Clarification on Feedback for AI-900 Practice Questions

[dionysus]

Hi everyone,

I recently completed the Timed Mode Set 1 – AI-900 Microsoft Azure AI Fundamentals practice questions and received feedback that marked several of my answers as incorrect. However, after reviewing the official Azure documentation and reasoning through the questions, I believe my original answers align with Azure’s services and capabilities. I’d appreciate it if someone could help clarify these discrepancies or confirm if my reasoning is correct.

Questions and My Reasoning

<hr>

1. Azure AI Document Intelligence Use Cases
Question: In what two situations can the Azure AI Document Intelligence service be employed?
Options:

• Determine the seller’s identity from a sales receipt.
• Analyze and extract data from insurance claim forms.
• Locate an image of a product in a product catalog.
• Convert a form written in Filipino to English.
• Identify emotions and sentiments from customer feedback forms.
  My Answer: Determine the seller’s identity from a sales receipt and Analyze and extract data from insurance claim forms.
  Feedback: Analyze and extract data from insurance claim forms and Locate an image of a product in a product catalog.

My Clarification:
Based on Azure’s Documentation (https://learn.microsoft.com/en-us/training/modules/analyze-receipts-form-recognizer/3-azure-document-intelligence), Document Intelligence can extract key details like the name, address, and phone number of a merchant from a receipt. This clearly aligns with “Determine the seller’s identity.” On the other hand, locating images of products within a catalog seems more aligned with Azure Vision, as there’s no documentation suggesting that Document Intelligence is designed for image recognition tasks.

<hr>

2. Customer Feedback Analysis
Question: Which Azure Cognitive Services service should be used to extract sentiment analysis and key phrases from customer reviews?
Options:

• Azure AI Language
• Azure AI Document Intelligence
• Azure AI Face
• Custom Vision
  My Answer: Azure AI Language.
  Feedback: Azure AI Document Intelligence.

My Clarification:
Sentiment analysis and key phrase extraction are core features of Azure AI Language, specifically its Natural Language Processing (NLP) capabilities. Azure AI Document Intelligence focuses on extracting structured data from documents (e.g., key-value pairs, tables) and does not perform sentiment analysis or key phrase extraction. This seems like an error in feedback.

<hr>

3. Predicting Network Intrusion Attempts
Question: Predicting network intrusion attempts by analyzing abnormal network traffic patterns is an example of anomaly detection.
Options: Yes / No
My Answer: Yes.
Feedback: No.

My Clarification:
Detecting abnormal network traffic is a textbook example of anomaly detection, as it involves identifying patterns that deviate significantly from the norm. This is a common application of anomaly detection in cybersecurity for identifying intrusions. I’m unsure why this was marked incorrect.

<hr>

4. Object Detection in Custom Vision
Question: In creating an object detection model in the Custom Vision service, you must choose a specific domain from a set of predetermined options.
Options: Yes / No
My Answer: Yes.
Feedback: No.

My Clarification:
Azure Documentation (https://learn.microsoft.com/en-us/azure/ai-services/custom-vision-service/getting-started-build-a-classifier) states that when creating a Custom Vision model, you select a domain (e.g., General, Retail, Food). While domain selection is more emphasized in classification tasks, it’s still part of the object detection workflow. Could someone clarify if this requirement differs for object detection?

<hr>

5. Facial Recognition and Demographics
Question: Which Microsoft guiding principle for responsible AI does addressing demographic exclusion in facial recognition technology relate to?
Options:

• Fairness
• Privacy and Security
• Accountability
• Inclusiveness
  My Answer: Fairness.
  Feedback: Inclusiveness.

My Clarification:
Fairness specifically involves mitigating bias in AI systems and ensuring equitable outcomes across demographics. Concerns about excluding certain demographics in facial recognition seem directly tied to fairness. Inclusiveness, while related, generally focuses on accessibility for people with diverse abilities or backgrounds. Can someone explain why Inclusiveness is considered correct here?

<hr>

6. Overfitting in Model Evaluation
Question: Evaluating a model using the same data used for training can lead to overfitting.
Options: Yes / No
My Answer: Yes.
Feedback: No.

My Clarification:
Evaluating on training data is a classic cause of overfitting because the model might memorize patterns in the training data, leading to poor generalization to new data. While it’s true that this also gives an overly optimistic performance estimate, this is inherently tied to overfitting. Could this feedback be an error?

<hr>

Request for Assistance

I would greatly appreciate any insights or clarifications from the community or anyone who has experienced similar issues. If there are nuances I might have missed, I’d love to understand them better to ensure I’m fully prepared for the AI-900 certification.

Thank you for your time!

Best regards,