AI Bot
AI-200: Deployment, Monitoring & Responsible AI Flashcards
Microsoft Azure AI Cloud Developer Associate AI-200 Flashcards
Study our AI-200: Deployment, Monitoring & Responsible AI flashcards for the Microsoft Azure AI Cloud Developer Associate AI-200 exam with 62+ flashcards. View as flashcards, a searchable table, or as a fun matching game.
| Front | Back |
| Compare ARM and Bicep | Bicep is a concise abstraction over ARM JSON templates |
| How do you detect model drift | Monitor input distributions prediction distributions and performance metrics over time |
| Name a bias mitigation technique | Pre processing reweighing in processing adversarial debiasing or post processing calibration |
| Name a CI CD tool for Azure | Azure DevOps pipelines or GitHub Actions |
| Name a global explainability method | Feature importance and partial dependence to explain overall model behavior |
| Name a local explainability method | SHAP or LIME that explains individual predictions |
| Name a scaling metric for AKS | CPU utilization or memory usage or custom application metrics |
| Name the blue green deployment pattern | Run two identical environments and switch traffic to the new one |
| What are counterfactual explanations | Show minimum changes to inputs that would change the model outcome |
| What is a canary deployment | Gradually shift a small percentage of traffic to a new version to validate changes |
| What is a container | Lightweight isolated runtime image that packages app code libraries and settings |
| What is a diagnostic log | Detailed event records emitted by a service for troubleshooting |
| What is a metric alert | Notification triggered when a telemetry metric crosses a threshold |
| What is a resource template parameter | Value passed to a template to customize deployments |
| What is a rolling deployment | Replace instances gradually to update an application with zero downtime |
| What is a sidecar container | Auxiliary container that provides supporting capabilities like logging or proxy |
| What is adversarial testing | Evaluate model robustness by generating inputs that cause incorrect predictions |
| What is AKS | Azure Kubernetes Service managed Kubernetes for container orchestration |
| What is Application Insights | Azure service for monitoring telemetry logs metrics and traces |
| What is autoscaling | Automatically adjust compute resources based on demand metrics |
| What is canary monitoring focus | Closely monitor errors latency and key metrics for the small percentage of new users |
| What is CI CD | Pipeline for Continuous Integration and Continuous Deployment |
| What is concept drift | Change in relationship between inputs and labels over time |
| What is containerization best practice | Use minimal base images and multi stage builds for smaller images |
| What is cost optimization for deployment | Right size resources use autoscaling and choose appropriate compute tier |
| What is data anonymization | Remove or transform identifiers to reduce re identifiability of individuals |
| What is data at rest encryption | Encrypt stored data using platform managed or customer managed keys |
| What is data drift | Change in input feature distribution compared to training data |
| What is data in transit encryption | Use TLS to secure data while moving between services |
| What is dataset card | Documentation describing dataset creation curation and known biases |
| What is differential privacy | Technique to add noise to data to protect individual privacy while enabling analytics |
| What is distributed tracing | Trace requests across services to diagnose latency and errors |
| What is explainability tool in Azure | InterpretML or Azure Machine Learning Responsible AI components |
| What is fairness metric to check | Statistical parity equalized odds or demographic parity for protected groups |
| What is GitOps | Use Git as the single source of truth for declarative infrastructure |
| What is GPU acceleration benefit | Faster model inference and training for parallel workloads |
| What is Infrastructure as Code | Define cloud infrastructure using machine readable templates |
| What is interpretability monitoring | Track shifts in explanation patterns to detect explanation drift |
| What is latency budget | Maximum allowable end to end time for a request to meet user expectations |
| What is logging retention policy | Define how long logs are kept to balance cost and forensic needs |
| What is Managed Identity | Azure feature that provides a secure identity for apps to access resources without secrets |
| What is model batching | Batch multiple inference requests to improve throughput and GPU utilization |
| What is model card | Documentation of model purpose performance limitations and evaluation results |
| What is model compression | Reduce model size using pruning quantization or distillation to improve latency |
| What is model distillation | Train a smaller model to mimic a larger model to improve efficiency |
| What is model explainability | Techniques to interpret model predictions and surface feature contributions |
| What is network isolation | Use VNets NSGs and private endpoints to limit network access |
| What is privacy preserving inference | Techniques such as homomorphic encryption or secure enclaves for encrypted computation |
| What is quantization | Convert model weights to lower precision to reduce memory and compute |
| What is RBAC | Role Based Access Control grants permissions to users groups and identities |
| What is request timeout tuning | Set limits on request duration to protect resources and improve SLA |
| What is resource limits in Kubernetes | Set CPU and memory requests and limits per container |
| What is responsible AI governance | Processes and policies to ensure models are safe fair and compliant |
| What is runbook | Documented operational procedures to troubleshoot and resolve incidents |
| What is secrets management | Store credentials and keys securely using services like Azure Key Vault |
| What is serverless | Compute model that abstracts servers and bills per execution time |
| What is shadow deployment | Run new model in parallel without impacting production responses to compare behavior |
| What is SLA | Service Level Agreement defines uptime and performance commitments |
| What is SLI | Service Level Indicator metric used to measure service performance |
| What is SLO | Service Level Objective measurable target within an SLA |
| What is structured logging | Log messages using a consistent schema to enable queries and analysis |
| When should you retrain a model | Significant model drift or degradation in production performance |
About the Flashcards
Flashcards for the Microsoft Azure AI Cloud Developer Associate exam review cloud application and machine learning deployment concepts, including containers, serverless computing, AKS, CI/CD pipelines, Infrastructure as Code, GitOps, and common release strategies.
Students can reinforce terminology related to monitoring, model and data drift, autoscaling, performance optimization, security, privacy, and responsible AI. The deck also covers model explainability, fairness, bias mitigation, operational reliability, service-level measurements, deployment testing, and documentation used to manage production AI systems.
Topics covered in this flashcard deck:
- Containers and AKS
- CI/CD deployment strategies
- Monitoring and model drift
- Security and data privacy
- Explainability and responsible AI
- Performance and reliability optimization