AI-900 Azure AI Fundamentals - Set 5 - Part 1

Correct Answer: Data Guard Rails

Text Analytics API is specifically designed to perform sentiment analysis and key phrase extraction from text at scale, making it ideal for analyzing customer reviews automatically.

Option a is incorrect because Custom Vision Service is designed for image classification and object detection tasks, not for analyzing text-based sentiment or extracting phrases.

Option c is incorrect because LUIS is primarily focused on understanding user intents and extracting entities from conversational text, not general sentiment analysis of product reviews.

Option d is incorrect because Face detection service analyzes facial features in images and cannot process text-based reviews or extract sentiment from written content.

Correct Answer: Form Recognizer with receipts

Form Recognizer has pre-built models specifically designed for receipts that can extract structured data like merchant names, amounts, and dates from receipt images efficiently.

Option a is incorrect because while OCR Read API can extract text from images, it does not understand the structure or semantics of receipts to extract specific fields.

Option c is incorrect because Computer Vision's describe image feature generates human-readable descriptions of images but does not extract structured data fields from documents like receipts.

Option d is incorrect because Text Analytics service processes text for sentiment and key phrases but cannot process images or extract structured data from visual documents.

Correct Answer: LUIS language understanding

LUIS is specifically designed to recognize user intents and extract entities from natural language, making it perfect for understanding what users want to accomplish in conversations.

Option a is incorrect because Q&A Maker is designed to create question-and-answer pairs from knowledge bases, not to understand complex user intents from varied natural language input.

Option b is incorrect because Text Analytics sentiment analysis determines emotional tone but does not identify specific user intentions like booking appointments or checking balances in conversations.

Option c is incorrect because Azure Bot Framework provides the infrastructure for building bots but does not include natural language understanding capabilities without integrating services like LUIS.

Correct Answer: Custom Vision object detection

Custom Vision provides a no-code interface specifically designed for image classification and object detection, allowing users without ML expertise to quickly train and deploy models.

Option a is incorrect because AutoML is powerful but requires more setup and understanding of datasets, whereas Custom Vision is specifically optimized for visual recognition tasks.

Option b is incorrect because Azure Machine Learning Designer requires understanding of ML pipelines and components, making it more complex than the visual, guided interface of Custom Vision.

Option d is incorrect because Jupyter notebooks with TensorFlow requires significant machine learning and programming expertise, which contradicts the requirement for limited ML knowledge.

Correct Answer: Named entity recognition

Named Entity Recognition specifically identifies and categorizes entities like person names, locations, and identification numbers in text, making it ideal for detecting PII in documents.

Option a is incorrect because sentiment analysis evaluates emotional tone and opinions in text but does not identify or extract specific entities like names and addresses.

Option b is incorrect because key phrase extraction identifies important topics and concepts but does not specifically categorize personal information entities that need to be redacted.

Option d is incorrect because language detection simply identifies which language text is written in and does not extract or categorize personal information entities from content.

Correct Answer: Sentiment and key phrases

Combining sentiment analysis to determine positive or negative feedback with key phrase extraction to identify specific topics like amenities provides the complete picture needed for analysis.

Option b is incorrect because language detection identifies languages and NER finds entities, but neither directly identifies topics or evaluates the emotional tone of feedback.

Option c is incorrect because OCR is for extracting text from images, which is not needed for analyzing text feedback, and sentiment alone does not identify topics.

Option d is incorrect because entity recognition alone identifies people, places, and organizations but does not evaluate sentiment or extract the topical themes from guest feedback.

Correct Answer: Custom Vision multiclass classification

Custom Vision with multiclass classification is designed to assign each image to exactly one category from predefined options, making it ideal for product categorization scenarios.

Option a is incorrect because Form Recognizer is designed for extracting structured data from documents and forms, not for categorizing general product images into categories.

Option c is incorrect because Computer Vision API tagging returns multiple generic tags for images but does not classify into custom, predefined business categories like electronics or clothing.

Option d is incorrect because Face API is specifically designed for analyzing facial features and attributes in images, not for categorizing different types of product images.

Correct Answer: Form Recognizer custom model

Form Recognizer custom models can be trained on specific form layouts to extract structured data from consistent document formats, including handwritten content and specific fields.

Option a is incorrect because while OCR extracts text, it does not understand form structure or field relationships needed to accurately extract specific data like applicant names.

Option b is incorrect because Custom Vision classification categorizes entire images into classes but does not extract structured field data or process form-based documents effectively.

Option d is incorrect because Computer Vision spatial analysis is designed for analyzing spatial relationships in scenes and videos, not for extracting structured data from form documents.

Correct Answer: Translator Service with NMT

Azure Translator Service uses neural machine translation to translate text across ninety plus languages with contextual understanding, making it ideal for multilingual content translation at scale.

Option a is incorrect because Text Analytics language detection only identifies which language text is written in but does not translate content between different languages.

Option b is incorrect because Speech Service translation focuses on real-time speech translation rather than comprehensive text translation across dozens of languages with full context preservation.

Option d is incorrect because LUIS extracts intents and entities from natural language but does not translate content between different languages or provide multilingual conversion capabilities.

Correct Answer: Face attributes with emotion

Face attributes with emotion returns emotion scores for detected faces, allowing analysis of emotional states over time when applied to video frames sequentially.

Option a is incorrect because face detection with landmarks identifies facial feature points like eyes and nose but does not analyze or classify emotional expressions from faces.

Option c is incorrect because face recognition verification compares faces to determine if they match the same person but does not analyze emotional expressions or states.

Option d is incorrect because face grouping clusters similar faces together based on appearance but does not analyze or extract emotional state information from facial expressions.

