{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2025,8,22]],"date-time":"2025-08-22T05:07:06Z","timestamp":1755839226611},"reference-count":47,"publisher":"Association for Computing Machinery (ACM)","issue":"13","content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":["Proc. VLDB Endow."],"published-print":{"date-parts":[[2021,9]]},"abstract":"<jats:p>\n            Differential privacy allows bounding the influence that training data records have on a machine learning model. To use differential privacy in machine learning, data scientists must choose privacy parameters (\u03f5,\n            <jats:italic>\u03b4<\/jats:italic>\n            ). Choosing meaningful privacy parameters is key, since models trained with weak privacy parameters might result in excessive privacy leakage, while strong privacy parameters might overly degrade model utility. However, privacy parameter values are difficult to choose for two main reasons. First, the theoretical upper bound on privacy loss (\u03f5, \u03b4) might be loose, depending on the chosen sensitivity and data distribution of practical datasets. Second, legal requirements and societal norms for anonymization often refer to individual identifiability, to which (\u03f5,\n            <jats:italic>\u03b4<\/jats:italic>\n            ) are only indirectly related.\n          <\/jats:p>\n          <jats:p>\n            We transform (\u03f5,\n            <jats:italic>\u03b4<\/jats:italic>\n            ) to a bound on the Bayesian posterior belief of the adversary assumed by differential privacy concerning the presence of any record in the training dataset. The bound holds for multidimensional queries under composition, and we show that it can be tight in practice. Furthermore, we derive an identifiability bound, which relates the adversary assumed in differential privacy to previous work on membership inference adversaries. We formulate an implementation of this differential privacy adversary that allows data scientists to audit model training and compute empirical identifiability scores and empirical (\u03f5,\n            <jats:italic>\u03b4<\/jats:italic>\n            ).\n          <\/jats:p>","DOI":"10.14778\/3484224.3484231","type":"journal-article","created":{"date-parts":[[2021,10,28]],"date-time":"2021-10-28T22:36:50Z","timestamp":1635460610000},"page":"3335-3347","source":"Crossref","is-referenced-by-count":6,"title":["Quantifying identifiability to choose and audit \u03f5 in differentially private deep learning"],"prefix":"10.14778","volume":"14","author":[{"given":"Daniel","family":"Bernau","sequence":"first","affiliation":[{"name":"SAP SE, Karlsruhe, Germany"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"G\u00fcnther","family":"Eibl","sequence":"additional","affiliation":[{"name":"Salzburg University of Applied Sciences, Salzburg, Austria"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Philip W.","family":"Grassal","sequence":"additional","affiliation":[{"name":"University of Heidelberg, Heidelberg, Germany"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Hannah","family":"Keller","sequence":"additional","affiliation":[{"name":"SAP SE, Karlsruhe, Germany"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Florian","family":"Kerschbaum","sequence":"additional","affiliation":[{"name":"University of Waterloo, Waterloo, Canada"}],"role":[{"role":"author","vocabulary":"crossref"}]}],"member":"320","published-online":{"date-parts":[[2021,10,28]]},"reference":[{"key":"e_1_2_1_1_1","doi-asserted-by":"publisher","DOI":"10.1145\/2976749.2978318"},{"key":"e_1_2_1_2_1","doi-asserted-by":"publisher","DOI":"10.1257\/aer.20170627"},{"key":"e_1_2_1_3_1","doi-asserted-by":"publisher","DOI":"10.1109\/FOCS.2014.56"},{"key":"e_1_2_1_4_1","doi-asserted-by":"crossref","unstructured":"Daniel Bernau G\u00fcnther Eibl Philip W. Grassal Hannah Keller and Florian Kerschbaum. 2021. Quantifying identifiability to choose and audit e in differentially private deep learning. arXiv:2103.02913 [cs.CR]  Daniel Bernau G\u00fcnther Eibl Philip W. Grassal Hannah Keller and Florian Kerschbaum. 2021. Quantifying identifiability to choose and audit e in differentially private deep learning. arXiv:2103.02913 [cs.CR]","DOI":"10.14778\/3484224.3484231"},{"key":"e_1_2_1_5_1","doi-asserted-by":"publisher","DOI":"10.1007\/978-3-030-81242-3_2"},{"key":"e_1_2_1_6_1","doi-asserted-by":"publisher","DOI":"10.1145\/3372297.3417238"},{"key":"e_1_2_1_7_1","doi-asserted-by":"publisher","DOI":"10.1109\/ICDEW.2013.6547433"},{"key":"e_1_2_1_8_1","volume-title":"UCI Machine Learning Repository. Retrieved","author":"Dua Dheeru","year":"2021"},{"key":"e_1_2_1_9_1","doi-asserted-by":"publisher","DOI":"10.1007\/11787006_1"},{"key":"e_1_2_1_10_1","doi-asserted-by":"publisher","DOI":"10.1007\/11761679_29"},{"key":"e_1_2_1_11_1","doi-asserted-by":"publisher","DOI":"10.1561\/0400000042"},{"key":"e_1_2_1_12_1","volume-title":"Rothblum","author":"Dwork Cynthia","year":"2016"},{"key":"e_1_2_1_13_1","doi-asserted-by":"publisher","DOI":"10.1109\/FOCS.2010.12"},{"key":"e_1_2_1_14_1","doi-asserted-by":"publisher","DOI":"10.1186\/s42162-018-0025-3"},{"key":"e_1_2_1_15_1","volume-title":"Proceedings of the Conference on Learning Representations (ICLR). IEEE Computer Society","author":"Goodfellow Ian","year":"2015"},{"key":"e_1_2_1_16_1","doi-asserted-by":"publisher","DOI":"10.2478\/popets-2019-0008"},{"key":"e_1_2_1_17_1","doi-asserted-by":"publisher","DOI":"10.1109\/CSF.2014.35"},{"key":"e_1_2_1_18_1","unstructured":"Thomas Humphries Matthew Rafuse Lindsey Tulloch Simon Oya Ian Goldberg and Florian Kerschbaum. 2020. Differentially Private Learning Does Not Bound Membership Inference. arXiv:2010.12112 [cs.CR]  Thomas Humphries Matthew Rafuse Lindsey Tulloch Simon Oya Ian Goldberg and Florian Kerschbaum. 2020. Differentially Private Learning Does Not Bound Membership Inference. arXiv:2010.12112 [cs.CR]"},{"key":"e_1_2_1_19_1","volume-title":"Proceedings of the Conference on Advances in Neural Information Processing Systems (NeurIPS). Curran Associates Inc.","author":"Jagielski Matthew","year":"2020"},{"key":"e_1_2_1_20_1","doi-asserted-by":"publisher","DOI":"10.5555\/3361338.3361469"},{"key":"e_1_2_1_21_1","unstructured":"Bargav Jayaraman Lingxiao Wang Katherine Knipmeyer Quanquan Gu and David Evans. 2020. Revisiting Membership Inference Under Realistic Assumptions. arXiv:2005.10881 [cs.CR]  Bargav Jayaraman Lingxiao Wang Katherine Knipmeyer Quanquan Gu and David Evans. 2020. Revisiting Membership Inference Under Realistic Assumptions. arXiv:2005.10881 [cs.CR]"},{"key":"e_1_2_1_22_1","doi-asserted-by":"publisher","DOI":"10.5555\/3045118.3045265"},{"key":"e_1_2_1_23_1","doi-asserted-by":"publisher","DOI":"10.1109\/TIT.2017.2685505"},{"key":"e_1_2_1_24_1","doi-asserted-by":"publisher","DOI":"10.29012\/jpc.v6i1.634"},{"key":"e_1_2_1_25_1","doi-asserted-by":"publisher","DOI":"10.5555\/3001460.3001502"},{"key":"e_1_2_1_26_1","doi-asserted-by":"publisher","DOI":"10.5555\/2051002.2051032"},{"key":"e_1_2_1_27_1","doi-asserted-by":"publisher","DOI":"10.1145\/2339530.2339695"},{"key":"e_1_2_1_28_1","doi-asserted-by":"publisher","DOI":"10.1145\/2508859.2516686"},{"key":"e_1_2_1_29_1","volume-title":"Bibby","author":"Mardia Kantilal Vardichand","year":"1979"},{"key":"e_1_2_1_30_1","volume-title":"Proceedings of the International Conference on Learning Representations (ICLR). OpenReview.net.","author":"McMahan Brendan","year":"2018"},{"key":"e_1_2_1_31_1","unstructured":"H. Brendan McMahan Galen Andrew Ulfar Erlingsson Steve Chien Ilya Mironov Nicolas Papernot and Peter Kairouz. 2019. A General Approach to Adding Differential Privacy to Iterative Training Procedures. arXiv:1812.06210 [cs.LG]  H. Brendan McMahan Galen Andrew Ulfar Erlingsson Steve Chien Ilya Mironov Nicolas Papernot and Peter Kairouz. 