Microsoft AI-102 (Designing and Implementing a Microsoft Azure AI Solution) Practice Test 57

Question 560

An onboarding system needs to populate name and date of birth from supported identity documents. What field-processing approach fits?

  1. Use the Invoice model's structured invoice fields and their confidence data
  2. Preserve page identifiers and text spans or polygon coordinates
  3. Use the Receipt model and map its receipt-specific output schema
  4. Use the prebuilt ID Document model and validate its typed identity fields

Explanation

Correct Answer

D:

The model extracts relevant fields from supported IDs; its output is data extraction, not independent proof of identity. In the shared-services expense document processor, confirm the fix by retesting the affected behavior after using the prebuilt ID Document model and validate its typed identity fields.

Incorrect Answers

A:

That change addresses another failure mode: prebuilt invoice extraction identifies domain-specific fields suitable for accounting workflows; it does not address this case, for which the relevant intervention is to use the prebuilt ID Document model and validate its typed identity fields.

B:

Read output includes location information on supported inputs, enabling verification against the source page in its own workflow, but that is a different correction from the present requirement to use the prebuilt ID Document model and validate its typed identity fields. The reported issue would remain.

C:

Although useful elsewhere in the shared-services expense document processor, point-of-sale receipts have a different document structure from supplier invoices; the present case is different and calls for the team to use the prebuilt ID Document model and validate its typed identity fields.

Question 561

Two annotators label the same date field differently, producing unstable predictions. What process should be added?

  1. Evaluate a custom neural extraction approach for variable supported layouts
  2. Rebalance and diversify validation examples and inspect per-class confusion
  3. Document annotation conventions and review inter-annotator consistency
  4. Evaluate routing classification and include representative examples of the new family

Explanation

Correct Answer

C:

Inconsistent label boundaries and semantics introduce training noise independent of model architecture. In the specialized insurance forms intake, confirm the fix by retesting the affected behavior after documenting annotation conventions and review inter-annotator consistency.

Incorrect Answers

A:

Template models favor predictable positions, whereas varied layouts need more flexible modeling in its own workflow, but that is a different correction from the present requirement to document annotation conventions and review inter-annotator consistency. The reported issue would remain.

B:

Although useful elsewhere in the specialized insurance forms intake, unrepresentative class frequencies can hide systematic routing mistakes; the present case is different and calls for the team to document annotation conventions and review inter-annotator consistency.

D:

That change addresses another failure mode: composition depends on accurate document-type selection; adding more extractors alone does not fix misrouting; it does not address this case, for which the relevant intervention is to document annotation conventions and review inter-annotator consistency.

Question 562

A neural extractor performs well on sample template A but fails on rare broker template F. What should be added?

  1. Rebalance and diversify validation examples and inspect per-class confusion
  2. Document annotation conventions and review inter-annotator consistency
  3. Evaluate routing classification and include representative examples of the new family
  4. Include correctly labeled examples of template F in training and independent evaluation

Explanation

Correct Answer

D:

Flexibility still depends on representative examples; a missing layout may behave as out-of-distribution input. In the specialized insurance forms intake, confirm the fix by retesting the affected behavior after including correctly labeled examples of template F in training and independent evaluation.

Incorrect Answers

A:

Unrepresentative class frequencies can hide systematic routing mistakes in its own workflow, but that is a different correction from the present requirement to include correctly labeled examples of template F in training and independent evaluation. The reported issue would remain.

B:

That change addresses another failure mode: inconsistent label boundaries and semantics introduce training noise independent of model architecture; it does not address this case, for which the relevant intervention is to include correctly labeled examples of template F in training and independent evaluation.

C:

Although useful elsewhere in the specialized insurance forms intake, composition depends on accurate document-type selection; adding more extractors alone does not fix misrouting; the present case is different and calls for the team to include correctly labeled examples of template F in training and independent evaluation.

Question 563

Hundreds of purchase forms have identical boxes, but an extractor struggles with a small field shift. What training data would help?

  1. Compose the appropriate compatible custom extractors with document routing
  2. Separate related source documents between model training and validation partitions
  3. Label representative template variants, including expected layout shifts
  4. Assign the true document-type label to each training example independent of filename conventions

Explanation

Correct Answer

C:

A template model learns a relatively stable structure; appropriate variation improves robustness within that document family. In the specialized insurance forms intake, confirm the fix by retesting the affected behavior after labeling representative template variants, including expected layout shifts.

Incorrect Answers

A:

Although useful elsewhere in the specialized insurance forms intake, a composed model selects an extraction path for supported form types rather than requiring callers to pick each extractor manually; the present case is different and calls for the team to label representative template variants, including expected layout shifts.

B:

That change addresses another failure mode: train-test leakage from shared document pages can exaggerate extraction accuracy even when the annotation fields are otherwise correct; it does not address this case, for which the relevant intervention is to label representative template variants, including expected layout shifts.

D:

A classifier learns from supervised labels; incorrect shortcuts such as filename prefixes create mislabeled categories and unreliable routing in its own workflow, but that is a different correction from the present requirement to label representative template variants, including expected layout shifts. The reported issue would remain.

