AI Bot
Responsible AI, Privacy & Compliance Flashcards
Microsoft Azure AI Fundamentals AI-901 Flashcards
Study our Responsible AI, Privacy & Compliance flashcards for the Microsoft Azure AI Fundamentals AI-901 exam with 55+ flashcards. View as flashcards, a searchable table, or as a fun matching game.
| Front | Back |
| Define demographic parity. | Fairness metric requiring equal positive outcome rates across groups |
| Define equal opportunity. | Fairness metric requiring equal true positive rates across groups |
| Define equalized odds. | Fairness metric requiring equal true positive and false positive rates across groups |
| Define fairness in AI. | Absence of systematic and unjustified bias in model outcomes across groups |
| How does Explainable AI support compliance? | Provides documentation and reasons for decisions to satisfy regulators and audits |
| List three bias sources in ML. | Data collection label errors and modeling choices |
| Name four core Responsible AI principles. | Fairness Transparency Accountability Privacy |
| Name three bias mitigation strategies. | Preprocessing re sampling in processing and postprocessing adjustments |
| Name three common compliance standards. | GDPR HIPAA SOC |
| What are counterfactual explanations? | Examples that show minimal changes to input to change a prediction |
| What are GDPR data subject rights? | Right to access rectification erasure objection and data portability |
| What are model cards used for? | Communicating model intended use evaluation metrics limitations and risks |
| What Azure service helps with DLP? | Microsoft Purview Data Loss Prevention |
| What is a datasheet for datasets? | Record that documents dataset creation composition recommended uses and limitations |
| What is a least privilege principle? | Granting users and services only the minimum permissions needed |
| What is a model card? | Documentation summarizing model purpose performance limitations and training data |
| What is anonymization? | Processing to irreversibly prevent identification of individuals in a dataset |
| What is Azure Blueprints? | Tool to deploy a repeatable set of governance artifacts like policies role assignments and templates |
| What is Azure Compliance Manager? | Service to assess and manage compliance posture and controls |
| What is Azure Key Vault used for? | Secure storage of keys secrets and certificates for encryption and access control |
| What is Azure Policy? | Service to enforce organization rules and ensure compliance across Azure resources |
| What is bias in AI? | Systematic error that causes unfairness in model predictions for certain groups |
| What is consent in data processing? | Freely given informed and specific permission from the data subject |
| What is continuous compliance? | Automated ongoing evaluation and enforcement of compliance controls |
| What is data classification? | Process to label data by sensitivity to apply appropriate controls |
| What is data minimization? | Principle to collect only data necessary for a specific purpose |
| What is data privacy? | Protection of personal data from unauthorized access and improper use |
| What is data residency? | Requirement to store data within a specific geographic region or jurisdiction |
| What is differential privacy? | Technique adding noise to data or queries to protect individual privacy |
| What is disparate impact? | Measure of adverse effect of a decision on a protected group often ratio based |
| What is encryption at rest? | Encryption of stored data to protect it from unauthorized access |
| What is encryption in transit? | Encryption of data while it moves between systems to prevent interception |
| What is Explainability technique SHAP? | Method that attributes contribution of each feature to individual predictions |
| What is feature importance? | Ranking of features by their overall influence on model predictions |
| What is Federated Learning? | Training models across decentralized devices without centralizing raw data |
| What is GDPR? | European regulation for data protection and privacy including rights for data subjects |
| What is homomorphic encryption? | Cryptographic technique enabling computation on encrypted data without decryption |
| What is interpretability? | Ability to explain how a model makes decisions in understandable terms |
| What is InterpretML? | Open source toolkit for model interpretability supporting glassbox and blackbox explanations |
| What is k anonymization? | Privacy technique that groups records to prevent re identification |
| What is LIME? | Local surrogate model that approximates complex model behavior near an instance |
| What is logging and monitoring for AI? | Collecting and analyzing logs and metrics to detect issues and support audits |
| What is model calibration? | Degree to which predicted probabilities match actual outcome frequencies |
| What is model retraining governance? | Policies and processes to control when how and who can retrain models |
| What is network isolation? | Segmentation of resources to restrict network access and reduce attack surface |
| What is partial dependence? | Global interpretability method showing relationship between a feature and model predictions |
| What is Privacy by Design? | Embedding privacy principles into systems and processes from the start |
| What is pseudonymization? | Replacing identifiers with pseudonyms to reduce identifiability while allowing linkage |
| What is purpose limitation? | Data should be used only for the purposes specified at collection time |
| What is Responsible AI? | Set of principles and practices to ensure AI is fair transparent accountable and respects privacy safety and human rights |
| What is Role Based Access Control RBAC? | Azure access management service that grants permissions by role |
| What is Secure Development Lifecycle? | Practices to integrate security checks across software development phases |
| What is secure multiparty computation? | Protocol allowing parties to jointly compute a function without revealing inputs |
| What is telemetry sanitization? | Removing or masking sensitive data from logs before storage or analysis |
| What is the Responsible AI dashboard? | Azure tool for evaluating model fairness interpretability and error analysis |
About the Flashcards
Flashcards for the Microsoft Azure AI Fundamentals exam review Responsible AI principles, including fairness, transparency, accountability, privacy, interpretability, and bias mitigation. Students can reinforce their understanding of fairness metrics, model calibration, explainability methods such as SHAP and LIME, and documentation tools like model cards and dataset datasheets.
The deck also covers privacy-enhancing technologies, Azure security controls, governance, and regulatory compliance. Key concepts include encryption, access control, network isolation, GDPR rights, data minimization, secure development, monitoring, and continuous compliance. These flashcards help students master terminology and key ideas related to trustworthy, secure, and compliant AI systems.
Topics covered in this flashcard deck:
- Responsible AI principles
- Fairness and bias metrics
- Model explainability techniques
- Privacy-enhancing technologies
- Azure security and governance
- GDPR and compliance