2019. A General Approach to Adding Differential Privacy to Iterative Training Procedures. arXiv:1812.06210 [cs.LG]"},{"key":"e_1_2_1_32_1","volume-title":"Proceedings of the Computer Security Foundations Symposium (CSF). IEEE Computer Society","author":"Mivonov Ilya","year":"2017"},{"key":"e_1_2_1_33_1","volume-title":"Proceedings of the Symposium on Security and Privacy (S&P). IEEE Computer Society","author":"Nasr Milad","year":"2019"},{"key":"e_1_2_1_34_1","volume-title":"Differential Privacy in Context: Conceptual and Ethical Considerations. In Four Facets of Differential Privacy Symposium. Presented at the Institute for Advanced Study","author":"Nissenbaum Helen","year":"2016"},{"key":"e_1_2_1_35_1","doi-asserted-by":"publisher","DOI":"10.1145\/1250790.1250803"},{"key":"e_1_2_1_36_1","article-title":"Is privacy privacy","volume":"376","author":"Nissim Kobbi","year":"2018","journal-title":"Philosophical Transactions of the Royal Society"},{"key":"e_1_2_1_37_1","volume-title":"Guidance Regarding Methods for De-identification of Protected Health Information in Accordance with the Health Insurance Portability and Accountability Act (HIPAA) Privacy Rule. Retrieved","author":"American Department of Health and Human Services. 2010.","year":"2021"},{"key":"e_1_2_1_38_1","volume-title":"General Data Protection Regulation. Official Journal of the European Union L119 59, 1 (May","author":"European Parliament and Council of the European Union. 2016.","year":"2016"},{"key":"e_1_2_1_39_1","unstructured":"Article 29 Data Protection Working Party. 2014. Opinion 05\/2014 on Anonymisation Techniques. Retrieved September 2 2021 from https:\/\/ec.europa.eu\/justice\/article-29\/documentation\/opinion-recommendation\/files\/2014\/wp216_en.pdf  Article 29 Data Protection Working Party. 2014. Opinion 05\/2014 on Anonymisation Techniques. Retrieved September 2 2021 from https:\/\/ec.europa.eu\/justice\/article-29\/documentation\/opinion-recommendation\/files\/2014\/wp216_en.pdf"},{"key":"e_1_2_1_40_1","first-page":"61","article-title":"Membership Inference Attack against Differentially Private Deep Learning Model","volume":"11","author":"Atiqur Rahman Md.","year":"2018","journal-title":"Transactions on Data Privacy"},{"key":"e_1_2_1_41_1","doi-asserted-by":"publisher","DOI":"10.1145\/2810103.2813687"},{"key":"e_1_2_1_42_1","doi-asserted-by":"publisher","DOI":"10.1109\/SP.2017.41"},{"key":"e_1_2_1_43_1","volume-title":"Proceedings of the Global Conference on Signal and Information Processing (GlobalSIP). IEEE Computer Society","author":"Song Shuang"},{"key":"e_1_2_1_44_1","unstructured":"Om Thakkar Galen Andrew and H. Brendan McMahan. 2019. Differentially Private Learning with Adaptive Clipping. arXiv:1905.03871 [cs.LG]  Om Thakkar Galen Andrew and H. Brendan McMahan. 2019. Differentially Private Learning with Adaptive Clipping. arXiv:1905.03871 [cs.LG]"},{"key":"e_1_2_1_45_1","doi-asserted-by":"publisher","DOI":"10.1109\/ICPR.1996.546912"},{"key":"e_1_2_1_46_1","volume-title":"Proceedings of the Symposium on Information Theory (ISIT). IEEE Computer Society","author":"van Erven Tim","year":"2010"},{"key":"e_1_2_1_47_1","doi-asserted-by":"publisher","DOI":"10.1109\/CSF.2018.00027"}],"container-title":["Proceedings of the VLDB Endowment"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/dl.acm.org\/doi\/pdf\/10.14778\/3484224.3484231","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2022,12,28]],"date-time":"2022-12-28T09:38:06Z","timestamp":1672220286000},"score":1,"resource":{"primary":{"URL":"https:\/\/dl.acm.org\/doi\/10.14778\/3484224.3484231"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2021,9]]},"references-count":47,"journal-issue":{"issue":"13","published-print":{"date-parts":[[2021,9]]}},"alternative-id":["10.14778\/3484224.3484231"],"URL":"https:\/\/doi.org\/10.14778\/3484224.3484231","relation":{},"ISSN":["2150-8097"],"issn-type":[{"value":"2150-8097","type":"print"}],"subject":[],"published":{"date-parts":[[2021,9]]}}}