
AI-102 EXAM QUESTIONS 2024 MICROSOFT AZURE AI 102 CERTIFICATION COURSE PART-9
AI Summary
This summary covers key insights, decisions, and conclusions from the transcript, focusing on Azure AI services for various tasks.
The discussion begins with question 71, where a company wants to reduce the time employees spend logging receipts in expense reports. All receipts are in English, and the goal is to extract top-level information like vendor and transaction totals while minimizing development effort. This implies using pre-built Azure services.
Custom Vision is deemed unsuitable because it requires custom models, increasing development effort, and is primarily for object detection and image classification, not specific receipt information extraction.
Personalizer is also unsuitable. While an Azure service, it's designed to deliver personalized content or experiences to users, similar to recommendations on YouTube, Netflix, or Amazon, not for extracting data from receipts.
Form Recognizer (now Azure AI Document Intelligence) is identified as the ideal solution. It’s a pre-built Azure service specifically designed to extract information from documents, including a pre-built receipt model. This minimizes development effort and can extract vendor details and transaction totals, making it perfect for the requirement.
Computer Vision can extract text (OCR) but is not as specialized as Form Recognizer for structured data from forms and receipts. It would also require more development effort compared to using pre-built receipt models. Therefore, Form Recognizer is the best answer.
The next question involves analyzing 5,000 scanned invoice images to extract invoice items, sales items, and customer details.
Custom Vision is again ruled out as it's primarily for image classification and object detection, not for extracting structured data like invoice details.
Azure AI Computer Vision can extract text and analyze images but isn't specifically designed for detailed, structured information like invoice items or customer details.
Immersive Reader is for assisting individuals with learning difficulties (e.g., dyslexia) by improving reading and comprehension, not for extracting structured data from documents.
Azure AI Document Intelligence (formerly Form Recognizer) is confirmed as the correct choice because it is specifically designed for extracting structured data from documents, including invoices. The speaker notes that Azure frequently renames its services (e.g., Cognitive Search to Azure AI Search, Form Recognizer to Azure AI Document Intelligence), emphasizing the need for familiarity with current names.
A case study question, based on a previously discussed scenario, concerns creating a chatbot to answer FAQs. The chatbot needs to provide answers only if the confidence score is 70% or above; otherwise, it should offer other options. Q&A Maker and an Azure Bot Framework SDK chatbot have already been created. The question asks which property should be used to integrate components to meet these requirements.
QnA MakerOptions.StrictFilter is incorrect because it applies strict filtering but doesn't control the confidence threshold for providing alternative options.
QnAMakerDialogResponse.NoText is also incorrect. It sets the message for when no match is found, not for handling matches with low confidence scores.
QnAMakerOptions.RankType ranks answers but doesn't provide control over the confidence threshold.
QnAMakerOptions.ScoreThreshold is the correct answer. This property directly controls the minimum confidence score required for an answer, allowing the chatbot to provide other options if the score falls below 70%.
The next scenario involves a food product factory needing a monitoring solution for staff compliance with Personal Protective Equipment (PPE), specifically identifying staff who have removed masks or safety glasses. The solution must perform checks every 15 minutes, minimize development effort, and minimize cost.
Face API is identified as the perfect service for this. It excels in scenarios involving face attributes like masks, glasses, and expressions. It minimizes development effort as a pre-built service and requires no machine learning expertise.
Computer Vision and Azure Video Analyzer for Media are mentioned but not suitable for this specific face-related detection.
Face API can detect faces, find similar faces, perform one-to-one verification (e.g., matching two photos of the same person) and one-to-many identification (e.g., searching for a criminal in a crowd), group faces by similarity (like in photo apps), and return face attributes such as masks, glasses, headwear, and head position. It can also estimate age, which can be used for targeted advertising.
Another question deals with 100 chatbots, each with its own language understanding (LUIS) model, requiring frequent programmatic updates to add the same phrases to each model. The question asks how to complete a blank in the code.
AddPhraseListAsync is the correct method. It's used to programmatically add phrase list features to a LUIS model, aligning with the requirement to update models with new phrases.
PhraseList is not a method but a concept/object. PhraseListCreateObject is a constructor or object type, not a method for adding phrase lists. Phrases refers to the actual phrases to be added, a property within the phrase list object, not a method. SavePhraseListAsync and UploadPhraseListAsync are not standard methods for this purpose.
For the second blank, New PhraseListCreateObject is the correct constructor to create an object that will contain the list of phrases.
The next task involves extracting text from press releases stored as PDFs. Azure AI Document Intelligence is the clear choice for extracting text from files, especially structured documents like PDFs. Azure AI Search is for creating search experiences, and AI Vision is for analyzing visual content, not specifically for text extraction from PDFs in this context.
Following text extraction, the next step is to perform sentiment analysis on the extracted text.
Azure Cognitive Search (now Azure AI Search) is for search capabilities. Computer Vision is for visual content analysis. Document Intelligence extracts structured data. Azure AI Language Service is the correct choice, as it is specifically designed for natural language processing tasks like sentiment analysis.
A crucial question involves an app (F1) using Azure AI Document Intelligence to analyze medical records and recommend pharmaceutical doses. A sample response from Document Intelligence is provided, and the task is to identify correct statements about it.
The response shows `status: succeeded`, `lastUpdatedDateTime` (indicating quick analysis), API version, and `modelId: prebuilt-health.insurance.card`. The `content` field shows "blood pressure 118 by 72". The `pages` array details detected words (e.g., "blood", "pressure", "118", "72") with their polygons (bounding boxes), confidence scores (e.g., 0.99 for "blood"), and spans. The `documents` array shows the document type identified as `health.insurance.card` with a confidence of 1.0, but `fields` is empty.
Statement 1: "The chosen model is suitable for the intended use case." This is incorrect. The intended use case is analyzing medical records and recommending pharmaceutical doses. The `prebuilt-health.insurance.card` model extracts information from health insurance cards (e.g., ID numbers, member details), not medical records or dose recommendations. A custom model would likely be needed.
Statement 2: "The text content was recognized with greater than 70% confidence." This is correct. The `pages` array shows individual words ("blood", "pressure", "118", "72") each recognized with confidence scores of 0.99 or 1.0, all above 70%.
Statement 3: "The form elements were recognized with greater than 70% confidence." This is incorrect. The `documents` array's `fields` object is empty, indicating no form elements were recognized, even though the model identified the document as a health insurance card with 100% confidence.
Therefore, only statement 2 is correct.
The final question concerns designing a content management system to optimize the reading experience for users with reduced comprehension and learning differences like dyslexia, while minimizing development effort.
Azure AI Immersive Reader is the correct service. It enhances reading and learning by improving text clarity (e.g., spacing between characters/words), displaying pictures for common words, highlighting parts of speech, converting text to speech, translating content in real-time, and splitting words into syllables for pronunciation guidance.
Translator is for language translation. Document Intelligence is for extracting structured information. AI Language is for natural language processing tasks. Immersive Reader is specifically designed for the described accessibility needs.