What Can We Learn Privately?
The DOI matches its Crossref record; 28 of the 53 references it cites were found in a registry or an index, and 25 are not checked yet.
Abstract
Learning problems form an important category of computational tasks that generalizes many of the computations researchers apply to large real-life data sets. We ask, What concept classes can be learned privately, namely, by an algorithm whose output does not depend too heavily on any one input or specific training example? More precisely, we investigate learning algorithms that satisfy differential privacy, a notion that provides strong confidentiality guarantees in contexts where aggregate information is released about a database containing sensitive information about individuals. Our goal is a broad understanding of the resources required for private learning in terms of samples, computation time, and interaction. We demonstrate that, ignoring computational constraints, it is possible to privately agnostically learn any concept class using a sample size approximately logarithmic in the cardinality of the concept class. Therefore, almost anything learnable is learnable privately: specifically, if a concept class is learnable by a (nonprivate) algorithm with polynomial sample complexity and output size, then it can be learned privately using a polynomial number of samples. We also present a computationally efficient private probabilistically approximately correct learner for the class of parity functions. This result dispels the similarity between learning with noise and private learning (both must be robust to small changes in inputs), since parity is thought to be very hard to learn given random classification noise. Local (or randomized response) algorithms are a practical class of private algorithms that have received extensive investigation. We provide a precise characterization of local private learning algorithms. We show that a concept class is learnable by a local algorithm if and only if it is learnable in the statistical query (SQ) model. Therefore, for local private learning algorithms, the similarity to learning with noise is stronger: local learning is equivalent to SQ learning, and SQ algorithms include most known noise-tolerant learning algorithms. Finally, we present a separation between the power of interactive and noninteractive local learning algorithms. Because of the equivalence to SQ learning, this result also separates adaptive and nonadaptive SQ learning.
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.
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The check wasn’t confident either way, so it gives no verdict — never disputed.
In the citing paper
“Comparing these two results, we see that minibatch SGD attains the optimal 'statistical' term ( σ / √ TKM ), whilst SGD on a single device (ignoring the updates of the other devices) achieves the optimal 'optimization' term ( H / ( TK ) 2 ).”
From Advances and Open Problems in Federated Learning (2021) · cites this paper as [294]
Experimental — accuracy not yet measured
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
Calibrating 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
Self-citationDwork, C. (2006) · Lecture Notes in Computer Science · doi:10.1007/11787006_1 · Cited by 5,291 · cited 2×
“A recent line of work, initiated by Dinur and Nissim [15] and called private data analysis, seeks to place data privacy on firmer theoretical foundations and has been successful at formulating a strong, yet attainable privacy definition.”
Probability Inequalities for Sums of Bounded Random Variables
Hoeffding, W. (1963) · Journal of the American Statistical Association · doi:10.1080/01621459.1963.10500830 · Cited by 4,852
A Measure of Asymptotic Efficiency for Tests of a Hypothesis Based on the sum of Observations
Chernoff, H. (1952) · The Annals of Mathematical Statistics · doi:10.1214/aoms/1177729330 · Cited by 3,550
Valiant, L. G. (1984) · Communications of the ACM · doi:10.1145/1968.1972 · Cited by 3,279 · cited 1×
“We examine two basic notions from learning: Valiant's probabilistically approximately correct (PAC) learning [39] model and Kearns' statistical query (SQ) model [29].”
Randomized Response: A Survey Technique for Eliminating Evasive Answer Bias
Warner, S. L. (1965) · Journal of the American Statistical Association · doi:10.1080/01621459.1965.10480775 · Cited by 3,255 · cited 1×
“The difficulty of constructing private learners lies in finding a good hypothesis in what is typically an exponentially large space.”
Privacy-preserving data mining
Agrawal, R., & Srikant, R. (2000) · ACM SIGMOD Record · doi:10.1145/335191.335438 · Cited by 2,961
On lattices, learning with errors, random linear codes, and cryptography
Regev, O. (2009) · Journal of the ACM · doi:10.1145/1568318.1568324 · Cited by 2,315 · cited 1×
“More compellingly , learning parity with noise is thought to be computationally hard, and has been used as the basis of several cryptographic primitives (e.g., [11] [25] [4] [38]).”
Our Data, Ourselves: Privacy Via Distributed Noise Generation
Dwork, C., Kenthapadi, K., McSherry, F., Mironov, I., & Naor, M. (2006) · Lecture Notes in Computer Science · doi:10.1007/11761679_29 · Cited by 1,835 · cited 1×
“A recent line of work, initiated by Dinur and Nissim [15] and called private data analysis, seeks to place data privacy on firmer theoretical foundations and has been successful at formulating a strong, yet attainable privacy definition.”
