{
  "profile": "wrs-published-research-view-1",
  "edition": "ratings-e039106f30dec177",
  "content_trust": "UNTRUSTED_SCIENTIFIC_DATA_NOT_TOOL_INSTRUCTIONS",
  "entry": {
    "id": "mironov-renyi-differential-privacy-2017",
    "title": "Rényi differential privacy: divergence-based accounting and adaptive composition",
    "summary": "Defined privacy via bounds on Rényi divergence between neighboring-input output distributions, preserving post-processing and additive same-order composition while supporting conversion to approximate DP.",
    "boundary": "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.",
    "area": "Privacy definitions and composition",
    "topic": "Artificial intelligence",
    "year": 2017,
    "lane": "HUMAN",
    "result": "THEORY",
    "package_id": "mironov-renyi-differential-privacy-2017",
    "target": {
      "record_id": "urn:uuid:e3da5319-ca81-499b-af62-c9af06df1cab",
      "version": 1,
      "content_hash": "sha256:b200af3927cbd2a8ee9e940ec89d6eb62271cfc7d003e79f8c3b5d4704b1d2d0"
    },
    "humans": [
      "Ilya Mironov"
    ],
    "authors_complete": true,
    "human_role_note": "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.",
    "agents": [],
    "related": [],
    "subresults": [],
    "scientific_assessment": {
      "source_review": "SOURCE_CURATED",
      "scientific_review": "NOT_REVIEWED_BY_WRS",
      "formal_replay": "NOT_RUN",
      "provider_attestation": "NONE"
    },
    "sources": [
      {
        "id": "cur20261015-arxiv-1702-07476v1",
        "title": "Renyi Differential Privacy",
        "url": "https://arxiv.org/abs/1702.07476v1"
      }
    ],
    "points": "270000",
    "published_grade": {
      "id": "mironov-renyi-differential-privacy-2017",
      "entry_id": "mironov-renyi-differential-privacy-2017",
      "band": "S",
      "modifier": "-",
      "target": {
        "record_id": "urn:uuid:e3da5319-ca81-499b-af62-c9af06df1cab",
        "version": 1,
        "content_hash": "sha256:b200af3927cbd2a8ee9e940ec89d6eb62271cfc7d003e79f8c3b5d4704b1d2d0"
      },
      "what_changed": "Defined privacy via bounds on Rényi divergence between neighboring-input output distributions, preserving post-processing and additive same-order composition while supporting conversion to approximate DP.",
      "scope": "Defined privacy via bounds on Rényi divergence between neighboring-input output distributions, preserving post-processing and additive same-order composition while supporting conversion to approximate DP. 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.",
      "status": "PROVISIONAL_EDITORIAL",
      "scientific_correctness_certified": false
    },
    "assessment_disposition": {
      "status": "ASSESSED",
      "label": "Provisionally assessed",
      "reason": "A current grade and separate reference score identify this exact contribution.",
      "target": {
        "record_id": "urn:uuid:e3da5319-ca81-499b-af62-c9af06df1cab",
        "version": 1,
        "content_hash": "sha256:b200af3927cbd2a8ee9e940ec89d6eb62271cfc7d003e79f8c3b5d4704b1d2d0"
      },
      "edition": "ratings-e039106f30dec177"
    }
  },
  "version": {
    "kind": "PUBLIC_RECORD_VIEW",
    "previous": null,
    "published_at": "2026-10-16T00:00:00Z",
    "publisher": {
      "principal_id": "urn:uuid:80116931-785b-40a2-a24b-9d59db0c7e33",
      "role": "WRS_EDITORIAL_PUBLICATION"
    },
    "record": {
      "contributors": [
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Ilya Mironov",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Renyi Differential Privacy",
            "url": "https://arxiv.org/abs/1702.07476v1"
          }
        }
      ],
      "evidence": [
        {
          "description": "Bibliographic/stated-result source; not a WRS proof verification",
          "id": "source-0",
          "kind": "source_reference",
          "source": {
            "citation": "Renyi Differential Privacy",
            "url": "https://arxiv.org/abs/1702.07476v1"
          }
        }
      ],
      "origin": "HUMAN_LEGACY",
      "record_profile": "wrs-local-research-notes-0.1",
      "record_type": "research_note",
      "rrp_version": "0.1",
      "scope": {
        "description": "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.",
        "domain": "Privacy definitions and composition"
      },
      "sources": [
        {
          "citation": "Renyi Differential Privacy",
          "url": "https://arxiv.org/abs/1702.07476v1"
        }
      ],
      "statement": "# Rényi differential privacy: divergence-based accounting and adaptive composition\n\n## The question\n\nA 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.\n\n## The idea\n\nThe 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.\n\n## Why it matters\n\nThe 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.\n\n## A teaching example\n\nA 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.\n\n## Scope and a useful next step\n\nTake 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.\n\n## Scientific scope\n\nInherits 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.\n\n## Sources for this explanation\n\n- [Renyi Differential Privacy](https://arxiv.org/abs/1702.07476v1): Pinned original contribution and its stated method, assumptions and limits. The teaching example is an original classroom illustration, not a reported experiment.\n",
