AZ-901

Artificial intelligence (AI)

software that mimics human cognitive abilities. Microsoft groups workloads into 6 categories.

Generative AI

produces new content (text, code, images, audio) from a prompt. Powered by large foundation models.

Machine learning

predicts outputs from patterns in data. Includes regression (numbers), classification (categories), clustering (groups).

Computer vision

interprets images & video: classification, object detection, OCR, image generation.

Natural language processing (NLP)

understands text & speech: sentiment, entities, translation, summarization.

Document intelligence / Information extraction

pulls structured data (fields, tables) from forms, invoices, IDs.

Knowledge mining

indexes large unstructured corpora so they're searchable and minable for insights.

Supervised learning

training data has known labels. Model learns input→output mapping. Used for regression & classification.

Unsupervised learning

no labels. Model finds structure (e.g., clustering customers by behavior).

Regression

predicts a numeric value (price, temperature, sales).

Classification predicts a category

binary (spam/not spam) or multi-class (cat/dog/rabbit).

Features

are the inputs (columns); the label is the target you're predicting.

Training =
Inference =

fitting a model on past data.
using the trained model to predict on new data.

Image classification

assigns ONE label to a whole image (cat / dog).

Object detection

finds MULTIPLE objects + bounding boxes (3 cars, 1 person at coordinates...).

Semantic segmentation

labels every PIXEL by class (sky, road, building).

OCR

extracts printed AND handwritten text from images.

Sentiment analysis

positive / neutral / negative / mixed at document & sentence level.

Named Entity Recognition (NER)

extracts people, places, organizations, dates, quantities.

Key phrase extraction

pulls main topics.

Translation

between languages.

Summarization

extractive (key sentences) or abstractive (rewrites).

Recite all six Microsoft Responsible AI principles and pair each with a real risk scenario. (F-R-P-I-T-A)

Fairness, Reliability, Privacy, Inclusiveness, Transparency, Accountability.

1. Fairness

treat all people fairly, avoid bias. Risk: a loan model that rejects more applicants from certain postcodes.

2. Reliability & Safety

works consistently, handles edge cases. Risk: a self-driving system fails in heavy rain.

3. Privacy & Security

protect personal/sensitive data. Risk: medical records leak through model logs.

4. Inclusiveness

empower people of all abilities and backgrounds. Risk: a chatbot fails users with regional accents.

5. Transparency

explain how the system works and its limits. Risk: an audit can't trace why the model rejected an application.

6. Accountability

humans remain responsible for outcomes. Risk: nobody owns when the AI makes a wrong decision.