The Role of Data in Effective AI Systems AIE Practice Questions (Page 6)

Questions 26–30

1. 26expert · hard

   A telecom company is building a model to predict customer churn. The data includes call records, billing history, and customer support interactions. The data is collected from different systems and has inconsistent formats and missing values. The team wants to train a model that performs well across all customer segments. What is the most important first step?

   • AApply normalization to all features before integrating the data.
   • BPerform data integration and cleaning to ensure consistency and completeness across all sources.
   • CTrain a separate model for each customer segment using the raw data.
   • DUse only the call records because they are the most reliable source.
   Reveal Solution Discussion Select an answer first
2. 27application · medium

   A healthcare startup is training a supervised model to detect skin lesions from images. Two annotators label the same set of images, but they disagree on 15% of cases. The team is about to train the model. What should they do first?

   • ATrain the model and rely on its confidence scores to identify mislabeled images.
   • BRemove all images where the annotators disagreed to ensure a clean dataset.
   • CUse the labels from the more experienced annotator for all images.
   • DHave a third expert review the disagreements and establish a labeling guideline to resolve them.
   Reveal Solution Discussion Select an answer first
4. 28application · medium

   A logistics company wants to build a model to predict delivery delays. They have two data sources: a small, clean dataset of 5,000 records from one region, and a large, noisy dataset of 500,000 records from multiple regions. The team wants good generalization across all regions. What is the best approach?

   • ATrain two separate models, one on each dataset, and average their predictions.
   • BUse only the small, clean dataset to ensure high data quality.
   • CCombine both datasets without cleaning and train on the merged data.
   • DUse the large dataset and apply preprocessing to clean it, then train the model.
   Reveal Solution Discussion Select an answer first
5. 29foundation · easy

   What is the primary purpose of data anonymization in AI systems?

   • ATo comply with data storage regulations
   • BTo increase the accuracy of the AI model
   • CTo speed up the training process
   • DTo protect individual privacy by removing or altering personally identifiable information
   Reveal Solution Discussion Select an answer first
6. 30expert · hard

   A multinational company is training a global customer-service AI. The data includes personal data from customers in different countries with varying privacy regulations. The company wants to maintain a single global model while complying with data residency requirements. What is the most appropriate strategy?

   • ATransfer all data to a central location and anonymize it before training.
   • BTrain separate models in each region using only local data and combine their insights through federated learning.
   • CTrain the model in one region and deploy it globally without local adaptation.
   • DUse only publicly available data to avoid privacy regulations.
   Reveal Solution Discussion Select an answer first