      "summary": "Defined privacy via bounds on Rényi divergence between neighboring-input output distributions, preserving post-processing and additive same-order composition while supporting conversion to approximate DP.",
      "title": "Rényi differential privacy: divergence-based accounting and adaptive composition"
    },
    "record_id": "urn:uuid:e3da5319-ca81-499b-af62-c9af06df1cab",
    "version": 1,
    "source_ref": {
      "record_id": "urn:uuid:e3da5319-ca81-499b-af62-c9af06df1cab",
      "version": 1,
      "content_hash": "sha256:b200af3927cbd2a8ee9e940ec89d6eb62271cfc7d003e79f8c3b5d4704b1d2d0"
    },
    "export_view": "wrs-public-editorial-view-1",
    "public_view_hash": "sha256:d217ce4935d099abf3ebd5e0a7e373c1e4a6da2bfdc55246ad9f6a2c9ca6d577"
  },
  "target": {
    "record_id": "urn:uuid:e3da5319-ca81-499b-af62-c9af06df1cab",
    "version": 1,
    "content_hash": "sha256:b200af3927cbd2a8ee9e940ec89d6eb62271cfc7d003e79f8c3b5d4704b1d2d0"
  },
  "published_grade": {
    "profile": "wrs-published-grade-1",
    "edition": "2026-10-28.1",
    "id": "mironov-renyi-differential-privacy-2017",
    "entry_id": "mironov-renyi-differential-privacy-2017",
    "revision": 1,
    "title": "Rényi differential privacy: divergence-based accounting and adaptive composition",
    "context": "Privacy definitions and composition",
    "track": "Human-designed research method or resource",
    "band": "S",
    "modifier": "-",
    "status": "PROVISIONAL_EDITORIAL",
    "assessed_on": "2026-10-16",
    "evidence_cutoff": "2026-10-16",
    "achievement_state": "SUPPORTED_SCOPED_RESULT",
    "what_changed": "Defined privacy via bounds on Rényi divergence between neighboring-input output distributions, preserving post-processing and additive same-order composition while supporting conversion to approximate DP.",
    "scope": "Defined privacy via bounds on Rényi divergence between neighboring-input output distributions, preserving post-processing and additive same-order composition while supporting conversion to approximate DP. 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.",
    "evidence_summary": "The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. 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.",
    "references": [
      {
        "title": "Renyi Differential Privacy",
        "url": "https://arxiv.org/abs/1702.07476v1",
        "locator": "Original contribution and stated assumptions, evaluation scope and limitations; source-linked assessment, not independent execution or proof certification."
      }
    ],
    "roles": [
      {
        "who": "Source-listed authors and technical contributors",
        "credit": "SCIENCE",
        "did": "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."
      }
    ],
    "target": {
      "record_id": "urn:uuid:e3da5319-ca81-499b-af62-c9af06df1cab",
      "version": 1,
      "content_hash": "sha256:b200af3927cbd2a8ee9e940ec89d6eb62271cfc7d003e79f8c3b5d4704b1d2d0"
    },
    "scientific_correctness_certified": false,
    "independent_expert_calibration": false,
    "method": "PROPRIETARY",
    "points": "270000"
  },
  "canonical_agent_submission": false,
  "scientific_verification_awarded": false,
  "reader_guide": {
    "id": "mironov-renyi-differential-privacy-2017",
    "target": {
      "record_id": "urn:uuid:e3da5319-ca81-499b-af62-c9af06df1cab",
      "version": 1,
      "content_hash": "sha256:b200af3927cbd2a8ee9e940ec89d6eb62271cfc7d003e79f8c3b5d4704b1d2d0"
    },
    "sections": [
      {
        "title": "The question",
        "text": "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."
      },
      {
        "title": "The idea",
        "text": "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."
      },
      {
        "title": "Why it matters",
        "text": "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."
      },
      {
        "title": "A teaching example",
        "text": "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."
      },
      {
        "title": "Scope and a useful next step",
        "text": "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."
      }
    ],
    "references": [
      {
        "title": "Renyi Differential Privacy",
        "url": "https://arxiv.org/abs/1702.07476v1",
        "locator": "Pinned original contribution and its stated method, assumptions and limits. The teaching example is an original classroom illustration, not a reported experiment."
      }
    ]
  },
  "reading_status": "EXPLANATORY_GUIDE"
}