Data · Established · Advanced

Differential Privacy

A mathematical guarantee that the presence or absence of any single individual's record barely changes a system's output.

What Differential Privacy is

Differential privacy replaces informal anonymisation with a measurable privacy budget, making claims about protection precise rather than rhetorical.

How it works

Calibrated noise is added to queries, gradients or aggregates. The privacy parameter quantifies leakage, and each query consumes part of a fixed budget.

Why it matters

It allows statistics and model training on sensitive data with a defensible guarantee, and is used in census and telemetry programmes.

Common uses

→Population statistics releases
→Private model training
→Device telemetry aggregation

Strengths

✓Provable, quantified guarantee
✓Composable across queries

Watch for

✓Accuracy cost
✓Budget management is complex

Continue exploring

Synthetic Data Data Privacy in AI