← Back to Home

Study Notes

Exam Overview

Passing Score: 700/1000

Duration: 20 min check-in + 45 min test | ~45-50 questions

Skills Measured (May 2025)

DomainWeight
AI Workloads & Responsible AI15-20%
Machine Learning Principles15-20%
Computer Vision15-20%
Natural Language Processing15-20%
Generative AI Workloads20-25%

Exam Tip

Generative AI is the largest section - prioritize studying Azure AI Foundry and Azure OpenAI Service!

Six Principles of Responsible AI

1. Fairness

AI systems should treat all people fairly without bias based on gender, race, religion, or other factors.

Example: A loan approval AI should not discriminate based on gender or race.

2. Reliability and Safety

AI should perform consistently and safely, handling unexpected situations and resisting manipulation.

Example: An autonomous vehicle must respond safely to unexpected road conditions.

3. Privacy and Security

Protect personal data, comply with regulations, and secure data storage and transmission.

Example: A health AI must protect patient medical records.

4. Inclusiveness

AI should empower everyone, including people with disabilities, considering diverse human experiences.

Example: Voice recognition should work for people with different accents.

5. Transparency

AI decisions should be understandable and explainable. Document algorithms and enable interpretability.

Example: A credit scoring AI should explain why an application was denied.

6. Accountability

People must be accountable for AI systems with clear governance, roles, and audit trails.

Example: Organizations must have processes to address AI system failures.

Machine Learning Concepts

ML Techniques

Regression

Predicts continuous numeric values

House prices, temperature, sales

Classification

Predicts categorical labels

Spam detection, disease diagnosis

Clustering

Groups similar data (unsupervised)

Customer segmentation

Key Terms

  • Features: Input variables (X) used for prediction
  • Labels: Output variable (Y) being predicted
  • Training Data: Data used to train the model (70-80%)
  • Validation Data: Data to evaluate performance (20-30%)

Exam Tip

Data should be split into ROWS (samples), NOT columns! AutoML's "Explain best model" = Transparency principle.

Computer Vision

Key Solutions

Image Classification

Assigns ONE label to entire image

Object Detection

Locates multiple objects with bounding boxes

OCR

Extracts text from images

Facial Detection

Detects faces and analyzes attributes

Face Service Operations

  • Detect: Find faces and attributes (glasses, age, emotion)
  • Verify: 1-to-1 comparison of two faces
  • Identify: 1-to-many matching against a group

Exam Tip

Use Face DETECT to check for attributes like sunglasses. Face VERIFY for identity confirmation.

Natural Language Processing

Key NLP Features

Sentiment Analysis

Positive/Negative/Neutral tone

Key Phrase Extraction

Main concepts from text

Named Entity Recognition

People, places, organizations

Language Detection

Identify text language

Azure Services

  • Azure AI Language: Text analysis (sentiment, NER, key phrases)
  • Azure AI Speech: Speech-to-text and text-to-speech
  • Azure AI Translator: Text and document translation

Exam Tip

Speech Recognition = Speech-to-Text. Speech Synthesis = Text-to-Speech.

Generative AI (20-25%)

This is the largest section - study thoroughly!

Key Concepts

  • LLMs: Large Language Models trained on massive text data
  • Transformers: Architecture using attention mechanisms
  • Prompt Engineering: Crafting effective inputs
  • RAG: Retrieval-Augmented Generation to reduce hallucinations
  • Hallucinations: AI generating false plausible information

Azure Services

Azure AI Foundry

Unified platform for AI development

Azure OpenAI Service

Access to GPT-4, DALL-E, Whisper

Model Catalog

Browse and deploy AI models

Prompting Techniques

  • Zero-shot: Direct question, no examples
  • Few-shot: Include examples in prompt
  • Chain-of-thought: Ask model to explain reasoning

Exam Tip

Know Azure AI Foundry and Azure OpenAI Service well. Understand RAG for grounding and reducing hallucinations.