{"paper":{"title":"Limits of Personalizing Differential Privacy Budgets","license":"http://creativecommons.org/licenses/by/4.0/","headline":"For mean estimation, a simple thresholding operator on privacy budgets captures nearly all the utility gains of full personalization.","cross_cats":["cs.LG"],"primary_cat":"cs.CR","authors_text":"Edwige Cyffers, Juba Ziani","submitted_at":"2026-05-13T13:24:50Z","abstract_excerpt":"A key technical difficulty in differential privacy is selecting a privacy budget that satisfies privacy requirements while maximizing utility. A natural and well-studied workaround is to use personalized privacy budgets, which may differ across agents. In this paper, we show that personalized budgets come with major limitations and that for mean estimation, the dominant factor is not full personalization, but rather choosing the right effective privacy budget. This can be achieved through a simple thresholding operator that we describe. Compared with this thresholding baseline, the gains obtai"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"we show that personalized budgets come with major limitations and that for mean estimation, the dominant factor is not full personalization, but rather choosing the right effective privacy budget. This can be achieved through a simple thresholding operator that we describe.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The analysis assumes standard additive-noise mechanisms for mean estimation and specific distributions of privacy requirements (mixed public-private or two-level), which may not capture all real-world data distributions or query types.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"For mean estimation, a simple thresholding operator on privacy budgets matches the performance of fully personalized differential privacy mechanisms up to constant factors.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"For mean estimation, a simple thresholding operator on privacy budgets captures nearly all the utility gains of full personalization.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"ac336d53e54dd76196459b62de22463e7cbff17e9701dbf8a4a2113fc2642ce3"},"source":{"id":"2605.13503","kind":"arxiv","version":1},"verdict":{"id":"d97d0b83-a8da-414e-8836-24a7fa066c7d","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T18:12:18.227737Z","strongest_claim":"we show that personalized budgets come with major limitations and that for mean estimation, the dominant factor is not full personalization, but rather choosing the right effective privacy budget. This can be achieved through a simple thresholding operator that we describe.","one_line_summary":"For mean estimation, a simple thresholding operator on privacy budgets matches the performance of fully personalized differential privacy mechanisms up to constant factors.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The analysis assumes standard additive-noise mechanisms for mean estimation and specific distributions of privacy requirements (mixed public-private or two-level), which may not capture all real-world data distributions or query types.","pith_extraction_headline":"For mean estimation, a simple thresholding operator on privacy budgets captures nearly all the utility gains of full personalization."},"references":{"count":32,"sample":[{"doi":"","year":2026,"title":"Personalized differential privacy for ridge regression under output perturbation.Naval Research Logistics (NRL), 73(4):525–537, 2026","work_id":"d074d0c6-09dc-4838-8c26-bea561592e00","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2015,"title":"Heterogeneous Differential Privacy","work_id":"f4a11c32-7a4d-4bc6-b1f2-e99ead412936","ref_index":2,"cited_arxiv_id":"1504.06998","is_internal_anchor":true},{"doi":"","year":2011,"title":"Anita Allen.Unpopular Privacy: What Must We Hide?OUP Usa, New York, US, 2011","work_id":"76e9b7ae-cb87-445c-9807-7293501efd6b","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2019,"title":"Limits of private learning with access to public data","work_id":"c628cea5-0c6f-4e6c-b510-2d0e5d2e328d","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2026,"title":"Data sharing with endogenous choices over differential privacy levels.arXiv preprint arXiv:2602.09357, 2026","work_id":"c119098d-13b9-467f-9098-82687c8e2f77","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":32,"snapshot_sha256":"4c03f9fccbc39faf9eccdb7fd872e9050dfd40466cb4c6f42cd8c04cddc5ba59","internal_anchors":1},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}