Implement an Azure Document Intelligence in Foundry Tools Solution AI-102 Practice Questions (Page 4)

Questions 16–20

1. 16application · medium

   A team is building a custom model for a new type of tax form. They have 150 scanned forms. They labeled 100 forms for training and kept 50 for testing. After training, the model performs poorly on the test set. What is the most likely cause?

   • AThe model needs to be composed with another model.
   • BThe training set is too small or not representative of the variety in the forms.
   • CThe test set should have been used for training as well.
   • DThe forms should be converted to a different file format before training.
   Reveal Solution Discussion Select an answer first
2. 17application · medium

   A company has a composed model that includes two custom models. They want to update one of the custom models with new training data. What is the correct process?

   • ARetrain the individual custom model, then create a new composed model with the updated model.
   • BDelete the composed model and retrain both custom models from scratch.
   • CAdd the new training data directly to the composed model.
   • DUpdate the composed model's configuration to point to the retrained model.
   Reveal Solution Discussion Select an answer first
4. 18expert · hard

   A custom model was trained on 200 labeled documents. The team evaluated it on a test set of 50 documents and got 90% accuracy. They want to publish it, but a stakeholder asks for evidence that the model will work on the full population of documents. What is the best way to provide this evidence?

   • AEvaluate the model on a larger, representative test set that reflects the population's variety.
   • BUse the confidence scores from the test set as evidence.
   • CPublish the model and monitor its performance in production for a few weeks.
   • DReport the training accuracy as the expected performance.
   Reveal Solution Discussion Select an answer first
5. 19expert · hard

   A company has three custom models: Model A for form type A, Model B for form type B, and Model C for form type C. They want to use a composed model. They test the composed model and find that a document of type A is being routed to Model B, resulting in poor extraction. What is the most likely cause?

   • AModel B has a higher confidence score for the document than Model A.
   • BThe document was not labeled correctly during training.
   • CModel A was not published before composing.
   • DThe composed model is not supported for this scenario.
   Reveal Solution Discussion Select an answer first
6. 20application · medium

   A team has trained a custom Document Intelligence model and wants to evaluate its performance before production deployment. They have a set of labeled documents that were not used during training. What should they do to evaluate the model?

   • ACompare the model's output to the training set labels.
   • BUse the model's built-in evaluation on the held-out test set and review confidence scores and field-level accuracy.
   • CDeploy the model to production and monitor real-time error rates.
   • DRe-train the model with the test set included to improve accuracy.
   Reveal Solution Discussion Select an answer first