Question 564

A template model extracts a monetary field but confuses its label with a nearby amount. What annotation issue should be examined?

  1. Verify field labels and bounding selections on the affected training documents
  2. Define and label the new field in suitable training examples before retraining
  3. Use evaluated confidence thresholds and a manual or fallback route for uncertain classifications
  4. Verify composition support and limits for the specific model types and API version

Explanation

Correct Answer

A:

Inconsistent supervised field positions can teach the extractor to pick the wrong neighboring content. In the specialized insurance forms intake, confirm the fix by retesting the affected behavior after verifying field labels and bounding selections on the affected training documents.

Incorrect Answers

B:

Although useful elsewhere in the specialized insurance forms intake, supervised extraction does not reliably learn an output target that no example defines; the present case is different and calls for the team to verify field labels and bounding selections on the affected training documents.

C:

Wrong class routing can corrupt field extraction; uncertainty should be managed explicitly in its own workflow, but that is a different correction from the present requirement to verify field labels and bounding selections on the affected training documents. The reported issue would remain.

D:

That change addresses another failure mode: document Intelligence composition capabilities depend on supported model families and service versions; it does not address this case, for which the relevant intervention is to verify field labels and bounding selections on the affected training documents.

Question 565

A team chooses a template model for loosely formatted handwritten legal letters. What mismatch should be reviewed?

  1. Document annotation conventions and review inter-annotator consistency
  2. Include correctly labeled examples of template F in training and independent evaluation
  3. Evaluate routing classification and include representative examples of the new family
  4. Evaluate a custom neural extraction approach for variable supported layouts

Explanation

Correct Answer

D:

Template models favor predictable positions, whereas varied layouts need more flexible modeling. In the specialized insurance forms intake, confirm the fix by retesting the affected behavior after evaluating a custom neural extraction approach for variable supported layouts.

Incorrect Answers

A:

Inconsistent label boundaries and semantics introduce training noise independent of model architecture in its own workflow, but that is a different correction from the present requirement to evaluate a custom neural extraction approach for variable supported layouts. The reported issue would remain.

B:

Although useful elsewhere in the specialized insurance forms intake, flexibility still depends on representative examples; a missing layout may behave as out-of-distribution input; the present case is different and calls for the team to evaluate a custom neural extraction approach for variable supported layouts.

C:

That change addresses another failure mode: composition depends on accurate document-type selection; adding more extractors alone does not fix misrouting; it does not address this case, for which the relevant intervention is to evaluate a custom neural extraction approach for variable supported layouts.

Question 566

A team responsible for a specialized insurance forms intake needs to handle documents whose field positions and layouts vary substantially; the acceptance check is that fields are extracted across several different document arrangements. Which capability should be selected?

  1. Compose multiple custom models
  2. Train a custom template extraction model
  3. Label custom extraction training samples
  4. Train a custom neural extraction model

Explanation

Correct Answer

D:

Custom neural extraction models learn more flexible patterns for varied supported document layouts and labeled data. For this specialized insurance forms intake, the intended evidence is that fields are extracted across several different document arrangements. They still require representative training examples and review of extraction accuracy.

Incorrect Answers

A:

A composed model combines document classification with selected custom extractors so the appropriate model processes each type; it may be used to route mixed form types through a single composed document-analysis entry point in a specialized insurance forms intake, but the current requirement is instead to handle documents whose field positions and layouts vary substantially. That separate capability does not prove that fields are extracted across several different document arrangements; One standalone extractor cannot be assumed to handle all unrelated form families reliably, so selecting that alternative would not meet the stated acceptance check.

B:

Custom template models learn field extraction for documents with relatively predictable visual structure; it may be used to capture stable field positions from repeated forms that share a consistent layout in a specialized insurance forms intake, but the current requirement is instead to handle documents whose field positions and layouts vary substantially. That separate capability does not prove that fields are extracted across several different document arrangements; They are not the preferred generic answer to many widely varying free-form layouts, so selecting that alternative would not meet the stated acceptance check.

C:

Field labels map sample content to the target extraction schema, enabling supervised custom model training; it may be used to annotate representative documents with the exact field values the model should learn in a specialized insurance forms intake, but the current requirement is instead to handle documents whose field positions and layouts vary substantially. That separate capability does not prove that fields are extracted across several different document arrangements; Supplying a schema name without accurately labeled examples is insufficient to teach the needed fields, so selecting that alternative would not meet the stated acceptance check.

Question 567

A document-labeling team uses the same scanned page in training and in a supposedly independent validation set. Which data control should it enforce?

  1. Label representative template variants, including expected layout shifts
  2. Separate related source documents between model training and validation partitions
  3. Compose the appropriate compatible custom extractors with document routing
  4. Assign the true document-type label to each training example independent of filename conventions

Explanation

Correct Answer

B:

Train-test leakage from shared document pages can exaggerate extraction accuracy even when the annotation fields are otherwise correct. In the specialized insurance forms intake, confirm the fix by retesting the affected behavior after separating related source documents between model training and validation partitions.

Incorrect Answers

A:

A template model learns a relatively stable structure; appropriate variation improves robustness within that document family in its own workflow, but that is a different correction from the present requirement to separate related source documents between model training and validation partitions. The reported issue would remain.

