Microsoft Azure AI Fundamentals (AI-900) Exam Preparation

Comprehensive vocabulary flashcards covering the core concepts, models, and Azure services tested on the Microsoft Azure AI Fundamentals (AI-900) certification exam.

DALL-E

The Azure OpenAI model designed for image generation from text descriptions.

Temperature

A parameter that controls the randomness of model responses; a lower value closer to $0$ makes outputs more deterministic and focused.

Fine-tuning

The process of training a pre-built model with custom data to specialize its performance.

Serverless API Deployment

A deployment option billed based on tokens used for inputs and outputs to the API.

Microsoft Foundry Portal

A portal providing model benchmarking capabilities to compare metrics across different AI models.

Whisper

The Azure OpenAI model designed primarily for speech-to-text and audio transcription tasks.

Codex

The Azure OpenAI model trained specifically on code and optimized for programming and code completion tasks.

Grounding

Providing an AI model with additional context or data (such as documents or databases) to improve response relevance and accuracy.

Fairness Principle

A responsible AI principle requiring that AI systems treat all people equitably and avoid discriminatory bias based on race, gender, age, or other protected characteristics.

Transparency Principle

A responsible AI principle stating that AI decision-making processes must be explainable, understandable, and recorded so approved users know why decisions were made.

Privacy and Security Principle

A responsible AI principle ensuring that AI systems protect sensitive information and keep personal data visible only to approved users.

Reliability and Safety Principle

A responsible AI principle aiming for ongoing monitoring, feedback, evaluation, and robust operation under unexpected conditions, such as correctly handling unusual or missing values.

Inclusiveness Principle

A responsible AI principle ensuring AI systems are accessible, empower everyone, and work equally well for people with disabilities.

Accountability Principle

A responsible AI principle dictating that the people and organizations who build and deploy AI systems are responsible for their outcomes and impacts.

Regression

A machine learning technique used to predict a continuous numeric value (e.g., house prices or taxi journey fares).

Feature

An input variable or predictor used to train a machine learning model (e.g., trip distance, age, or education level).

Label

The known output value or target variable that a supervised machine learning model is trained to predict.

Confusion Matrix

A grid or table displaying the counts of predicted and actual true positives, false positives, true negatives, and false negatives.

Clustering

An unsupervised learning technique that groups similar data points together without relying on predefined, labeled data.

Multi-class Classification

A classification task that categorizes inputs into one of several discrete classes (e.g., mild, moderate, or severe).

Overfitting

A scenario where a model learns patterns and noise in the training data too closely, resulting in poor performance on unseen data.

Root Mean Squared Error (RMSE)

An evaluation metric that measures the average magnitude of prediction errors in a regression model.

Compute Instance

A managed cloud-based workstation in Azure Machine Learning used by developers for interactive development and experimentation.

Compute Cluster

A scalable Azure Machine Learning resource used for parallel processing, training workloads, and batch inference.

Inference Cluster

An Azure Machine Learning compute target used to attach an existing Azure Kubernetes Service (AKS) cluster for model deployment.

Automated Machine Learning (AutoML)

A capability in Azure Machine Learning that automatically tests algorithms and tunes hyperparameters to identify the best model.

Validation Dataset

A dataset used during model development to evaluate performance and tune hyperparameters prior to testing on the final test set.

Sentiment Analysis

A natural language processing workload that analyzes text to determine the emotional tone (positive, negative, or neutral).

Key Phrase Extraction

A natural language processing capability that identifies the main talking points or concepts within text documents.

Named Entity Recognition (NER)

A natural language processing feature that detects and categorizes entities (such as companies, organizations, people, and places) in text.

Language Detection

A natural language processing capability that identifies the language in which a given text is written.

Object Detection

A computer vision workload that identifies and locates multiple objects within an image, outputting bounding box coordinates.

Optical Character Recognition (OCR)

A computer vision technique that extracts machine-readable printed or handwritten text from images and documents.

Facial Analysis

A computer vision capability that evaluates facial attributes (such as exposure quality, noise, and occlusion) rather than identity.

Speech Recognition

A speech service technology (speech-to-text) that converts spoken audio into written text or captions.

Speech Synthesis

A speech service technology (text-to-speech) that converts written text into synthesized spoken audio.

Custom Neural Voice

A feature in Azure Speech used to create a customized synthetic voice model based on specific human voice samples.

Azure Document Intelligence

An Azure Foundry service (formerly Form Recognizer) used to extract structured data (such as invoice numbers or vendor names) from forms, receipts, and invoices.

Azure AI Custom Vision

A service designed for creating, training, and deploying customized image classification and object detection models.

Prediction Endpoint

The published endpoint in Azure AI Custom Vision used to submit new images and receive predictions from a trained model.

Azure Machine Learning Pipeline

A sequence of connected steps used to prepare data, train models, evaluate performance, and orchestrate ML workflows.

Transfer Learning

A technique where knowledge from a model pre-trained on a large dataset is used as the starting point for a new, related task.

Prompt Engineering

The practice of designing and structuring input prompts strategically to guide generative AI models to produce desired outputs.

AI Democratization

Microsoft's strategy to make AI tools, platforms, and services accessible to a wide variety of users and organizations, beyond specialized data scientists.

Precision

A classification metric defined as $\frac{\text{correct positive predictions}}{\text{all positive predictions}}$.

Recall

A classification metric defined as $\frac{\text{correct positive predictions}}{\text{all actual positives}}$.

Accuracy

A model evaluation metric measuring the proportion of correctly classified instances out of all total instances.

Knowledge Mining

The extraction of latent insights, entities, and structured information from vast volumes of unstructured content like text, images, and audio.

Reinforcement Learning

A machine learning paradigm where an agent learns through trial and error by receiving rewards for desired behaviors and penalties for undesired ones.

Multimodal AI

AI systems that process, understand, and integrate multiple forms of data (such as text, images, and audio) simultaneously.

Azure AI Content Safety

A service that detects, monitors, and filters harmful, offensive, or inappropriate text and images to ensure safe AI outputs.

Microsoft AI-900 Exam

A foundational certification exam designed to validate core knowledge of artificial intelligence concepts and Microsoft Azure AI services.