A fruitful direction for future data mining research will be the development of techniques that incorporate privacy concerns. Specifically, we address the following question. Since the primary task in data mining is the development of models about aggregated data, can we develop accurate models without access to precise information in individual data records? We consider the concrete case of building a decision-tree classifier from tredning data in which the values of individual records have been perturbed. The resulting data records look very different from the original records and the distribution of data values is also very different from the original distribution. While it is not possible to accurately estimate original values in individual data records, we propose a-novel reconstruction procedure to accurately estimate the distribution of original data values. By using these reconstructed distributions, we are able to build classifiers whose accuracy is comparable to the accuracy of classifiers built with the original data.

Publishing data about individuals without revealing sensitive information about them is an important problem. In recent years, a new definition of privacy called k-anonymity has gained popularity. In a k-anonymized dataset, each record is indistinguishable from at least k − 1 other records with respect to certain “identifying ” attributes. In this paper we show using two simple attacks that a k-anonymized dataset has some subtle, but severe privacy problems. First, an attacker can discover the values of sensitive attributes when there is little diversity in those sensitive attributes. This kind of attack is a known problem [60]. Second, attackers often have background knowledge, and we show that k-anonymity does not guarantee privacy against attackers using background knowledge. We give a detailed analysis of these two attacks and we propose a novel and powerful privacy criterion called ℓ-diversity that can defend against such attacks. In addition to building a formal foundation for ℓ-diversity, we show in an experimental evaluation that ℓ-diversity is practical and can be implemented efficiently. 1.

This paper considers the problem of providing security to statistical databases against disclosure of confidential information. Security-control methods suggested in the literature are classified into four general approaches: conceptual, query restriction, data perturbation, and output perturbation. Criteria for evaluating the performance of the various security-control methods are identified. Security-control methods that are based on each of the four approaches are discussed, together with their performance with respect to the identified evaluation criteria. A detailed comparative analysis of the most promising methods for protecting dynamic-online statistical databases is also presented. To date no single security-control method prevents both exact and partial disclosures. There are, however, a few perturbation-based methods that prevent exact disclosure and enable the database administrator to exercise “statistical disclosure control. ” Some of these methods, however introduce bias into query responses or suffer from the O/l query-set-size problem (i.e., partial disclosure is possible in case of null query set or a query set of size 1). We recommend directing future research efforts toward developing new methods that prevent exact disclosure and provide statistical-disclosure control, while at the same time do not suffer from the bias problem and the O/l query-set-size problem. Furthermore, efforts directed toward developing a bias-correction mechanism and solving the general problem of small query-set-size would help salvage a few of the current perturbation-based methods.

The Hippocratic Oath has guided the conduct of physicians for centuries. Inspired by its tenet of preserving privacy, we argue that future database systems must include responsibility for the privacy of data they manage as a founding tenet. We enunciate the key privacy principles for such Hippocratic database systems. We propose a strawman design for Hippocratic databases, identify the technical challenges and problems in designing such databases, and suggest some approaches that may lead to solutions. Our hope is that this paper will serve to catalyze a fruitful and exciting direction for future database research. 1

We examine the tradeoff between privacy and usability of statistical databases. We model a statistical database by an n-bit string d1,.., dn, with a query being a subset q ⊆ [n] to be answered by � i∈q di. Our main result is a polynomial reconstruction algorithm of data from noisy (perturbed) subset sums. Applying this reconstruction algorithm to statistical databases we show that in order to achieve privacy one has to add perturbation of magnitude Ω ( √ n). That is, smaller perturbation always results in a strong violation of privacy. We show that this result is tight by exemplifying access algorithms for statistical databases that preserve privacy while adding perturbation of magnitude Õ(√n). For time-T bounded adversaries we demonstrate a privacy-preserving access algorithm whose perturbation magnitude is ≈ √ T. 1

Abstract. Over the past five years a new approach to privacy-preserving

Abstract. Data anonymization techniques based on the k-anonymity model have been the focus of intense research in the last few years. Although the k-anonymity model and the related techniques provide valuable solutions to data privacy, current solutions are limited only to static data release (i.e., the entire dataset is assumed to be available at the time of release). While this may be acceptable in some applications, today we see databases continuously growing everyday and even every hour. In such dynamic environments, the current techniques may suffer from poor data quality and/or vulnerability to inference. In this paper, we analyze various inference channels that may exist in multiple anonymized datasets and discuss how to avoid such inferences. We then present an approach to securely anonymizing a continuously growing dataset in an efficient manner while assuring high data quality. 1

Today's globally networked society places great demand on the dissemination and sharing of person-specific data for many new and exciting uses. Even situations where aggregate statistical information was once the reporting norm now rely heavily on the transfer of microscopically detailed transaction and encounter information. This happens at a time when more and more historically public information is also electronically available. When these data are linked together, they provide an electronic shadow of a person or organization that is as identifying and personal as a fingerprint even when the information contains no explicit identifiers, such as name and phone number. