Free AI-900 Exam Flashcards

Machine learning lets systems learn patterns from data instead of relying only on explicit rules. It matters because many AI solutions improve by training on examples.

Prediction uses historical data to estimate future values or outcomes. Common examples include sales forecasts, demand planning, and churn risk.

Anomaly detection identifies unusual patterns or outliers. It is used for fraud detection, equipment monitoring, and security alerts.

Computer vision analyzes images and video. It supports tasks such as image classification, object detection, OCR, and spatial analysis.

Natural language processing helps systems understand, analyze, or generate human language. It powers sentiment analysis, translation, summarization, and chat experiences.

Conversational AI uses bots or agents to interact with users through natural language. It matters for support, self-service, and guided workflows.

Knowledge mining extracts searchable information from large content collections. Azure AI Search is commonly used to index documents and enrich them with AI skills.

Fairness means AI systems should avoid unjust bias and treat people equitably. It matters because biased data or design can create harmful outcomes.

Reliability and safety means AI systems should work consistently and fail safely. Testing, monitoring, and fallback plans reduce risk in real use.

Privacy and security protect data used by AI systems and prevent unauthorized access. This matters because AI workloads often process sensitive information.

Inclusiveness means AI should work for people with different abilities, languages, and contexts. It encourages accessible and broadly usable solutions.

Transparency means users should understand what an AI system does and how decisions are influenced. It supports trust and appropriate human review.

Accountability means people and organizations remain responsible for AI outcomes. Governance, documentation, and review processes help enforce responsibility.

Supervised learning trains a model with labeled examples. It is used when the expected output, such as a category or numeric value, is known during training.

Unsupervised learning finds patterns in unlabeled data. Clustering similar customers or documents is a common exam example.

Classification predicts a category or class label. Examples include spam detection, disease risk category, and sentiment labels.

Regression predicts a numeric value. Examples include price, demand, temperature, or delivery time estimates.

Clustering groups similar items without predefined labels. It helps discover natural segments in customers, products, or documents.

A training dataset is used to teach a model patterns. Good training data should represent the problem and avoid avoidable bias.

A validation dataset helps tune and compare models during development. It checks performance on data not directly used for training.

A test dataset evaluates final model performance on held-out data. It matters because high training performance alone can hide overfitting.

Azure Machine Learning is a cloud platform for building, training, deploying, and managing machine learning models. It supports code-first and designer workflows.

Image classification assigns one or more labels to an image. It answers questions such as whether an image contains a product, defect, or scene type.

Object detection identifies objects and their locations in an image. It returns bounding boxes, not just image-level labels.

Optical character recognition extracts printed or handwritten text from images and documents. It matters for forms, receipts, IDs, and scanned files.

Face detection locates human faces in images. It differs from face recognition, which tries to identify or verify a person.

Azure AI Vision provides image analysis, OCR, and related visual AI capabilities through managed APIs. It reduces the need to build custom models from scratch.

Document Intelligence extracts fields, tables, and text from forms and documents. It is used when structured information must be pulled from PDFs or images.

Sentiment analysis estimates whether text expresses positive, negative, neutral, or mixed opinion. It is common in customer feedback analysis.

Key phrase extraction identifies important words or phrases in text. It helps summarize documents and tag content for search or review.

Named entity recognition detects entities such as people, places, organizations, dates, and quantities. It helps structure unstructured text.

Language detection identifies the language of text. It is useful before translation, routing, or applying language-specific analysis.

Translation converts text from one language to another. Azure AI Translator supports multilingual applications without custom language models.

Speech to text converts spoken audio into written text. It supports transcription, captions, and voice-driven applications.

Text to speech converts written text into synthetic speech. It is used for accessibility, voice assistants, and audio responses.

Speech translation translates spoken input across languages. It combines speech recognition, translation, and sometimes speech synthesis.

Azure AI Language provides text analytics and language understanding features. It supports sentiment, key phrases, entities, question answering, and custom text tasks.

Question answering returns answers from a knowledge base or source content. It is often used for FAQ bots and support assistants.

Generative AI creates new content such as text, code, images, or summaries from learned patterns. It differs from AI that only classifies or predicts.

A large language model predicts and generates text based on patterns learned from large text datasets. It powers chat, summarization, and content generation.

A prompt is the input or instruction given to a generative AI model. Clear prompts improve relevance, formatting, and task completion.

A system message gives high-level instructions that guide model behavior. It is useful for setting role, tone, constraints, and safety expectations.

Grounding connects model responses to supplied data or trusted sources. It reduces unsupported answers and helps responses fit the organization's content.

Retrieval augmented generation retrieves relevant content and provides it to a model before generation. It helps answer with current or private knowledge without retraining the model.

Azure OpenAI provides access to advanced generative AI models through Azure governance, security, and integration features. It is central to many Azure generative AI scenarios.

A copilot is an AI assistant embedded in a product or workflow. It helps users complete tasks through natural language and contextual suggestions.

Content filters help detect and reduce harmful or inappropriate generated content. They support responsible use of generative AI systems.

A token is a chunk of text processed by a language model. Token limits affect prompt size, response length, and cost.

A model deployment makes a selected model available through an endpoint. In Azure AI services, deployment configuration controls how applications call the model.

Responsible generative AI includes grounding, human review, content filtering, transparency, and privacy controls. These practices reduce risk when models produce open-ended content.