The Role of Data in Effective AI Systems AIE Practice Questions (Page 2)

Questions 6–10

1. 6application · medium

   A mobile app company collects user location data to train a recommendation model. The company must comply with GDPR, which requires that users can request deletion of their data. The ML team wants to ensure that the model does not retain information about deleted users. What is the most appropriate approach?

   • AAnonymize the user data so that it cannot be linked back to individuals, and keep it in the training set.
   • BImplement a data deletion pipeline that removes user data from the training set and retrains the model periodically.
   • CStop collecting location data entirely to avoid future GDPR compliance issues.
   • DKeep the data in the model's training set but ensure it is encrypted to protect user privacy.
   Reveal Solution Discussion Select an answer first
2. 7application · medium

   A manufacturing company uses sensor data from machines to predict maintenance needs. The data is stored in a time-series database, and the company wants to ensure the model remains accurate over time as machines age. What is the most important data lifecycle practice?

   • AContinuously collect new sensor data and periodically retrain the model to adapt to changing machine behavior.
   • BDelete old sensor data to save storage costs, and keep only the most recent month of data.
   • CEncrypt all sensor data and store it in a separate database to ensure security.
   • DArchive all historical data and never delete it, to preserve the original training set.
   Reveal Solution Discussion Select an answer first
4. 8foundation · easy

   Which practice is an example of protecting data security in an AI system?

   • AEncrypting data both at rest and in transit
   • BNormalizing feature values
   • CUsing a larger dataset for training
   • DIncreasing the number of training epochs
   Reveal Solution Discussion Select an answer first
5. 9expert · hard

   A social media company is developing a sentiment-analysis model to detect hate speech. They have a massive dataset of user posts, but the data is highly imbalanced: only 1% of posts are hate speech. The model trained on this data achieves high overall accuracy but fails to detect most hate speech. What is the most effective strategy to improve recall for the minority class?

   • AUse accuracy as the primary metric and accept the current model because it is correct 99% of the time.
   • BRemove the minority class examples to make the dataset balanced, and retrain the model.
   • CApply resampling techniques such as oversampling the minority class or undersampling the majority class, and use precision-recall metrics for evaluation.
   • DCollect more data from the majority class to make the dataset even larger, assuming more data always helps.
   Reveal Solution Discussion Select an answer first
6. 10expert · hard

   A social media company is building a content-moderation AI. The training data is collected from user reports, which are more frequent for certain types of content and languages. The model may be biased against minority languages. The company must also protect user privacy. What is the most effective approach?

   • ATrain the model on the existing data and apply post-processing to reduce bias.
   • BUse only the most common language to simplify the model.
   • CRemove all user-generated content and use only synthetic data to avoid privacy issues.
   • DAugment the training data with additional examples from minority languages and apply privacy-preserving techniques like differential privacy.
   Reveal Solution Discussion Select an answer first