Task Statement 1.3: Describe the AI/ML Development Lifecycle. AIF-C01 Practice Questions (Page 3)

Questions 11–15

1. 11application · medium

   A healthcare company needs to build a model to classify medical images. They have a small labeled dataset and limited ML expertise. They want to use a pre-trained model to reduce training time and improve accuracy. Which approach should they take?

   • ATrain a custom model from scratch using Amazon SageMaker
   • BUse Amazon QuickSight to build the image classification model
   • CDeploy a generic pre-trained model without any fine-tuning
   • DUse transfer learning with an open-source pre-trained model and fine-tune it on their data
   Reveal Solution Discussion Select an answer first
2. 12expert · hard

   A company has a production ML model that is performing well, but the data science team is spending significant time manually retraining and redeploying the model whenever new data arrives. They want to automate the retraining process while maintaining the ability to roll back to a previous model version if the new model performs poorly. Which MLOps practice should they implement?

   • AManually retrain the model every month and deploy it after a quick check
   • BSet up a CI/CD pipeline that automatically retrains the model on new data, evaluates it, and deploys it with a rollback mechanism
   • CDeploy every retrained model directly to production without evaluation
   • DUse a single model and never retrain it
   Reveal Solution Discussion Select an answer first
4. 13foundation · easy

   Which AWS service provides a managed API to access foundation models from various providers, such as AI21 Labs, Anthropic, and Meta, without managing the underlying infrastructure?

   • AAmazon SageMaker AI
   • BAmazon QuickSight
   • CAmazon Q
   • DAmazon Bedrock
   Reveal Solution Discussion Select an answer first
5. 14foundation · easy

   Which source of foundation models involves using a model that has been pre-trained on large datasets by a third party and is made available for public use, often with the ability to fine-tune it for specific tasks?

   • AOpen-source pre-trained models
   • BCustom-trained models
   • CModels from academic research
   • DProprietary models via API
   Reveal Solution Discussion Select an answer first
6. 15application · medium

   A company has deployed a machine learning model to production. The model's performance has been declining gradually over the past few months, and the data science team suspects the underlying data distribution is changing. Which MLOps practice should they implement to detect and address this issue?

   • AIncrease the model's learning rate during training
   • BFreeze the model version to ensure consistent predictions
   • CUse Amazon QuickSight to visualize the model's predictions
   • DSet up model monitoring to track performance metrics and data drift, then retrain the model when thresholds are breached
   Reveal Solution Discussion Select an answer first