Building Models Given the Task Considering Cost, Complexity, Latency, and Scalability PROFESSIONAL-MACHINE-LEARNING-ENGINEER Practice Questions (Page 1)

Questions 1–5

1. 1foundation · easy

   A mobile app uses an ML model to detect objects in photos. The app must work offline and provide results in under 100 milliseconds. Which deployment strategy is most appropriate?

   • AServerless prediction
   • BEdge deployment
   • CBatch prediction
   • DOnline prediction
   Reveal Solution Discussion Select an answer first
2. 2application · medium

   A bank is developing a model to approve loan applications. Regulatory requirements mandate that the bank must be able to explain each decision to customers. The bank's data includes hundreds of features, and they have observed that a complex ensemble model performs slightly better than a logistic regression. What should the bank do to balance interpretability and performance?

   • AUse the complex model but only provide global feature importance to customers.
   • BUse the complex ensemble model and apply SHAP to explain individual predictions.
   • CUse logistic regression exclusively because it is inherently interpretable.
   • DUse a black-box model and refuse to provide explanations due to proprietary reasons.
   Reveal Solution Discussion Select an answer first
4. 3application · medium

   A healthcare startup needs to classify medical images (X-rays) into disease categories. They have a small data science team with limited deep learning experience, and data residency requirements mandate that all data remain in the EU. They need a solution that minimizes development time while allowing some customization of the model architecture. What should they choose?

   • AVertex AI AutoML Vision with data stored in an EU region.
   • BBigQuery ML to train a custom CNN on image data stored in the EU.
   • CDeploy a pre-trained LLM to classify images via text descriptions.
   • DVertex AI Pipelines to build a custom TensorFlow training pipeline with data in the US.
   Reveal Solution Discussion Select an answer first
5. 4application · medium

   A manufacturing company wants to deploy a defect detection model to run on cameras at each factory line. The model must make predictions locally with minimal latency and continue operating even if internet connectivity is lost. The model is a lightweight CNN. Which deployment strategy is most appropriate?

   • ADeploy the model to edge devices at each factory line.
   • BDeploy the model as an online endpoint in the cloud.
   • CUse BigQuery ML to run predictions on streaming data.
   • DUse batch prediction on images uploaded nightly.
   Reveal Solution Discussion Select an answer first
6. 5application · medium

   A retail company needs to forecast daily store-level sales for 500 stores to plan inventory replenishment. They require predictions for the next 14 days, updated nightly, with a strict budget. Historical data spans 3 years and shows strong seasonal patterns. The team has limited ML expertise and needs to minimize ongoing operational overhead. Which approach is most appropriate?

   • ADeploy a pre-trained LLM to generate forecasts from textual sales summaries.
   • BUse BigQuery ML to train an ARIMA model and generate batch forecasts.
   • CBuild a custom Prophet model on a Kubernetes cluster with auto-scaling.
   • DTrain a DNN with embeddings for store and date features, deploy as an online endpoint.
   Reveal Solution Discussion Select an answer first