Correct Answer: Q&A Maker service

Q&A Maker is specifically designed to create question-and-answer knowledge bases from existing documents and FAQs, enabling automatic responses without extensive NLP training required.

Option a is incorrect because LUIS is designed for understanding intents and extracting entities from conversational input, not for creating document-based question-and-answer knowledge bases directly.

Option c is incorrect because Text Analytics API analyzes text for sentiment, key phrases, and entities but does not create interactive question-answering systems from documentation.

Option d is incorrect because Custom Vision model is designed for image classification and object detection tasks, not for processing text documents or answering questions.

Correct Answer: Speech Service with translation

Speech Service includes speech translation capabilities that can convert spoken language into both text and audio in multiple target languages simultaneously for real-time scenarios.

Option a is incorrect because Translator service handles text translation but does not process or translate live speech audio directly from speakers in real-time conference scenarios.

Option c is incorrect because Text Analytics analyzes written text for various insights but does not process spoken audio or provide real-time speech translation functionality.

Option d is incorrect because LUIS recognizes intents from natural language but does not translate speech between languages or provide multilingual audio output for conference scenarios.

Correct Answer: Pipelines for workflow

Pipelines in Azure ML Studio are specifically designed to create executable workflows that connect data preparation, training, and deployment steps in a sequential, reusable process.

Option a is incorrect because experiments are logical groupings of runs for organizing training attempts but do not define the sequential workflow of connected data processing steps.

Option b is incorrect because notebooks provide interactive coding environments but do not inherently create reusable, visual workflows that orchestrate multiple connected steps automatically in production.

Option d is incorrect because endpoints are used for serving deployed models to receive inference requests, not for organizing the training and preparation workflow steps themselves.

Correct Answer: Compute clusters for training

Compute clusters in Azure ML can scale from zero to maximum nodes automatically, making them ideal for training jobs that need cost-effective scaling when not actively running.

Option a is incorrect because compute instances are designed for notebook development and typically remain running continuously rather than scaling to zero automatically when inactive.

Option c is incorrect because inference clusters are specifically designed for deploying and serving models in production, not for training models with automatic cost-saving scaling capabilities.

Option d is incorrect because attached compute brings existing external resources like Databricks into Azure ML but does not provide the same automatic scaling to zero functionality.

Correct Answer: Automatic featurization process

Automatic featurization in AutoML handles feature engineering tasks including scaling, normalization, and dimensionality reduction to identify and optimize the most relevant features automatically.

Option a is incorrect because model selection tests different algorithms to find the best performing one but does not specifically handle feature optimization and selection processes.

Option c is incorrect because cross validation ensures model reliability by splitting data into training and validation sets but does not identify or select the most relevant features.

Option d is incorrect because primary metric configuration defines how model performance is measured but does not automatically identify or optimize which features should be used in training.

Correct Answer: Regression prediction task

Regression tasks in AutoML predict continuous numeric values, making them ideal for predicting specific dollar amounts like daily sales revenue based on various input features.

Option a is incorrect because binary classification predicts one of two categorical outcomes rather than continuous numeric values like specific revenue dollar amounts.

Option b is incorrect because multiclass classification assigns inputs to one of multiple predefined categories but does not predict continuous numeric values like revenue amounts.

Option d is incorrect because while time series forecasting predicts future values, regression is more appropriate when using diverse features like weather and promotions beyond just temporal patterns.

Correct Answer: Environments for dependencies

Environments in Azure ML define Python package versions and dependencies, ensuring consistent configurations across different team members, compute targets, and training runs for reproducibility.

Option a is incorrect because datastores provide connections to storage locations for data access but do not manage Python packages or software dependencies for training scripts.

Option b is incorrect because datasets manage data references and versioning but do not control Python package versions or script dependencies needed for model training consistency.

Option d is incorrect because experiments organize and group runs logically but do not define or enforce the software environment and package dependencies required for execution.

Correct Answer: Custom Vision project

Custom Vision is specifically optimized for image classification with small datasets and provides a no-code interface, making it perfect for quick deployment without deep ML expertise.

Option a is incorrect because while AutoML can handle image classification, it requires more setup and understanding compared to Custom Vision's specialized, user-friendly interface for vision tasks.

Option c is incorrect because ML Designer requires understanding of pipeline components and data flow, making it more complex than Custom Vision's guided approach for simple classification.

Option d is incorrect because using notebooks with PyTorch requires significant programming and deep learning expertise, contradicting the requirement for limited machine learning knowledge and quick results.

Correct Answer: Object detection project

Object detection projects in Custom Vision are specifically designed to detect multiple objects in images, draw bounding boxes around them, and label each detected object appropriately.

Option a is incorrect because multiclass classification assigns one label to the entire image rather than detecting and localizing multiple separate objects with individual bounding boxes.

Option b is incorrect because multilabel classification assigns multiple tags to an entire image but does not detect individual objects or draw bounding boxes around specific defects.

Option d is incorrect because image tagging is a general Computer Vision feature that returns tags but does not train custom models or draw bounding boxes around detected objects.

Correct Answer: Misses many actual products often

Low recall means the model fails to identify many true positive cases, so it misses numerous actual products that should be detected even though identified ones are correct.

Option a is incorrect because high precision means predictions are accurate when made, but low recall indicates many correct instances are missed, not that most are identified.

Option c is incorrect because high precision specifically indicates few false positives, meaning the model does not predict too many incorrect matches when it makes predictions.

Option d is incorrect because the combination of high precision and low recall represents an imbalanced performance where accuracy exists but coverage is poor, not balanced overall.