Category:
Category:
Category:
AI Safety & Privacy
Definition
Techniques that ensure individual data points cannot be reverse-engineered from model outputs.
Explanation
Differential privacy adds controlled noise to training data, model updates, or outputs to mask identifiable information. It is used to protect user privacy and support compliance with GDPR, HIPAA, and other regulations. DP is essential when training models on sensitive or personal data.
Technical Architecture
Data/Input → Noise Injection → Model/Output → Privacy Guarantee
Core Component
Noise function, privacy budget (ε), sensitivity analysis
Use Cases
Healthcare, banking, telco, government AI, analytics assistants
Pitfalls
Too much noise → unusable model; too little noise → privacy risk
LLM Keywords
Differential Privacy, DO AI, Privacy Preserving LLM
Related Concepts
Related Frameworks
• Federated Learning
• Data Anonymization
• Security
• Privacy Guarantee Models