## Universally Utility-Maximizing Privacy Mechanisms

### Cached

### Download Links

Citations: | 51 - 1 self |

### BibTeX

@MISC{Ghosh_universallyutility-maximizing,

author = {Arpita Ghosh and Tim Roughgarden and Mukund Sundararajan},

title = {Universally Utility-Maximizing Privacy Mechanisms},

year = {}

}

### OpenURL

### Abstract

A mechanism for releasing information about a statistical database with sensitive data must resolve a trade-off between utility and privacy. Publishing fully accurate information maximizes utility while minimizing privacy, while publishing random noise accomplishes the opposite. Privacy can be rigorously quantified using the framework of differential privacy, which requires that a mechanism’s output distribution is nearly the same whether or not a given database row is included or excluded. The goal of this paper is strong and general utility guarantees, subject to differential privacy. We pursue mechanisms that guarantee near-optimal utility to every potential user, independent of its side information (modeled as a prior distribution over query results) and preferences (modeled via a loss function). Our main result is: for each fixed count query and differential privacy level, there is a geometric mechanism M ∗ — a discrete variant of the simple and well-studied Laplace mechanism — that is simultaneously expected loss-minimizing for every possible user, subject to the differential privacy constraint. This is an extremely strong utility guarantee: every potential user u, no matter what its side information and preferences, derives as much utility from M ∗ as from interacting with a differentially private mechanism Mu that is optimally tailored to u. More precisely, for every user u there is an optimal mecha-

### Citations

1606 | Microeconomic Theory - Mas-Colell, Whinston, et al. - 1995 |

364 |
Introduction to Linear Optimization. Athena Scientific
- Bertsimas, Tsitsiklis
- 1997
(Show Context)
Citation Context ...x(i+1)r ≥ 0 ∀r ∈ N \ {n}, ∀i ∈ N (3) α · xir − x(i+1)r ≤ 0 ∀r ∈ N \ {n}, ∀i ∈ N (4) r∈R xir = 1 ∀i ∈ N (5) xir ≥ 0 ∀i ∈ N, ∀r ∈ N (6) Since the LP is bounded and feasible, we have the following (e.g. =-=[3]-=-). Lemma 5.1 Every user-specific LP has an optimal solution that is a vertex. For the rest of this section, fix a user with prior {pi} and a loss function l(i, r) that is monotone in |i − r| for every... |

341 | Calibrating noise to sensitivity in private data analysis
- Dwork, Mcsherry, et al.
- 2006
(Show Context)
Citation Context ...ion schemes initially thought to be “private” have succumbed to privacy attacks [19, 17, 1], highlighting the urgent need for mechanisms that are provably private. The differential privacy literature =-=[10, 8, 16, 18, 5, 7, 12, 4]-=- has proposed a rigorous and quantifiable definition of privacy, as well as provably privacy-preserving mechanisms for diverse applications including statistical queries, machine learning, and pricing... |

330 | Differential privacy
- Dwork
- 2006
(Show Context)
Citation Context ...ion schemes initially thought to be “private” have succumbed to privacy attacks [19, 17, 1], highlighting the urgent need for mechanisms that are provably private. The differential privacy literature =-=[10, 8, 16, 18, 5, 7, 12, 4]-=- has proposed a rigorous and quantifiable definition of privacy, as well as provably privacy-preserving mechanisms for diverse applications including statistical queries, machine learning, and pricing... |

208 | Revealing information while preserving privacy
- Dinur, Nissim
- 2003
(Show Context)
Citation Context ...ion schemes initially thought to be “private” have succumbed to privacy attacks [19, 17, 1], highlighting the urgent need for mechanisms that are provably private. The differential privacy literature =-=[10, 8, 16, 18, 5, 7, 12, 4]-=- has proposed a rigorous and quantifiable definition of privacy, as well as provably privacy-preserving mechanisms for diverse applications including statistical queries, machine learning, and pricing... |

165 | Practical privacy: the sulq framework
- Blum, Dwork, et al.
- 2005
(Show Context)
Citation Context |

158 | Robust de-anonymization of large sparse datasets
- Narayanan, Shmatikov
- 2008
(Show Context)
Citation Context ...gregate information is sufficiently anonymous so as not to breach the privacy of any individual. Unfortunately, publication schemes initially thought to be “private” have succumbed to privacy attacks =-=[19, 17, 1]-=-, highlighting the urgent need for mechanisms that are provably private. The differential privacy literature [10, 8, 16, 18, 5, 7, 12, 4] has proposed a rigorous and quantifiable definition of privacy... |

153 | Wherefore art thou r3579x?: anonymized social networks, hidden patterns, and structural steganography
- Backstrom, Dwork, et al.
- 2007
(Show Context)
Citation Context ...gregate information is sufficiently anonymous so as not to breach the privacy of any individual. Unfortunately, publication schemes initially thought to be “private” have succumbed to privacy attacks =-=[19, 17, 1]-=-, highlighting the urgent need for mechanisms that are provably private. The differential privacy literature [10, 8, 16, 18, 5, 7, 12, 4] has proposed a rigorous and quantifiable definition of privacy... |

