The Role of Data in Effective AI Systems AIE Practice Questions (Page 3)

Questions 11–15

1. 11application · medium

   A healthcare startup is developing a supervised model to classify skin lesions from images as benign or malignant. They have a large dataset of images, but only a small subset is labeled by dermatologists. To improve model performance, they plan to use the unlabeled images. Which approach is most appropriate?

   • ATrain the model only on the labeled subset, ignoring the unlabeled images entirely.
   • BUse active learning to have the model identify the most uncertain images for dermatologists to label first.
   • CAutomatically label all unlabeled images using a rule-based color threshold, then train on the combined dataset.
   • DRandomly select a large sample of unlabeled images and have a non-expert annotate them to save time.
   Reveal Solution Discussion Select an answer first
2. 12application · medium

   A team is labeling images for an object-detection model. The labeling guidelines are vague, and different annotators label similar objects differently. The model's performance is expected to suffer. What is the most effective way to improve the labeling process?

   • ADevelop detailed labeling guidelines and conduct regular quality checks on annotations.
   • BUse a pre-trained model to auto-label the images and skip manual annotation.
   • CIncrease the number of annotators to get more labels for each image.
   • DTrain the model on the current labels and hope it learns to ignore inconsistencies.
   Reveal Solution Discussion Select an answer first
4. 13application · medium

   A financial institution is training a fraud-detection model using customer transaction data. The data includes personally identifiable information (PII) such as names and account numbers. The compliance team requires that the data be protected and that the model not expose individual customer details. What should the data engineering team do?

   • AObtain customer consent for using their data, and proceed with the original dataset.
   • BEncrypt the data at rest and in transit, but keep the raw PII in the training dataset for accuracy.
   • CDelete all PII fields and train only on aggregated transaction summaries.
   • DAnonymize the data by removing direct identifiers and applying techniques like k-anonymity before training.
   Reveal Solution Discussion Select an answer first
5. 14foundation · easy

   Which data preprocessing step involves converting data to a common scale so that features with larger magnitudes do not dominate the model?

   • AData transformation
   • BNormalization
   • CData cleaning
   • DData annotation
   Reveal Solution Discussion Select an answer first
6. 15expert · hard

   A medical imaging company is labeling X-rays for a pneumonia-detection model. The labeling is done by a mix of junior and senior radiologists. The junior radiologists label faster but with more errors. The company has a tight deadline and a limited budget. What is the most effective way to ensure label quality while meeting the deadline?

   • AHave senior radiologists review a sample of junior labels and provide feedback to improve consistency.
   • BUse only senior radiologists to label all images, even if it delays the project.
   • CAutomatically discard any image where junior and senior labels differ.
   • DUse junior labels for all images and rely on the model to learn from the noise.
   Reveal Solution Discussion Select an answer first