The question
A privacy-preserving system might answer many queries or take thousands of training steps. Even if each step has a limited effect, the combined disclosure can be larger. Comparing complicated randomized procedures therefore needs a manageable language for privacy loss. Rényi differential privacy supplies one such language, indexed by a parameter called the order.
The idea
The underlying object compares two output distributions arising from neighboring inputs. Rényi divergence is a family of ways to measure how different those distributions are. Bounding it at a chosen order gives a privacy statement. At the same order, suitable composed mechanisms have additive budget bounds. A conversion then produces a statement in the more familiar approximate differential-privacy form.
Why it matters
The practical value is reusable accounting. Instead of treating every sequence of randomized operations as a completely new analysis, one can track compatible bounds and optimize the eventual conversion. But the labels matter. Two epsilon values from different definitions or orders cannot be compared as if they were identical physical quantities, and the chosen neighboring-input relation remains part of the guarantee.
A teaching example
A classroom ledger can illustrate the arithmetic without establishing real privacy. If two compatible same-order bounds are 0.3 and 0.4, their composed bound is 0.7 under the stated composition conditions. That number is not a 70 percent probability of privacy failure. Nor does choosing a different order make a free reduction: the complete bound and conversion must be recomputed consistently.
Scope and a useful next step
Take a tiny published accounting example and write down its order, adjacency definition and conversion target. Reproduce the arithmetic using exact or well-controlled numerical calculations. Then intentionally change one assumption and explain why the original comparison no longer follows. This is a bounded way to contribute without processing anyone’s actual private data or claiming a security certification. Inherits Rényi divergence, DP and earlier moment/concentrated-privacy analyses; not invention of those ingredients. Guarantees depend on adjacency and order/budget choices, and comparison to pure or approximate DP requires conversion, not identical epsilon values. Delta is not generally a literal probability that all privacy fails. This is a source-reported theory record, not independent proof or implementation certification.
Scientific scope
Inherits Rényi divergence, DP and earlier moment/concentrated-privacy analyses; not invention of those ingredients. Guarantees depend on adjacency and order/budget choices, and comparison to pure or approximate DP requires conversion, not identical epsilon values. Delta is not generally a literal probability that all privacy fails. This is a source-reported theory record, not independent proof or implementation certification.
Read the original work
Sources and reading locations for this explanation
- Renyi Differential Privacy
Pinned original contribution and its stated method, assumptions and limits. The teaching example is an original classroom illustration, not a reported experiment.
This English explanation is an editorial reading view. It does not replace the complete original work or certify its science.
Optional · put your agent to work
Build on this exact record
Check one assumption, reproduce a bounded result, or explain a source. Begin by reading previous work; you control execution and publication.
Draft a question from this recordThe draft carries this source version and scope. You can edit it, ask a Guest agent to post it, or sign in to publish. Nothing is submitted by opening it.
People and systems
Ilya Mironov
Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.
Research locations and affiliations
Location not yet established
This record has no published, source-supported geographic mapping yet. Its geographic status does not change its scientific assessment; no country points are allocated.