Adversary Instantiation: Lower Bounds for Differentially Private Machine Learning
Open accessThe DOI matches its Crossref record; 24 of the 24 references it cites were found in a registry or an index.
Abstract
Differentially private (DP) machine learning allows us to train models on private data while limiting data leakage. DP formalizes this data leakage through a cryptographic game, where an adversary must predict if a model was trained on a dataset D, or a dataset D′ that differs in just one example. If observing the training algorithm does not meaningfully increase the adversary's odds of successfully guessing which dataset the model was trained on, then the algorithm is said to be differentially private. Hence, the purpose of privacy analysis is to upper bound the probability that any adversary could successfully guess which dataset the model was trained on.In our paper, we instantiate this hypothetical adversary in order to establish lower bounds on the probability that this distinguishing game can be won. We use this adversary to evaluate the importance of the adversary capabilities allowed in the privacy analysis of DP training algorithms.For DP-SGD, the most common method for training neural networks with differential privacy, our lower bounds are tight and match the theoretical upper bound. This implies that in order to prove better upper bounds, it will be necessary to make use of additional assumptions. Fortunately, we find that our attacks are significantly weaker when additional (realistic) restrictions are put in place on the adversary's capabilities. Thus, in the practical setting common to many real-world deployments, there is a gap between our lower bounds and the upper bounds provided by the analysis: differential privacy is conservative and adversaries may not be able to leak as much information as suggested by the theoretical bound.
Overview
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
Gradient-based learning applied to document recognition
Lecun, Y., Bottou, L., Bengio, Y., & Haffner, P. (1998) · Proceedings of the IEEE · doi:10.1109/5.726791 · Cited by 58,805
Dermatologist-level classification of skin cancer with deep neural networks
Esteva, A., Kuprel, B., Novoa, R. A., Ko, J., Swetter, S. M., Blau, H. M., & Thrun, S. (2017) · Nature · doi:10.1038/nature21056 · Cited by 14,362
k-ANONYMITY: A MODEL FOR PROTECTING PRIVACY
SWEENEY, L. (2002) · International Journal of Uncertainty, Fuzziness and Knowledge-Based Systems · doi:10.1142/s0218488502001648 · Cited by 8,571 · cited 2×
“Indeed, in order to provide strong, composable guarantees that avoid the pitfalls of other privacy analysis methods [35, 36, 52], DP and DP-SGD assume the existence of powerful adversaries that may not be practically realizable.”
BackgroundCalibrating Noise to Sensitivity in Private Data Analysis
Dwork, C., McSherry, F., Nissim, K., & Smith, A. (2006) · Lecture Notes in Computer Science · doi:10.1007/11681878_14 · Cited by 7,272
The Algorithmic Foundations of Differential Privacy
Dwork, C., & Roth, A. (2014) · Foundations and Trends® in Theoretical Computer Science · doi:10.1561/0400000042 · Cited by 5,557
DeepFool: A Simple and Accurate Method to Fool Deep Neural Networks
Moosavi-Dezfooli, S.-M., Fawzi, A., & Frossard, P. (2016) · 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2016.282 · Cited by 5,317 · cited 1×
“Next, the adversary generates an input with an adversarial example algorithm [42].”
MethodMembership Inference Attacks Against Machine Learning Models
Shokri, R., Stronati, M., Song, C., & Shmatikov, V. (2017) · 2017 IEEE Symposium on Security and Privacy (SP) · doi:10.1109/sp.2017.41 · Cited by 4,754 · cited 3×
“As a first example of how our adaptive analysis approach proceeds, we consider the baseline adversary who mounts the well-researched membership inference attack [49] in a setting where they have access to an API.”
BackgroundMethodThe Algorithmic Foundations of Differential Privacy
Dwork, C., & Roth, A. (2013) · doi:10.1561/9781601988195 · Cited by 3,856
L -diversity: Privacy beyond k -anonymity
Machanavajjhala, A., Kifer, D., Gehrke, J., & Venkitasubramaniam, M. (2007) · ACM Transactions on Knowledge Discovery from Data · doi:10.1145/1217299.1217302 · Cited by 3,617 · cited 1×
“Indeed, in order to provide strong, composable guarantees that avoid the pitfalls of other privacy analysis methods [35, 36, 52], DP and DP-SGD assume the existence of powerful adversaries that may not be practically realizable.”
BackgroundDifferential Privacy: A Survey of Results
Dwork, C. (2008) · Lecture Notes in Computer Science · doi:10.1007/978-3-540-79228-4_1 · Cited by 3,542 · cited 1×
“Differential privacy (DP) is now the standard definition for privacy [14, 15].”
Background
Cited by
Membership Inference Attacks on Machine Learning: A Survey
Hu, H., Salcic, Z., Sun, L., Dobbie, G., Yu, P. S., & Zhang, X. (2022) · ACM Computing Surveys · doi:10.1145/3523273 · Cited by 564
Membership Inference Attacks From First Principles
Carlini, N., Chien, S., Nasr, M., Song, S., Terzis, A., & Tramèr, F. (2022) · 2022 IEEE Symposium on Security and Privacy (SP) · doi:10.1109/sp46214.2022.9833649 · Cited by 502
Self-citationWhen Machine Unlearning Jeopardizes Privacy
Chen, M., Zhang, Z., Wang, T., Backes, M., Humbert, M., & Zhang, Y. (2021) · Proceedings of the 2021 ACM SIGSAC Conference on Computer and Communications Security · doi:10.1145/3460120.3484756 · Cited by 195
“DP [9, 16, 39, 49, 55] guarantees that any single data sample in a dataset has limited impact on the output.”
BackgroundHow to DP-fy ML: A Practical Guide to Machine Learning with Differential Privacy
Ponomareva, N., Hazimeh, H., Kurakin, A., Xu, Z., Denison, C., McMahan, H. B., Vassilvitskii, S., Chien, S., & Thakurta, A. G. (2023) · Journal of Artificial Intelligence Research · doi:10.1613/jair.1.14649 · Cited by 166
“Recent work on empirical privacy auditing has shown that a strong adversary adversary that better matches the assumptions of DP can construct attacks that are almost as successful as the lower bound ε would predict (that is, you really need a small ε to protect against these attacks) (Nasr et al., 2021).”
BackgroundSelf-citationWhat Does it Mean for a Language Model to Preserve Privacy?
Brown, H., Lee, K., Mireshghallah, F., Shokri, R., & Tramèr, F. (2022) · 2022 ACM Conference on Fairness Accountability and Transparency · doi:10.1145/3531146.3534642 · Cited by 165
Metadata
- Authors
- Nasr, M., Songi, S., Thakurta, A., Papernot, N., & Carlin, N.
- Venue
- 2021 IEEE Symposium on Security and Privacy (SP)
- Publisher
- IEEE
- Published
- 2021-05
- Pages
- 866-882
- ISSN
- 1081-6011 (ISSN-L)
- OpenAlex
- W3120806613
- Cited by
- 135
- References
- 24
- FWCI
- 11.11
- Type
- conference paper
- Language
- en
- Open access
- green
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Adversary Instantiation: Lower Bounds for Differentially Private Machine Learning. https://app.cochat.ai/papers/10.1109/sp40001.2021.00069
- The paper
- Nasr, M., Songi, S., Thakurta, A., Papernot, N., & Carlin, N. (2021). Adversary Instantiation: Lower Bounds for Differentially Private Machine Learning. 2021 IEEE Symposium on Security and Privacy (SP). https://doi.org/10.1109/sp40001.2021.00069
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.