Draft AI Solution PMI-CPMAI Practice Questions (Page 4)

Questions 16–20

1. 16expert · hard

   A pharmaceutical company wants to use AI to discover new drug candidates. They have a large dataset of molecular structures and known drug interactions. The solution must generate novel molecular structures that are likely to be effective. Which AI model type and deployment strategy are most appropriate?

   • AUnsupervised clustering, with an on-premises training pipeline and an on-premises inference service.
   • BReinforcement learning, with a cloud-based training pipeline and an edge-based inference service.
   • CSupervised classification, with a cloud-based training pipeline and a cloud-based inference service.
   • DGenerative AI (e.g., variational autoencoders or GANs), with a cloud-based training pipeline and an on-premises inference service for security.
   Reveal Solution Discussion Select an answer first
2. 17expert · hard

   A bank wants to detect fraudulent credit card transactions. They have labeled transaction data, but the fraud rate is very low (0.1%). The solution must minimize false positives to avoid annoying customers, while still catching as many frauds as possible. Which model type and data processing approach is most appropriate?

   • ASupervised classification without any class imbalance handling, and a batch pipeline that retrains monthly.
   • BReinforcement learning, and a real-time pipeline that learns from each transaction's outcome to adjust fraud detection policy.
   • CUnsupervised anomaly detection, and a real-time pipeline that flags transactions that deviate from normal patterns.
   • DSupervised classification with class imbalance handling (e.g., SMOTE), and a batch pipeline that retrains the model daily with new labeled data.
   Reveal Solution Discussion Select an answer first
4. 18foundation · easy

   Which deployment strategy is most appropriate for an AI application that must operate with low latency in a factory setting where internet connectivity is unreliable?

   • AOn-premise deployment
   • BCloud deployment
   • CHybrid deployment
   • DEdge deployment
   Reveal Solution Discussion Select an answer first
5. 19expert · hard

   An e-commerce company wants to build a recommendation system that updates user preferences in real-time as they browse. They have a large user base and a legacy product catalog system. The solution must provide low-latency recommendations and integrate with the catalog via a REST API. Which data flow design best meets these requirements?

   • AStreaming ingestion of user browsing events, real-time feature engineering, a collaborative filtering model, and a REST API to fetch product details from the catalog.
   • BBatch ingestion of browsing events every hour, offline feature engineering, a content-based filtering model, and a REST API to fetch product details.
   • CStreaming ingestion of browsing events, real-time feature engineering, a supervised classification model, and a direct database connection to the catalog.
   • DBatch ingestion of browsing events daily, offline feature engineering, a collaborative filtering model, and a file-based export to the catalog.
   Reveal Solution Discussion Select an answer first
6. 20application · medium

   A telecom company wants to build an AI system to predict customer churn. They have customer usage data, billing data, and support tickets. The solution must provide a dashboard for business analysts and integrate with the CRM system. Which high-level architecture design is most appropriate?

   • AA data lake for raw data, a batch processing layer for feature engineering, a supervised classification model, a dashboard, and a REST API for CRM integration.
   • BA data lake, a real-time processing layer, a supervised classification model, a dashboard, and a file-based export to the CRM.
   • CA real-time streaming pipeline, an unsupervised clustering model, a dashboard, and direct database writes to the CRM.
   • DA data warehouse, a batch processing layer, a reinforcement learning model, and a REST API for CRM integration, but no dashboard.
   Reveal Solution Discussion Select an answer first