Microsoft Azure AI Fundamentals (AI-900) Quick Reference Guide
This quick reference guide provides concise information about key concepts, services, and terminology covered in the AI-900 exam. Use it for last-minute review and to reinforce your understanding of core concepts.
Exam Domains at a Glance
| Domain | Weightage | Key Focus |
|---|---|---|
| AI Workloads & Considerations | 15-20% | AI workload types, responsible AI principles |
| Machine Learning Fundamentals | 15-20% | ML techniques, concepts, Azure ML capabilities |
| Computer Vision | 15-20% | Vision solution types, Azure vision services |
| Natural Language Processing | 15-20% | NLP workloads, Azure language services |
| Generative AI | 20-25% | Gen AI features, Azure OpenAI, AI Foundry |
Key Azure AI Services Memory Map
AI & Machine Learning
- Azure Machine Learning: End-to-end ML platform
- Automated ML: Automates algorithm selection and hyperparameter tuning
- ML Designer: Visual interface for creating ML models
- ML Studio: Web portal for ML projects
Computer Vision
- Azure AI Vision: Image analysis, OCR, spatial analysis
- Azure Custom Vision: Custom image classification and object detection
- Azure AI Face: Face detection, recognition, and analysis
- Azure AI Document Intelligence: Extract data from documents (formerly Form Recognizer)
Natural Language Processing
- Azure AI Language:
- Text analysis (sentiment, key phrases, entities)
- Language understanding
- Question answering
- Conversational language understanding
- Azure AI Speech:
- Speech-to-text
- Text-to-speech
- Speech translation
- Speaker recognition
- Azure AI Translator: Text translation
Generative AI
- Azure OpenAI Service:
- GPT models (text generation)
- DALL-E (image generation)
- Embeddings (vector representations)
- Azure AI Foundry:
- Model catalog
- Model customization
- Deployment and monitoring
AI Workload Types - Quick Reference
Computer Vision Workloads
- Image Classification: “What is this image?”
- Object Detection: “What objects are in this image and where?”
- Semantic Segmentation: “Which pixel belongs to which object?”
- OCR: “What text is in this image?”
- Face Analysis: “Who is in this image and what are their characteristics?”
NLP Workloads
- Sentiment Analysis: “Is this text positive, negative, or neutral?”
- Key Phrase Extraction: “What are the important phrases in this text?”
- Named Entity Recognition: “What people, places, organizations are mentioned?”
- Language Detection: “What language is this text written in?”
- Translation: “Convert this text to another language”
- Speech Recognition: “Convert speech to text”
- Speech Synthesis: “Convert text to speech”
Document Processing
- Form Recognition: Extract data from forms
- Receipt Analysis: Extract data from receipts
- Invoice Processing: Extract data from invoices
- ID Document Analysis: Extract data from IDs
Generative AI
- Text Generation: Create human-like text
- Image Generation: Create images from text descriptions
- Code Generation: Create code from natural language
- Conversational AI: Create chatbots and assistants
Machine Learning Concepts - Quick Reference
ML Types
- Supervised Learning: Training with labeled data
- Classification: Predict categories
- Regression: Predict numeric values
- Unsupervised Learning: Finding patterns in unlabeled data
- Clustering: Group similar items
- Dimensionality Reduction: Reduce feature space
- Deep Learning: Neural networks with multiple layers
- CNNs: For images
- RNNs: For sequences
- Transformers: For text and more
ML Workflow
- Data Collection: Gather relevant data
- Data Preparation: Clean, transform, feature engineering
- Model Training: Fit model to training data
- Model Evaluation: Test on validation data
- Model Deployment: Deploy to production
- Model Monitoring: Track performance over time
ML Terminology
- Features: Input variables
- Labels: Output variables (what you’re predicting)
- Training Data: Data used to train the model
- Validation Data: Data used to tune hyperparameters
- Test Data: Data used to evaluate final model
