team has collected newer high-quality examples. Which action best continues the documented cycle?

1. AKeep the deployed model unchanged and stop monitoring its inferences
2. BEvaluate for drift, update the training data, and retrain the model
3. CReplace data preparation with permanent ownership of fixed hardware
4. DDiscard the new examples and repeat deployment without evaluation

Question 10 of 10

A team has examples in several repositories, inconsistent category values, and two attributes that work better when combined. What sequence prepares the data before training?

1. ADelete the examples before evaluation
2. BDeploy first, monitor an untrained model, then choose the business problem
3. CFetch into one repository, clean values, then transform the attributes
4. DRetrain before the initial training run, then preserve every inconsistency