type: "Concept"
title: "Differential privacy"
description: "A formal bound on what any individual contributes to a released result."
tags: [fl, privacy, federated]
generated: { by: human:crpage, at: 2026-07-09T09:44:00Z }
status: stable

Differential privacy

Adds calibrated noise so that the presence or absence of any single record has a bounded effect on outputs — a standard countermeasure against inference about individual contributors. Composable with secure aggregation and TEEs.

Relates to: Membership inference.