AI-900: Cheatsheet
Domain Weights
| Domain | Weight |
|---|---|
| AI workloads and considerations | 15-20% |
| Machine learning on Azure | 15-20% |
| Computer vision workloads | 15-20% |
| NLP workloads | 15-20% |
| Generative AI workloads | 20-25% |
Responsible AI
| Principle | Remember |
|---|---|
| Fairness | Avoid bias; evaluate across groups |
| Reliability and safety | Consistent, safe behavior under expected conditions |
| Privacy and security | Protect data, access, prompts, and outputs |
| Inclusiveness | Design for diverse users and accessibility needs |
| Transparency | Explain AI use, limits, and behavior |
| Accountability | Human ownership, review, audit, and escalation |
ML Technique Lookup
| Scenario | Answer |
|---|---|
| Predict a numeric value | Regression |
| Predict a category or class | Classification |
| Find natural groupings without labels | Clustering |
| Learn complex image/audio/text patterns | Deep learning |
| Language models and attention | Transformer architecture |
| Input column used for prediction | Feature |
| Target value to predict | Label |
| Try multiple ML algorithms automatically | Automated ML |
Vision Lookup
| Scenario | Answer |
|---|---|
| Assign label to entire image | Image classification |
| Locate objects in image | Object detection |
| Extract text from image | OCR |
| General image analysis/OCR | Azure AI Vision |
| Face-specific detection/analysis | Azure AI Face |
NLP and Speech Lookup
| Scenario | Answer |
|---|---|
| Extract important terms | Key phrase extraction |
| Extract people, places, dates, quantities | Entity recognition |
| Determine positive/negative/neutral opinion | Sentiment analysis |
| Convert speech to text | Speech recognition |
| Convert text to speech | Speech synthesis |
| Convert between languages | Translation |
| Text analytics service | Azure AI Language |
| Speech input/output service | Azure AI Speech |
Generative AI Lookup
| Scenario | Answer |
|---|---|
| Generate text, summaries, code, images | Generative AI |
| Ground answers in private/fresh data | RAG |
| Numeric representation for similarity | Embedding |
| Model can process multiple input types | Multimodal model |
| OpenAI models through Azure | Azure OpenAI Service |
| Build AI apps and manage model workflows | Azure AI Foundry |
| Discover and deploy models | Azure AI Foundry model catalog |
Quick Decision Rules
Input is image/video? -> Azure AI Vision, object detection, classification, OCR, or Face depending on output.
Input is text? -> Azure AI Language for analytics, Azure OpenAI for generation, Translator for language conversion.
Input/output is audio? -> Azure AI Speech.
Need custom ML lifecycle? -> Azure Machine Learning.
Need generative AI app platform? -> Azure AI Foundry.
Need OpenAI models with Azure controls? -> Azure OpenAI Service.
Retirement Reminder
AI-900 is retiring on June 30, 2026. If studying after that date, check AI-901: Azure AI Fundamentals instead.