Security, privacy, and fairness · MLD-0310
Differential privacy
A mathematical guarantee limiting how much an algorithm's output distribution changes when one protected unit's data changes.
Original reference page · detailed expansion pending. Original definitions, examples, notes and citations are preserved below. New examples are illustrative, not benchmark results.
Original example
Train with bounded individual contribution and calibrated noise.
Original caveat
The adjacency definition and privacy parameters are essential.
Context and aliases
Level: Intermediate. Stages: Data · Training · Evaluation · Deployment · Governance.
Tags: ML · DL · SLM · LLM · Security · Privacy · Safety.
Aliases: DP.
Related concepts
Concepts linking here
Technology connections
- Firecracker — Lightweight virtual-machine monitor for microVMs.
- Flower — Federated learning framework.
- gVisor — Sandboxed container runtime with a user-space kernel boundary.
- Microsoft Presidio — Toolkit for detecting and anonymizing sensitive information.
- NVIDIA FLARE — Framework for federated learning and analytics.
- Opacus — Differentially private training library for PyTorch.
- Open Policy Agent — General-purpose policy engine.
- TensorFlow Privacy — Differential privacy tools for TensorFlow.
Original references
- The Algorithmic Foundations of Differential Privacy ↗ · Reference link
- Google Machine Learning Glossary ↗ · Consulted
- OWASP Top 10 for LLM Applications ↗ · Reference link
- Original reference URL ↗