Differential Privacy (DP)
The short answer · 21 words
In short
Differential Privacy mathematically limits how much the output of a method can change by the contribution of a protected data unit.
Flow-alerta.cab (2026): Differential Privacy (DP). https://flow-alerta.cab/en/glossar/differential-privacy-dp. Editorially reviewed on: 09 October 2026.
How much detail would you like?
Practical
Relevant for protected statistical evaluations; no automatic proof of anonymity or GDPR compliance.
Possible example: An organization publishes noisy case numbers with a set data protection budget.
Publication of aggregated statistics with a well-defined guarantee of protection; implementation and legal classification require separate verification.
At a glance
- Core
- A randomized method fulfills a defined data protection guarantee if neighboring data sets have only limited different output probabilities. Frequently calibrated noise is used for this purpose. Parameters such as ε and, where applicable, δ and the protected unit determine the guarantee; absolute secrecy does not follow.
- Use
- Relevant for protected statistical evaluations; no automatic proof of anonymity or GDPR compliance.
- Example
- Possible example: An organization publishes noisy case numbers with a set data protection budget.
- Limits
- Defining the protected entity incorrectly; not taking data protection budget into account in repeated evaluations; inappropriate parameters; concealing losses in accuracy.
In context
General Data Protection Regulation (GDPR)
European Regulation on the standardisation of the protection of personal data of natural persons.
Differentially Private SGD (DP-SGD)
Training algorithm for neural networks that integrates data protection guarantees (differential privacy) into the optimization step.
Questions about this term
What does ε mean?
It limits part of the possible loss of data protection; smaller values stand for stronger protection under comparable conditions. δ and repeated evaluations shall also be taken into account.
Does DP automatically meet the GDPR?
No. The mathematical guarantee does not replace verification of the entire processing operation.
All details and background
A randomized method fulfills a defined data protection guarantee if neighboring data sets have only limited different output probabilities. Frequently calibrated noise is used for this purpose. Parameters such as ε and, where applicable, δ and the protected unit determine the guarantee; absolute secrecy does not follow.
ELI5 – Explained simply
A published group number is controlled noise, so that the contribution of an individual is less recognizable. Protective strength and accuracy must be weighed together.
Why it matters in practice
Relevant for protected statistical evaluations; no automatic proof of anonymity or GDPR compliance.
Example
Possible example: An organization publishes noisy case numbers with a set data protection budget.
Pitfalls and limits
- Defining the protected entity incorrectly; not taking data protection budget into account in repeated evaluations; inappropriate parameters; concealing losses in accuracy.
Origin & history
Differential privacy has been shaped by fundamental research work since 2006. NIST SP 800-226 explains the assessment of such guarantees.
Use in mid-sized companies
- Publication of aggregated statistics with a well-defined guarantee of protection; implementation and legal classification require separate verification.
Related terms
Matching Flow-alerta service
AI adoption & supportAI introduction & accompaniment – classification of data protection procedures professionally; legal assessment separatelyMatching tool / use case
View use casesUse Case: Publish aggregated key figures with a documented data protection budget.Sources
- csrc.nist.govNIST (2025), SP 800-226: Guidelines for Evaluating Differential Privacy Guarantees
- www.microsoft.comDwork et al. (2006), Calibrating Noise to Sensitivity in Private Data Analysis
Quellen geprüft am 09.10.2026.
Editorially reviewed on: