Implement an Azure Document Intelligence in Foundry Tools Solution AI-102 Practice Questions (Page 7)

Questions 31–35

1. 31expert · hard

   A team trained a custom model on 500 labeled invoices and evaluated it on a held-out test set of 100 invoices. The overall accuracy is 92%, but the accuracy for a specific field (tax amount) is only 60%. The team needs to improve tax amount extraction before production. What is the best approach?

   • APublish the model and set a high confidence threshold for the tax amount field.
   • BCreate a composed model that includes the current model and a prebuilt invoice model.
   • CAdd more labeled invoices that contain tax amounts, especially those with varied formats, and retrain.
   • DUse the layout model to extract the tax amount and merge it with the custom model output.
   Reveal Solution Discussion Select an answer first
2. 32foundation · easy

   You are training a custom document intelligence model. What is the primary purpose of using labeled data during training?

   • ATo validate the model's accuracy after training
   • BTo generate the model's endpoint
   • CTo teach the model where the fields are located in the document
   • DTo reduce the number of training samples needed
   Reveal Solution Discussion Select an answer first
4. 33foundation · easy

   You have a PDF of a purchase order that includes vendor name, date, and line items. You want to extract these fields without training a custom model. Which approach should you use?

   • AUse the prebuilt ID document model
   • BUse the prebuilt read model
   • CUse the prebuilt receipt model
   • DUse the prebuilt invoice model
   Reveal Solution Discussion Select an answer first
5. 34foundation · easy

   You have trained a custom document intelligence model and are satisfied with its performance. What must you do before you can use it in a production application?

   • ARe-train the model with more data
   • BCreate a composed model
   • CPublish the model to get an endpoint
   • DCopy the model to another region
   Reveal Solution Discussion Select an answer first
6. 35application · medium

   A healthcare organization uses a custom Document Intelligence model to extract patient intake form fields. After training, they evaluate the model with a test set and find that the confidence scores for the 'insurance provider' field are consistently below 0.8. They have already labeled 50 forms. What should they do to improve the model's accuracy?

   • ARetrain the model with the same labeled data but more training iterations.
   • BAdd more labeled examples of forms where the insurance provider field appears in different locations or formats.
   • CIncrease the number of test documents to get a better evaluation.
   • DSwitch to the prebuilt identity document model.
   Reveal Solution Discussion Select an answer first