Microsoft Azure AI Fundamentals (AI-900) flashcards

Facial Analysis extracts attributes (age, gender, emotion) from faces. Face Recognition identifies specific individuals.

• Facial Analysis: Focuses on characteristics, often anonymous.
• Face Recognition: Focuses on identity, matching to known individuals.
• Both are subsets of computer vision, but serve different purposes and have different privacy implications.

Memory trick: Analyze faces for traits, don't recognize names.

A computer vision task that predicts the distance of each pixel in an image from the camera, creating a depth map that represents the 3D structure of the scene.

• Essential for augmented reality (AR) and virtual reality (VR).
• Enables realistic placement and interaction of virtual objects in real environments.
• Can be performed using stereo cameras or monocular (single) images with AI.

Memory trick: Vision measures depth to build a 3D home.

An NLP task that identifies and categorizes key information (entities) like names of persons, organizations, locations, expressions of times, quantities, monetary values, percentages, etc., in unstructured text.

• Categorizes entities into predefined types.
• Fundamental for information extraction and search.
• Widely used in media monitoring and data structuring.

Memory trick: NER Finds Names Everywhere Readily

Using computer vision to calculate the distance of objects in a scene from the camera, enabling accurate 3D spatial understanding for augmented reality applications.

• Generates a depth map, where pixel values represent distance.
• Crucial for realistic virtual object placement and interaction.
• Can be achieved using stereo cameras, LiDAR, or monocular depth prediction models.

Memory trick: Estimate depth to correctly place AR.

A computer vision capability that extracts attributes from human faces within an image, such as age, gender, emotion, and head pose, without identifying the person.

• Focuses on attributes of a face, not identity.
• Commonly used for demographic analysis and emotional understanding.
• Differs from Face Recognition, which aims to identify individuals.

Memory trick: Faces can be identified, analyzed, or simply detected.

Custom Named Entity Recognition is an NLP capability that enables users to train models to identify and categorize specific, domain-specific entities from text, beyond what pre-trained models can offer.

• Requires labeled training data relevant to the custom entities.
• Essential for specialized industries like legal, medical, finance.
• Part of Azure Cognitive Service for Language's custom features.

Memory trick: Custom Entities Need Custom Recognition.

A specialized capability within the Azure Computer Vision service designed for highly accurate OCR, capable of extracting text from various types of images, including handwritten and printed documents, even with complex layouts or challenging conditions.

• Optimized for both printed and handwritten text.
• Handles various languages and orientations.
• More robust than basic OCR for complex scenarios like historical documents.

Memory trick: Read API deciphers even the oldest scrolls.

The process of automatically identifying the natural language in which a piece of text is written.

• Crucial for multilingual data processing.
• Often a prerequisite for other NLP tasks.
• Supported by Azure Cognitive Service for Language.

Memory trick: Detect the language before you can speak it.

An Azure AI service feature used to train a custom image classification model that can categorize images into specific, user-defined classes based on provided training data.

• You provide the images and define the labels (classes).
• Ideal for niche classification tasks not covered by generic pre-trained models.
• Can be used to classify entire images or identify regions within images (object detection).

Memory trick: Custom Vision recognizes unique images with custom training.

Computer vision techniques used to enable augmented and virtual reality experiences, often involving real-time scene understanding.

• Key capabilities include 3D reconstruction, object tracking, and semantic segmentation.
• Enables realistic placement and interaction of virtual objects in real environments.
• Requires high precision and real-time performance.

Memory trick: For AR, segment the scene semantically.

A cloud-based neural machine translation service that enables fast, accurate, and scalable text translation across many languages.

• Supports real-time text translation.
• Offers translation for over 100 languages and dialects.
• Integrates easily with other Azure AI services.

Memory trick: Translator Turns Text Towards Target Tones

An advanced NLP technique that identifies specific aspects or entities in text and then determines the sentiment expressed towards each of those individual aspects.

• Provides fine-grained sentiment analysis.
• Identifies target entities (e.g., product features).
• Crucial for detailed product feedback analysis.

Memory trick: ABSA Analyzes By Specific Aspects

The process of identifying and filtering inappropriate or policy-violating content across multiple languages.

• Often requires Machine Translation as a precursor.
• Utilizes NLP for sentiment, toxicity, and policy checks.
• Essential for global communication platforms.

Memory trick: Translate to understand, then moderate to comply.

The NLP task of assigning one or more predefined categories or labels to a piece of text.

• Used for spam detection, content routing, sentiment analysis (as a binary classification).
• Requires a set of predefined categories.
• Can be custom-trained for specific domains.

Memory trick: Classify text to put it in the right folder.

A technology capable of identifying or verifying a person from a digital image or a video frame.

• Compares detected faces against a database of known faces.
• Requires a 'face print' or 'template' for each individual.
• Used for security, access control, and personalized experiences.

Memory trick: To recognize a face, you need Face Recognition.

A cloud-based service for building, deploying, and improving custom image classification and object detection models.

• Train models with your own specific image data.
• Ideal for niche use cases where pre-trained models aren't sufficient.
• Supports both classification (tagging images) and object detection (locating objects).

Memory trick: Azure offers vision tools: general, custom, faces, forms.

An Azure AI service that uses machine learning to identify and extract key-value pairs, text, and tables from documents.

• Specializes in structured data extraction from forms and documents.
• Can handle both standardized and non-standardized document layouts.
• Offers pre-built models and custom model training.

Memory trick: Forms Recognize Data from Documents.

An NLP capability that allows users to train models to identify and extract domain-specific or organization-specific entities from text.

• Extends standard NER for unique terms.
• Requires labeled training data.
• Useful for specialized industries (medical, legal, finance).

