Gather Required Data PMI-CPMAI Practice Questions (Page 2)

Questions 6–10

1. 6application · medium

   A government agency is migrating citizen survey data from an on-premises database to Azure for AI-based sentiment analysis. The data contains personally identifiable information (PII) and must be anonymized before it is used for model training. The team needs to ensure that the anonymization process is applied during the data transfer. What should they do?

   • AUse AzCopy to copy the data to Azure Blob Storage, then run a separate anonymization script on the data.
   • BManually anonymize the data on the on-premises server before uploading it to Azure.
   • CUse Azure Data Factory with a data flow that applies anonymization transformations (e.g., hashing PII fields) before writing to Azure Data Lake Storage.
   • DUse Azure Database Migration Service to migrate the data, which automatically anonymizes PII.
   Reveal Solution Discussion Select an answer first
2. 7application · medium

   A weather forecasting company collects atmospheric data from sensors every minute. The data is used to train a model that predicts severe weather events. The model must be updated with the latest data as soon as possible, but the retraining process takes several hours. What is the best approach to ensure the model uses the most current data?

   • ARetrain the model only once a week to reduce resource usage.
   • BUse a manual process to retrain the model when the team has time.
   • CRetrain the model after every new data point is collected.
   • DSet up a continuous data ingestion pipeline that updates the training dataset in real-time, and schedule model retraining every few hours.
   Reveal Solution Discussion Select an answer first
4. 8expert · hard

   A multinational corporation is consolidating customer data from regional databases in the US, EU, and Asia into a single Azure data lake for a global AI initiative. The data must comply with regional data residency laws, meaning EU customer data must remain in an EU region. The team needs to extract data from each region and ensure that the data is correctly partitioned and validated. What is the most appropriate architecture?

   • AUse a single Data Factory pipeline that extracts data from all regions and stores it in a data lake in the EU region.
   • BUse Azure Database Migration Service to migrate all regional databases to a single Azure SQL Database in the US, then export to a data lake.
   • CUse separate Azure Data Factory pipelines for each region, each extracting data to a data lake in the corresponding region, and then use a global pipeline to copy data to a central location while maintaining region tags.
   • DExtract all data to a single data lake in the US region and rely on Azure's compliance certifications to meet data residency requirements.
   Reveal Solution Discussion Select an answer first
5. 9application · medium

   A news website wants to train a model to recommend articles based on user reading history. The data is updated continuously as users read articles. The team needs to ensure the model is retrained with fresh data regularly, but the retraining process is resource-intensive and takes several hours. What is the best approach to balance data freshness and resource usage?

   • ARetrain the model only when a user requests a recommendation, using the latest data at that time.
   • BRetrain the model every hour to ensure the most up-to-date recommendations.
   • CUse a manual process to retrain the model whenever the team remembers to do it.
   • DSet up a weekly retraining schedule that uses the latest data, and implement incremental data refreshes daily to keep the training dataset current.
   Reveal Solution Discussion Select an answer first
6. 10application · medium

   An e-commerce company has a daily batch process that refreshes a product recommendation dataset. The process runs at midnight and takes about two hours to complete. Recently, the data team noticed that the recommendations are based on stale data because the refresh sometimes fails and is not retried. What should the team implement to ensure data currency and reliability?

   • AChange the batch process to run every hour instead of daily to reduce the chance of stale data.
   • BConfigure the batch process with retry policies and alerting so that failures are automatically retried and the team is notified if the refresh fails after multiple attempts.
   • CUse a manual process to run the refresh when the automated process fails.
   • DMove the batch process to run on a different schedule, such as every Monday, to avoid conflicts with other jobs.
   Reveal Solution Discussion Select an answer first