AI-900 Azure AI Fundamentals - Set 4 - Part 1

Correct Answer: Automated ML

Option d is correct because Content Moderator is part of Decision Services and is specifically designed to detect inappropriate or offensive content in images, text, and videos for moderation purposes.

Option a is incorrect because Language Services focus on processing text-based content like translation and sentiment analysis, not visual content moderation or detecting inappropriate images.

Option b is incorrect because Vision Services analyze and extract information from images but do not specifically focus on detecting inappropriate or harmful content for moderation purposes.

Option c is incorrect because Speech Services handle audio processing tasks like speech recognition and synthesis, not visual content moderation or detecting inappropriate images in user-generated content.

Correct Answer: AutoML

Option a is correct because AutoML completely automates the process of building and training multiple machine learning models, allowing users to compare algorithms and select the best one without writing code.

Option b is incorrect because Designer is a visual drag-and-drop tool for building machine learning pipelines that still requires understanding of workflow construction rather than automated model selection and training.

Option c is incorrect because Notebooks provide a Jupyter environment for writing Python code to build models manually, which contradicts the requirement of not writing code for model creation.

Option d is incorrect because Pipelines are machine learning workflows that organize steps in a process but do not automatically train and compare multiple models without manual configuration.

Correct Answer: Deploy Custom Vision Service

Option c is correct because Custom Vision Service is specifically designed for creating custom image classification models by uploading your own labeled images to teach the model your unique product categories.

Option a is incorrect because Computer Vision API provides general-purpose pre-built vision capabilities but does not allow you to create custom models tailored to your specific product categories and styles.

Option b is incorrect because Face Detection is designed specifically for identifying and analyzing human faces in images, not for classifying retail products into categories like clothing items.

Option d is incorrect because Form Recognizer is designed for extracting text and structure from documents and forms, not for classifying product images into retail categories.

Correct Answer: Form Recognizer

Option d is correct because Form Recognizer, also known as Document Intelligence, is specifically designed to extract text and structure from complex documents including forms with tables and mixed content types.

Option a is incorrect because Text Analytics focuses on analyzing text content for sentiment, key phrases, and entities but does not extract text from document images or handle complex form layouts.

Option b is incorrect because Translator is designed for converting text from one language to another and does not extract information from document images or process form structures.

Option c is incorrect because Computer Vision can extract basic text using OCR but does not understand form structure, tables, or the relationships between fields like Form Recognizer does.

Correct Answer: Q and A Maker

Option a is correct because Q and A Maker is specifically designed to create question-answering services from existing content, understanding user intent and matching questions to appropriate answers.

Option b is incorrect because Content Moderator is used for detecting inappropriate content in text, images, and videos rather than understanding questions and providing answers from a knowledge base.

Option c is incorrect because Personalizer is a Decision Service that provides personalized recommendations based on user behavior, not a question-answering system for FAQ content.

Option d is incorrect because Computer Vision analyzes images and extracts visual information rather than understanding text-based questions and providing answers from a knowledge base.

Correct Answer: Speech Translation

Option b is correct because Speech Translation combines speech recognition and translation capabilities to convert spoken audio from one language to another in real-time for live interpretation scenarios.

Option a is incorrect because Text to Speech only converts written text into spoken audio and does not handle incoming speech recognition or translation from one language to another.

Option c is incorrect because Speaker Recognition is designed to identify who is speaking based on voice characteristics rather than translating speech content from one language to another.

Option d is incorrect because Speech to Text only converts spoken words into written text without performing any language translation, which is essential for the international conference scenario.

Correct Answer: Designer

Option c is correct because Designer provides a visual drag-and-drop interface specifically for constructing end-to-end machine learning pipelines with pre-built components for all workflow stages.

Option a is incorrect because AutoML automates model training and selection but does not provide a visual interface for designing custom pipelines with data transformation and feature engineering steps.

Option b is incorrect because Notebooks provide a code-based Jupyter environment for writing Python scripts rather than a visual drag-and-drop interface for pipeline construction.

Option d is incorrect because Experiments are logical groupings of training runs that track execution history rather than a tool for visually designing and constructing machine learning workflows.

Correct Answer: Model Registry

Option a is correct because Model Registry allows creation, management, and tracking of registered models as incremental versions under the same name with metadata tags for comparison.

Option b is incorrect because Pipelines are executable workflows of machine learning tasks rather than a storage system for tracking and versioning trained models over time.

Option c is incorrect because Datasets store and manage training data with versioning capabilities but do not track or version trained machine learning models and their performance metrics.

Option d is incorrect because Compute Instances are development workstations for data scientists to work with data and models, not a registry for versioning and tracking model performance.

Correct Answer: Time Series Forecasting

Option d is correct because Time Series Forecasting is specifically designed for predicting future values based on historical time-based data, handling seasonality and multiple contextual variables automatically.

Option a is incorrect because Classification categorizes data into predefined classes or labels rather than predicting continuous numerical values like revenue over time based on historical patterns.

Option b is incorrect because standard Regression predicts continuous values but does not specifically handle time-based patterns, seasonality, or the temporal relationships essential for revenue forecasting.

Option c is incorrect because Anomaly Detection identifies unusual patterns or outliers in data rather than forecasting future values based on historical time-series patterns and trends.

Correct Answer: Anomaly Detector

Option b is correct because Anomaly Detector is specifically designed to identify unusual patterns and outliers in data streams, making it ideal for detecting abnormal transaction behaviors.

Option a is incorrect because Content Moderator is designed for detecting inappropriate content in text, images, and videos rather than identifying unusual patterns in numerical transaction data.

Option c is incorrect because Personalizer provides personalized recommendations based on user context and behavior rather than detecting anomalous or unusual patterns that indicate potential fraud.