Memory trick: Custom NER for custom names.

Azure Translator is a cloud-based neural machine translation service that enables fast and accurate language translation.

• Supports over 100 languages and dialects.
• Used for real-time translation, document translation, and website localization.
• Can be integrated into applications, websites, and workflows.

Memory trick: Translator Bridges Language Gaps Swiftly.

Combining Speech-to-Text with Text Analytics allows for the conversion of spoken audio into text, followed by the extraction of insights like sentiment, keywords, and entities from the transcribed content.

• Common in call center analytics, voice assistants, meeting transcription.
• Azure Speech Service provides the transcription part.
• Azure Text Analytics provides the NLP analysis part.

Memory trick: Speak, Transcribe, Analyze, Understand.

Using object detection models to identify and localize individual instances of objects within an image, thereby enabling accurate counting of those objects.

• Each detected object gets a bounding box and a class label.
• Useful for inventory, traffic analysis, and wildlife monitoring.
• Provides both 'what' and 'where' information.

Memory trick: Detect objects to count and place them.

A cloud-based service that provides access to advanced image analysis algorithms.

• Offers pre-trained models for common vision tasks.
• Capabilities include object detection, image captioning, OCR, and celebrity recognition.
• Can analyze images and videos to extract information.

Memory trick: Computer Vision is the general-purpose eye for Azure.

The NLP task of automatically producing human-like text, often based on a given input or context.

• Creates new, original text.
• Used in chatbots, content creation, summarization.
• Leverages deep learning models.

Memory trick: Generate text, like a robot author.

An NLP capability that allows users to define their own categories and train a model to classify text into those specific, custom-defined categories.

• Requires labeled data for training.
• Highly adaptable to specific domain requirements.
• Used for custom content moderation, routing, or policy enforcement.

Memory trick: Custom Classification Checks Compliance Critically

Computer Vision focuses on interpreting visual data (images, videos), while Natural Language Processing (NLP) focuses on interpreting textual and spoken language data.

• Computer Vision tasks include object detection, image classification, face recognition.
• NLP tasks include sentiment analysis, text summarization, language translation.
• Some tasks, like OCR, bridge both by extracting text (CV) for further NLP processing.

Memory trick: AI's mind sees and speaks.

An NLP task that condenses a longer text document into a shorter, coherent, and fluent summary while retaining the most important information.

• Can be extractive (pulls sentences) or abstractive (generates new sentences).
• Useful for quickly grasping main points of long documents.
• Reduces information overload.

Memory trick: Summarize to get the short story.

Leveraging Azure Custom Vision to train AI models that can automatically identify specific defects or anomalies on products by learning from labeled images.

• Enables recognition of highly specific visual features.
• Reduces manual inspection time and improves consistency.
• Requires a dataset of images showing both good and defective products.

Memory trick: Custom Vision spots the custom flaws.

Applications of object detection beyond basic identification, often involving real-time analysis.

• Quality control in manufacturing.
• Autonomous vehicles for obstacle avoidance.
• Security surveillance for anomaly detection.
• Safety monitoring in industrial environments.

Memory trick: Detecting specific safety items requires object detection.

Sentiment Analysis is an NLP technique used to determine the emotional tone (positive, negative, neutral) of text.

• Used for customer feedback, social media monitoring, brand reputation.
• Can operate at document, sentence, or aspect level.
• Often provides a confidence score along with the sentiment label.

Memory trick: Opinions Reveal Sentiments Clearly.

An Azure AI service that detects and flags potentially harmful or unwanted user-generated content in text and images across various categories, providing moderation scores.

• Identifies categories like hate, sexual, self-harm, violence.
• Provides confidence scores for detected categories.
• Supports multilingual content moderation.

Memory trick: Content Safety Controls Classification

Personally Identifiable Information (PII) Detection is an NLP capability that identifies and optionally redacts sensitive personal data from text.

• Helps with data privacy and compliance (e.g., GDPR, HIPAA).
• Can identify names, addresses, credit card numbers, email addresses, etc.
• Often used in conjunction with redaction for security.

Memory trick: Privacy Is Paramount in Data.

A comprehensive Azure AI service that helps organizations create safer online experiences by detecting harmful content across text, image, and video modalities with customizable severity levels.

• Moderates text, images, and videos.
• Detects hate, sexual, self-harm, and violence content.
• Provides severity scores and classifications.

Memory trick: Safety's eye watches all media's content.

A powerful combination for extracting domain-specific information from scanned or image-based documents, where OCR converts the image to text and Custom NER identifies specialized entities.

• OCR is crucial for non-digital text sources.
• Custom NER allows training for unique entity types.
• Ideal for legal, medical, or industry-specific document processing.

Memory trick: Scanned words need seeing, then custom finding.

An Azure AI service that creates a conversational Q&A layer over data. It allows users to query information in natural language and receive direct, accurate answers.

• Builds a knowledge base from structured/unstructured content.
• Understands natural language questions.
• Returns precise answers, not just relevant documents.

Memory trick: QnA: Quick, Nifty Answers for all your docs.

Language Detection identifies the natural language of a text, while Azure Translator provides machine translation services to convert text from one language to another.

• Language Detection identifies over 120 languages.
• Azure Translator supports over 100 languages for translation.
• Often used together for multilingual content processing.

Memory trick: Detect the tongue, then Translate the words.

An Azure AI service that helps detect potentially offensive, risky, or otherwise undesirable user-generated content across text, images, and videos.

• Automates content review processes.
• Identifies hate speech, self-harm, sexual, violent content.
• Supports multiple languages and content types.

Memory trick: Safety is key when users are chatting online.