131 | Differential Privacy: A Survey of Results
- Dwork
(Show Context)
Citation Context ...m is almost independent of whether a given user “opted in” or “opted out” of the database [10, 14]. Below we discuss the papers in the differential privacy literature closest to the present work; see =-=[9]-=- for a recent, thorough survey of the state of the field. Dinur and Nissim [7] showed that for a database with n rows, answering O(n log 2 n) randomly chosen subset count queries with o( √ n) error al... |

128 | A learning theory approach to non-interactive database privacy
- Blum, Ligett, et al.
- 2008
(Show Context)
Citation Context |

111 | Smooth sensitivity and sampling in private data analysis
- Nissim, Raskhodnikova, et al.
- 2007
(Show Context)
Citation Context |

109 | K.: Mechanism design via differential privacy
- McSherry, Talwar
(Show Context)
Citation Context |

89 | Privacy-preserving datamining on vertically partitioned databases
- Dwork, Nissim
- 2004
(Show Context)
Citation Context |

62 | What can we learn privately
- Kasiviswanathan, Lee, et al.
- 2011
(Show Context)
Citation Context ...; they obtain non-interactive mechanisms that provide simultaneous good accuracy (in terms of worst-case error) for all count queries from a class with polynomial VC dimension. Kasiviswanathan et al. =-=[13]-=- give further results for privately learning hypotheses from a given class. The use of abstract “utility functions” in McSherry and Talwar [16] has a similar flavor to our use of loss functions, thoug... |

59 |
The price of privacy and the limits of LP decoding
- Dwork, McSherry, et al.
- 2007
(Show Context)
Citation Context ...n rows, answering O(n log 2 n) randomly chosen subset count queries with o( √ n) error allows an adversary to reconstruct most of the rows of the database (a blatant privacy breach); see Dwork et al. =-=[11]-=- for a more robust impossibility result of the same type. Most of the differential privacy literature circumvents these impossibility results by focusing on interactive models where a mechanism suppli... |

39 | Privacy, accuracy, and consistency too: A holistic solution to contingency table release - Barak, Chaudhuri, et al. - 2007 |

20 |
A note on differential privacy: Defining resistance to arbitrary side information
- Kasiviswanathan, Smith
- 2008
(Show Context)
Citation Context ...ver databases a potential attacker has, its posterior after interacting with a differentially private mechanism is almost independent of whether a given user “opted in” or “opted out” of the database =-=[10, 14]-=-. Below we discuss the papers in the differential privacy literature closest to the present work; see [9] for a recent, thorough survey of the state of the field. Dinur and Nissim [7] showed that for ... |

3 |
Census Bureau (2008). Statistical Abstract of the United States Population
- S
(Show Context)
Citation Context ...ial privacy, privacy, utility, linear programming 1. INTRODUCTION Organizations including the census bureau, medical establishments, and Internet companies collect and publish statistical information =-=[6, 19]-=-. The census bureau may, for instance, publish the result of a query such as: “How many individuals have incomes that exceed $100,000?”. An implicit hope in this approach is that aggregate information... |

2 |
search data scandal. http://en.wikipedia.org/wiki/AOL_search_data_ scandal
- AOL
(Show Context)
Citation Context ...ial privacy, privacy, utility, linear programming 1. INTRODUCTION Organizations including the census bureau, medical establishments, and Internet companies collect and publish statistical information =-=[6, 19]-=-. The census bureau may, for instance, publish the result of a query such as: “How many individuals have incomes that exceed $100,000?”. An implicit hope in this approach is that aggregate information... |

2 |
The 2009 Statistical Abstract. http://www.census.gov/compendia/statab/ (accessed 20
- Bureau
- 2009
(Show Context)
Citation Context ...ic mechanism via suitable remappings of its range. 1 Introduction Organizations including the census bureau, medical establishments, and Internet companies collect and publish statistical information =-=[6, 19]-=-. The census bureau may, for instance, publish the result ∗ A preliminary version of this paper appeared in the Proceedings of the 40th Annual Symposium on Theory of Computing, June 2008. † Yahoo! Res... |

1 |
What can we learn privately? http://www.citebase.org/abstract?id=oai:arXiv.org:0803.0924
- Kasiviswanathan, Lee, et al.
- 2008
(Show Context)
Citation Context ...; they obtain non-interactive mechanisms that provide simultaneous good accuracy (in terms of worst-case error) for all count queries from a class with polynomial VC dimension. Kasiviswanathan et al. =-=[12]-=- give further results for privately learning hypotheses from a given class. The use of abstract “utility functions” in McSherry and Talwar [15] has a similar flavor to our use of loss functions, thoug... |