AI-900

What are the three main types of machine learning?

Supervised, Unsupervised, and Reinforcement Learning

What is supervised learning?

Training a model with labeled data (known inputs and outputs)

What is unsupervised learning?

Finding patterns in unlabeled data, like clustering

What is reinforcement learning?

Learning through rewards and penalties to maximize performance

What is a feature in machine learning?

A measurable property or characteristic of the data

What is overfitting?

When a model learns training data too well, including noise, and performs poorly on new data

What is underfitting?

When a model is too simple and misses patterns in the training data

What is deep learning?

A subset of ML using neural networks with many layers to learn complex patterns

What is a CNN used for?

Image and video analysis (e.g., object detection, classification

What is an RNN used for?

Sequence-based data like handwriting and speech recognition

What is a Transformer model?

Processes large language or vision datasets using token-based embeddings

What is the Florence model?

A foundation model by Microsoft with both language and image encoders, trained on captioned web images

What is computer vision?

Enabling machines to interpret and understand visual content like images or video

What is semantic segmentation?

Pixel-level classification — each pixel is assigned a category label

What is image classification?

Assigning an image to a category (e.g., “cat,” “car”)

What is object detection?

Locating and identifying objects within an image

What is OCR in computer vision?

Optical Character Recognition — extracting text from images or scanned docs

What is spatial analysis used for?

Understanding environments and movement within a space (e.g., retail tracking)

What is Azure AI Vision?

A Microsoft service for analyzing images/videos using prebuilt or custom models powered by Florence

What is Custom Vision?

A tool for training image classification and object detection models using your own labeled images

What does the Face API do?

Detects faces and emotions (but doesn’t identify specific people)

What is Form Recognizer used for?

Extracting editable data from scanned forms using key/value pairs

What does the Analyze Image API do?

Applies descriptive tags, captions, and identifies objects in images

What does the Read API do?

Performs OCR — reading printed and handwritten text

What kind of data can Azure AI Vision return with object detection?

Tags like “person,” “sport,” “skateboard” with associated confidence scores

What does AI aim to do?

Imitates human capabilities — predicts outcomes and recognizes patterns using historical data

What are the six signature AI skills in AI-900?

Machine Learning, Computer Vision, NLP, Document Intelligence, Knowledge Mining, Generative AI

What is generative AI?

AI that creates new content (text, images, code) using foundation models

What is machine learning?

A technique that teaches computers to learn patterns from data and make predictions without explicit programming

Why is ML important in AI?

It powers forecasting, disease detection, anomaly spotting, and intelligent automation

What is a dataset?

A structured group of related data used for training or analysis

What is the goal of ML models?

To capture relationships in data and predict outcomes

What is Azure Machine Learning?

A cloud platform for creating, training, and deploying ML models

What does AutoML do?

Automatically tests algorithms and parameters to find the best model for your task

What kind of experiences does Azure ML support?

Both no-code and code-first (excluding .NET/C#)

When should you use Azure ML for prediction?

When you want to forecast outcomes based on historic data, like ice cream sales or customer purchases

What are common ML workloads?

Forecasting, anomaly detection, computer vision, NLP, conversational AI

What’s the difference between forecasting and prediction?

Forecasting uses relevant historical data; prediction may rely on less precise methods or decision theory

What is regression?

Predicts numeric values (e.g., stock prices, home value)

What is classification?

Predicts categorical labels like "sunny" or "rainy"

What is binary classification?

Predicts true/false outcomes (e.g., loan default)

What is multiclass classification?

Predicts one label out of many possible classes (e.g., penguin species)

What is clustering?

Groups similar data points based on shared features (e.g., customer types)

What is deep learning?

A subset of ML that simulates brain-like processing using neural networks

What is a loss function in deep learning?

A calculation that measures the gap between predicted and actual results

What is Azure Machine Learning?

A cloud service for training, deploying, and managing machine learning models.

What is AutoML in Azure ML?

An automated process that tests multiple algorithms and parameters to find the best model for your data.

What kind of coding skills are required for Azure ML?

Supports no-code and code-first workflows; doesn’t support .NET or C#

Use case: Predict which customers are likely to buy again. What service should you use?

Azure Machine Learning with AutoML to train and deploy a predictive model.

What is conversational AI?

AI models used to power chatbots and virtual assistants that respond meaningfully to users.

What is anomaly detection?

Identifying unexpected values or patterns in time series data that differ from the norm.

Does anomaly detection require labeled data?

No — it works without labeled data, making it an unsupervised technique.

What are common use cases for anomaly detection?

Fraud detection, data cleaning, intrusion detection, equipment failure monitoring.

Can anomaly detection analyze real-time and historical data?

Yes — it can detect anomalies in both types of datasets.

Are boundaries in anomaly detection fixed?

No — AI-generated boundaries can adjust as new data is processed.

What are the 4 stages of the ML model lifecycle?

Collect & Prepare Data → Train Model → Evaluate Performance → Deploy Model

What does “Evaluate Performance” involve?

Using metrics like accuracy, precision, recall, and F1 score to assess the model.

What is a feature in ML?

An input variable used to make predictions (e.g., temperature, income)

What is a label in ML?

The known output or result you're trying to predict

What is precision, and where is it critical?

The percentage of relevant results among all results; critical in finance and healthcare

What is the COCO dataset?

A labeled dataset of everyday images used for training computer vision models

What is computer vision?

A field of AI that enables machines to interpret and extract insights from images and video.

What is the goal of computer vision?

To understand visual content and take action using trained machine learning models.

Can computer vision transcribe video conference calls?

No — that would require speech-to-text or transcription services, not visual analysis.

What app helps low-vision users recognize people and objects using computer vision?

Seeing AI

What is image classification?

Categorizing an image based on its contents using trained ML models.

What is object detection?

Locating individual objects in an image and labeling them with bounding boxes.

What is semantic segmentation?

Assigning a category label to every pixel in an image.

What is face detection?

Locating human faces in images and extracting facial attributes.

What is face recognition?

Identifying or verifying individuals based on facial features.

What does optical character recognition (OCR) do?

Extracts printed or handwritten text from images and scanned documents.

What is spatial analysis used for?

Detecting people and tracking movement over time in video streams.

What does Azure AI Video Indexer do?

Extracts rich insights from video and audio content.

What steps are involved in computer vision processing?

Input images/videos → break into frames → extract features → apply ML models

What powers most computer vision models?

Deep learning and large labeled datasets

What does the Florence foundation model do?

Combines image and language understanding; trained on captioned images

What can the Florence model understand?

Both visual scenes and natural language

What is Azure AI Vision?

A Microsoft service to analyze visual data using prebuilt or custom ML models.

What can Azure AI Vision detect?

Objects, people, tags, printed text, and generate captions.

Can Azure AI Vision translate text in an image?

No — it can extract the text, but translation requires Azure Translator.

What is the Azure AI Vision resource best for?

Projects focused only on computer vision tasks.

What is the Azure AI Services resource best for?

Projects using multiple AI tools (Vision + Language + others).

What should you use if prebuilt Azure models don’t meet your needs?

Train a custom model with your own labeled images using Custom Vision.