Option d is incorrect because Custom Vision is used for image classification and object detection tasks rather than analyzing numerical transaction data to identify behavioral anomalies.

Correct Answer: Text Analytics

Option d is correct because Text Analytics is specifically designed to analyze text content for sentiment, extract key phrases, identify entities, and detect topics across multiple languages.

Option a is incorrect because Q and A Maker is designed for creating question-answering services from knowledge bases rather than performing sentiment analysis on customer reviews and feedback.

Option b is incorrect because Translator converts text from one language to another but does not analyze sentiment, extract topics, or determine whether content is positive or negative.

Option c is incorrect because LUIS (Language Understanding) is designed for understanding user intents in conversational applications rather than performing sentiment analysis on large volumes of product reviews.

Correct Answer: Model Validation

Option d is correct because Model Validation is the process of comparing training data results to test data results to ensure the model generalizes well to new data.

Option a is incorrect because Data Labeling is the process of applying tags or labels to training data for supervised learning rather than evaluating how well a trained model performs.

Option b is incorrect because Feature Engineering involves creating and transforming input features to improve model performance rather than testing the model's performance on unseen validation data after training completes.

Option c is incorrect because Automatic Featurization applies scaling and normalization techniques during model training rather than evaluating model performance on validation data after training completes.

Correct Answer: Speech to Text

Option b is correct because Speech to Text converts spoken audio into written text continuously, which is exactly what is needed for transcribing doctor dictations into electronic records.

Option a is incorrect because Text to Speech converts written text into spoken audio rather than transcribing spoken words into written text for medical record documentation.

Option c is incorrect because Speech Translation converts speech from one language to another rather than simply transcribing speech into text within the same language for medical records.

Option d is incorrect because Speaker Recognition identifies who is speaking based on voice characteristics rather than transcribing the content of speech into written text for documentation.

Correct Answer: Compute Instances

Option a is correct because Compute Instances are development workstations that data scientists can use to work with code, access powerful compute, and open tools like Jupyter, VS Code.

Option b is incorrect because Designer is a visual drag-and-drop tool for building pipelines without code rather than a collaborative environment for writing and executing Python code.

Option c is incorrect because AutoML automates model training without requiring code rather than providing a collaborative development environment for writing and sharing Python scripts.

Option d is incorrect because Model Registry stores and versions trained models rather than providing a development environment with compute resources for writing collaborative Python code.

Correct Answer: General A1

Option c is correct because General A1 domain is optimized for better accuracy with comparable inference time, recommended for larger datasets or more difficult scenarios requiring precision.

Option a is incorrect because General Compact domains are optimized for edge deployment with real-time constraints rather than maximizing accuracy for quality control in manufacturing environments.

Option b is incorrect because Food domain is specifically optimized for photographs of dishes and food items as they appear on restaurant menus, not manufacturing product defects.

Option d is incorrect because Retail domain is optimized for images found in shopping catalogs or websites with high precision for clothing classification, not manufacturing defect detection.

Correct Answer: Endpoint

Option d is correct because Endpoints host deployed models and provide remote access through REST APIs, allowing applications to invoke the model for real-time predictions.

Option a is incorrect because Pipelines are machine learning workflows that define the sequence of steps for training or batch processing rather than hosting deployed models for predictions.

Option b is incorrect because Compute Clusters provide scalable compute resources for processing and training tasks rather than hosting deployed models accessible via REST APIs.

Option c is incorrect because Experiments are logical groupings of training runs that track execution details rather than hosting deployed models accessible for making predictions.

Correct Answer: Data Stores

Option a is correct because Data Stores securely connect to storage services on Azure without putting authentication credentials at risk, providing safe access to data sources.

Option b is incorrect because Datasets are registered collections of data for machine learning workloads rather than the connection mechanism that securely links to underlying storage services.

Option c is incorrect because Compute refers to the underlying computing instances used for processing and training rather than secure connections to storage services containing training data.

Option d is incorrect because Environments are reproducible Python configurations for machine learning experiments rather than secure connection mechanisms to cloud storage services.

Correct Answer: Machine Learning Assisted Labeling

Option c is correct because Machine Learning Assisted Labeling uses machine learning to suggest labels after some training data is provided, accelerating the labeling process significantly.

Option a is incorrect because Data Stores provide secure connections to storage services containing data rather than helping with the actual labeling process or providing intelligent suggestions.

Option b is incorrect because AutoML automates the entire model training process rather than specifically assisting with the data labeling task for supervised learning preparation.

Option d is incorrect because Compute Instances are development workstations for data scientists to work with code and models rather than tools for accelerating the data labeling process.

Correct Answer: Personalizer

Option b is correct because Personalizer is specifically designed to provide personalized recommendations that learn and adapt based on user context, actions, and feedback over time.

Option a is incorrect because Anomaly Detector identifies unusual patterns and outliers in data rather than providing personalized recommendations based on user preferences and behavior.

Option c is incorrect because Content Moderator detects inappropriate content in text, images, and videos rather than providing personalized recommendations for e-commerce products.

Option d is incorrect because Text Analytics analyzes text for sentiment and entities rather than creating adaptive recommendation systems that learn from user interactions and preferences.

Correct Answer: Pipelines

Option d is correct because Pipelines are executable workflows with encapsulated steps that can be reused, shared, and worked on independently by multiple data scientists.

Option a is incorrect because Experiments are logical groupings of training runs that track execution history rather than reusable workflows with independent steps that can be shared.

Option b is incorrect because Compute provides underlying computing resources for processing tasks rather than defining reusable workflows with multiple independent steps that teams can share.

Option c is incorrect because Datasets are registered collections of data used for training rather than reusable workflows that encapsulate multiple processing and training steps.