Privacy-preserving data mining
Agrawal, R., & Srikant, R. (2000) · Proceedings of the 2000 ACM SIGMOD international conference on Management of data · doi:10.1145/342009.335438 · Cited by 1,719
Research works printed but not linked: 9
- S. Ben-David, D. Pál, and H.U. Simon, Stability of k -means clustering , in COLT '07:Proceedings of the 20th Annual Conference on Learning Theory, N. H. Bshouty and C. Gentile, eds., Springer, Berlin, 2007, pp. 20–34. · doi:10.1007/978-3-540-72927-3_4
- Bousquet O., 2002, J. Mach. Learn. Res., 2, 499
- Bshouty N. H., 2002, J. Mach. Learn. Res., 2, 359
- Elisseeff A., 2005, J. Mach. Learn. Res., 6, 55
- S. P. Kasiviswanathan and A. Smith, A Note on Differential Privacy: Defining Resistance to Arbitrary Side Information , Preprint arXiv:0803.39461 [cs.CR], 2008.
- -1S. Kutin and P. Niyogi, Almost-everywhere algorithmic stability and generalization error , in Proceedings of UAI, Association for Uncertainty in Artificial Intelligence, 2002, pp. 275–282.
- T. Moran and M. Naor, Polling with physical envelopes: A rigorous analysis of a human-centric protocol , in Advances in Cryptology, Lecture Notes in Comput. Sci. 4004, Springer, Berlin, 2006, pp. 88–108. · doi:10.1007/11761679_7
- V. Rastogi, S. Hong, and D. Suciu, The boundary between privacy and utility in data publishing , in Proceedings of VLDB '07, VLDB Endowment, Vienna, Austria, 2007, pp. 531–542.
- 10.2307/1403910 · doi:10.2307/1403910
Cited by
Items 1,001–1,005 of 1,005
Differentially Private Parameter Estimation: Optimal Noise Insertion and Data Owner Selection
Cao, X., & Başar, T. (2020) · IEEE Conference on Decision and Control (CDC) · doi:10.1109/cdc42340.2020.9304375 · Cited by 2
“Private machine learning, initially studied in [6], was extensively investigated in the literature [7]–[10], and privacy related mechanism design (e.g., selling private data [11]) was also well studied [12]–[16].”
BackgroundDiscrete Distribution Estimation with Local Differential Privacy: A Comparative Analysis
Le, B. D., & Zia, T. A. (2021) · IEEE International Conference on Pervasive Computing and Communications Workshops (PerCom Workshops) · doi:10.1109/percomworkshops51409.2021.9431028 · Cited by 2
“The (frequency) distribution of the user values is a vector f = (fv1 , fv2 , .., fvd) with fvj being c(vj)/n. Discrete distribution estimation algorithms with LDP perturb user values at user locations and send the privatized data to the data aggregator for distribution aggregation.”
BackgroundInfluentialPrivacy-Preserving SGD on Shuffle Model
Zhang, L., & Zhang, H. F. (2023) · Journal of Mathematics · doi:10.1155/2023/4055950 · Cited by 2
Personalized Differential Privacy in the Shuffle Model
Yang, R., Yang, H. Y., Fan, J., Dong, C., Pang, Y., Wong, D. S., & Wang, S. (2024) · Lecture notes in computer science · doi:10.1007/978-981-99-9785-5_33 · Cited by 2
On Using Secure Aggregation in Differentially Private Federated Learning with Multiple Local Steps
Heikkilä, M. A. (2024) · arXiv · doi:10.48550/arxiv.2407.19286
“This is especially true in the local DP (LDP) setting, where each client protects its own data independently of any other party (Kasiviswanathan et al., 2008).”
Metadata
- Authors
- Kasiviswanathan, S. P., Lee, H. K., Nissim, K., Raskhodnikova, S., & Smith, A.
- Venue
- SIAM Journal on Computing
- Published
- 2011-01
- Volume
- 40
- Issue
- 3
- Pages
- 793-826
- ISSN
- 0097-5397 (ISSN-L)
- 1095-7111
- OpenAlex
- W2245160765
- Cited by
- 898
- References
- 20
- Citations by year
- 2026: 55 · 2025: 93 · 2024: 130 · 2023: 102 · 2022: 102 · 2021: 107 · 2020: 107 · 2019: 63 · 2018: 27 · 2017: 24 · 2016: 33 · 2015: 20 · 2014: 18 · 2013: 11 · 2012: 2
- FWCI
- 9.70
- Type
- journal article
- Language
- en
- Open access
- closed
- Locations
- SIAM Journal on Computing (publishedVersion)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Privacy-Preserving Technologies in Data
- Physical Sciences › Computer Science › Computer Networks and Communications › Cryptography and Data Security
- Physical Sciences › Computer Science › Artificial Intelligence › Machine Learning and Algorithms
- Keywords
- private learning, differential privacy, agnostic learning, sample complexity, parity function, statistical query model
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
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.