C:

That change addresses another failure mode: a composed model selects an extraction path for supported form types rather than requiring callers to pick each extractor manually; it does not address this case, for which the relevant intervention is to separate related source documents between model training and validation partitions.

D:

Although useful elsewhere in the specialized insurance forms intake, a classifier learns from supervised labels; incorrect shortcuts such as filename prefixes create mislabeled categories and unreliable routing; the present case is different and calls for the team to separate related source documents between model training and validation partitions.

Question 568

A release for a specialized insurance forms intake is blocked until the application can identify which document category a submitted page or file belongs to before extraction. The tester will verify that the system returns the predicted form type. What should be implemented?

  1. Compose multiple custom models
  2. Train a custom document classifier
  3. Train a custom neural extraction model
  4. Train a custom template extraction model

Explanation

Correct Answer

B:

A classifier predicts document types from trained categories and can support routing to appropriate extractors. For this specialized insurance forms intake, the intended evidence is that the system returns the predicted form type. Classification by itself does not produce every custom form field value.

Incorrect Answers

A:

A composed model combines document classification with selected custom extractors so the appropriate model processes each type; it may be used to route mixed form types through a single composed document-analysis entry point in a specialized insurance forms intake, but the current requirement is instead to identify which document category a submitted page or file belongs to before extraction. That separate capability does not prove that the system returns the predicted form type; One standalone extractor cannot be assumed to handle all unrelated form families reliably, so selecting that alternative would not meet the stated acceptance check.

C:

Custom neural extraction models learn more flexible patterns for varied supported document layouts and labeled data; it may be used to handle documents whose field positions and layouts vary substantially in a specialized insurance forms intake, but the current requirement is instead to identify which document category a submitted page or file belongs to before extraction. That separate capability does not prove that the system returns the predicted form type; They still require representative training examples and review of extraction accuracy, so selecting that alternative would not meet the stated acceptance check.

D:

Custom template models learn field extraction for documents with relatively predictable visual structure; it may be used to capture stable field positions from repeated forms that share a consistent layout in a specialized insurance forms intake, but the current requirement is instead to identify which document category a submitted page or file belongs to before extraction. That separate capability does not prove that the system returns the predicted form type; They are not the preferred generic answer to many widely varying free-form layouts, so selecting that alternative would not meet the stated acceptance check.

Question 569

A classifier reports many false invoices because its validation set contains almost no claims. What should be fixed?

  1. Evaluate a custom neural extraction approach for variable supported layouts
  2. Include correctly labeled examples of template F in training and independent evaluation
  3. Document annotation conventions and review inter-annotator consistency
  4. Rebalance and diversify validation examples and inspect per-class confusion

Explanation

Correct Answer

D:

Unrepresentative class frequencies can hide systematic routing mistakes. In the specialized insurance forms intake, confirm the fix by retesting the affected behavior after rebalancing and diversify validation examples and inspect per-class confusion.

Incorrect Answers

A:

Although useful elsewhere in the specialized insurance forms intake, template models favor predictable positions, whereas varied layouts need more flexible modeling; the present case is different and calls for the team to rebalance and diversify validation examples and inspect per-class confusion.

B:

That change addresses another failure mode: flexibility still depends on representative examples; a missing layout may behave as out-of-distribution input; it does not address this case, for which the relevant intervention is to rebalance and diversify validation examples and inspect per-class confusion.

C:

Inconsistent label boundaries and semantics introduce training noise independent of model architecture in its own workflow, but that is a different correction from the present requirement to rebalance and diversify validation examples and inspect per-class confusion. The reported issue would remain.

Question 570

The application trusts a low-confidence class prediction to choose a sensitive downstream extractor. What guard is suitable?

  1. Verify field labels and bounding selections on the affected training documents
  2. Define and label the new field in suitable training examples before retraining
  3. Update the labeling schema and corresponding annotated examples before training the next model
  4. Use evaluated confidence thresholds and a manual or fallback route for uncertain classifications

Explanation

Correct Answer

D:

Wrong class routing can corrupt field extraction; uncertainty should be managed explicitly. In the specialized insurance forms intake, confirm the fix by retesting the affected behavior after using evaluated confidence thresholds and a manual or fallback route for uncertain classifications.

Incorrect Answers

A:

Although useful elsewhere in the specialized insurance forms intake, inconsistent supervised field positions can teach the extractor to pick the wrong neighboring content; the present case is different and calls for the team to use evaluated confidence thresholds and a manual or fallback route for uncertain classifications.

B:

That change addresses another failure mode: supervised extraction does not reliably learn an output target that no example defines; it does not address this case, for which the relevant intervention is to use evaluated confidence thresholds and a manual or fallback route for uncertain classifications.

C:

Model output names and labels must be aligned; changing application schema alone does not retrain field mappings in its own workflow, but that is a different correction from the present requirement to use evaluated confidence thresholds and a manual or fallback route for uncertain classifications. The reported issue would remain.

Designing and Implementing a Microsoft Azure AI Solution (AI-102) Practice Tests