- Overfitting: Model performs well on training data but poorly on new data
- Underfitting: Model fails to capture the underlying pattern
Responsible AI Principles - Mnemonic: “FAIR-PT”
- Fairness: AI systems should treat all people fairly
- Accountability: Take responsibility for how AI systems operate
- Inclusiveness: Design AI systems that work for everyone
- Reliability & Safety: Build systems that perform reliably and safely
- Privacy & Security: Protect user data and privacy
- Transparency: Make AI systems understandable
Common ML Algorithms - Quick Reference
Classification Algorithms
- Logistic Regression: Simple, interpretable
- Decision Trees: Hierarchical decisions
- Random Forest: Ensemble of decision trees
- Support Vector Machines: Find optimal boundary
- Neural Networks: Multi-layer networks
Regression Algorithms
- Linear Regression: Simple, interpretable
- Decision Forest Regression: Ensemble of trees
- Neural Network Regression: Multi-layer networks
Clustering Algorithms
- K-means: Partition into k clusters
- Hierarchical Clustering: Build cluster hierarchy
- DBSCAN: Density-based clustering
Azure Machine Learning Components
Compute Options
- Compute Instances: Development workstations
- Compute Clusters: Scalable training clusters
- Inference Clusters: For model deployment
- Attached Compute: Link existing resources
Data Storage
- Datastores: Connections to storage services
- Datasets: References to specific data
Experiment Tracking
- Runs: Individual training sessions
- Metrics: Performance measurements
- Logs: Detailed output
Model Management
- Registration: Store models in registry
- Versioning: Track model versions
- Deployment: Deploy as web services
- Monitoring: Track model performance
Computer Vision Concepts
Image Analysis Techniques
- Feature Extraction: Identify key features
- Transfer Learning: Reuse pre-trained models
- Data Augmentation: Create variations of training images
Evaluation Metrics
- Accuracy: Overall correctness
- Precision: Exactness of positive predictions
- Recall: Completeness of positive predictions
- F1-Score: Harmonic mean of precision and recall
- IoU: Intersection over Union for object detection
- mAP: Mean Average Precision for object detection
NLP Concepts
Text Processing Techniques
- Tokenization: Split text into tokens
- Stemming/Lemmatization: Reduce words to base form
- Stop Word Removal: Remove common words
- Vectorization: Convert text to numeric form
- Embeddings: Represent words as vectors
Speech Processing
- Acoustic Modeling: Map audio to phonemes
- Language Modeling: Predict word sequences
- SSML: Speech Synthesis Markup Language
Generative AI Concepts
Model Types
- Autoregressive Models: Generate one token at a time
- Diffusion Models: Gradually denoise random patterns
- GANs: Generator and discriminator networks
Techniques
- Prompt Engineering: Craft effective prompts
- Fine-tuning: Adapt pre-trained models
- Few-shot Learning: Learn from few examples
- Zero-shot Learning: Perform tasks without specific training
Evaluation
- Human Evaluation: Human judgment of quality
- Automated Metrics: BLEU, ROUGE, etc.
- Alignment: Adherence to human values and preferences
Exam Tips
General Tips
- Read questions carefully - look for keywords
- Eliminate obviously wrong answers first
- Watch for “NOT” or “EXCEPT” in questions
- Manage time - 45 minutes for 40-60 questions
Question Types
- Multiple Choice: Select one correct answer
- Multiple Selection: Select all that apply
- Case Studies: Questions based on scenarios
- Drag and Drop: Match items to categories
Key Areas to Focus
- Azure service capabilities and use cases
- Responsible AI principles
- ML concepts and terminology
- Differences between AI workload types
- Azure OpenAI Service features
Last-Minute Review
- Review Azure service names and capabilities
- Memorize responsible AI principles
- Understand ML types and scenarios
- Know the differences between vision, NLP, and generative AI workloads
- Familiarize yourself with Azure AI Foundry and OpenAI Service
Good luck on your exam!