AI-500: Core Exam Objectives Flashcards
Microsoft Multi-Agent AI Solutions Expert AI-500 Flashcards
| Front | Back |
| Common metrics for classification | Accuracy precision recall F1 AUC |
| Common metrics for regression | RMSE MAE R squared |
| Define reinforcement learning | Learning by trial and error using rewards and penalties |
| Define supervised learning | Training models with labeled data to predict outcomes |
| Define unsupervised learning | Finding patterns in unlabeled data such as clustering and dimensionality reduction |
| Describe data labeling best practice | Clear guidelines quality checks and inter annotator agreement |
| Example of authentication method for Azure AI services | Managed identities service principals or API keys |
| Explain concept of model interpretability | Techniques to explain model predictions such as SHAP LIME and feature importance |
| Explain model drift | Performance degradation over time due to changing data distributions |
| Give one coordination strategy for multi agent systems | Centralized coordinator to assign tasks and resolve conflicts |
| How to mitigate hallucinations in LLMs | Use grounding retrieval chains prompt constraints and answer verification |
| How to monitor model performance in production | Collect input output metrics drift alerts and user feedback |
| Key exam study tip for AI 500 | Focus on service capabilities security and real world trade offs |
| Key responsibilities of an AI Engineer for Azure AI | Design implement and monitor AI solutions using Azure AI services |
| Key security considerations for AI solutions | Data privacy access control model integrity and auditing |
| Multi agent system definition | Multiple agents interacting to achieve individual or shared goals |
| Name Azure service for building conversational agents | Azure Bot Service and Language Service for chatbots |
| Purpose of Azure Cognitive Services | Prebuilt APIs for vision speech language and decision capabilities |
| Typical failure modes of conversational AI | Off topic responses hallucinations context loss and misinterpretation |
| What is a large language model LLM | Neural network trained on vast text to generate and understand language |
| What is a responsible AI checklist item | Bias assessment and mitigation plan |
| What is conversational memory | Storing dialogue state to maintain context across turns |
| What is data labeling in Azure Machine Learning | Process of annotating training data using labeling projects and workers |
| What is deployment slot in Azure ML | An environment to stage validate and roll out model versions |
| What is feature engineering | Creating transforming and selecting features to improve model performance |
| What is prompt engineering | Designing inputs to guide LLM outputs effectively |
| What is Responsible AI | Designing AI that is fair transparent accountable and safe |
| What is retrieval augmented generation RAG | Combining external documents retrieval with LLM generation for grounded answers |
| When to choose custom model vs prebuilt API | Use custom for domain specific needs prebuilt for quick general tasks |
| When to use containers for model deployment | When you need portability scalability and custom runtime dependencies |
About the Flashcards
Flashcards for the Microsoft Multi-Agent AI Solutions Expert exam give a compact, practice-oriented review of core machine learning and cloud platform topics you need to know. Cards define learning paradigms (supervised, unsupervised, reinforcement), large language models, and key managed services such as cognitive, bot, and machine learning platform features.
They emphasize terminology, concepts, and practical exam-ready ideas such as model training and evaluation metrics, data labeling and feature engineering, deployment and monitoring (deployment slots, containers, authentication), model interpretability, responsible practices and security, prompt engineering, retrieval-augmented generation, and conversational system failure modes and mitigation. They are arranged for quick drilling and confident recall.
Topics covered in this flashcard deck:
- Learning paradigms
- Large language models
- Cloud platform services
- Model deployment & monitoring
- Metrics & interpretability
- Responsible practices & security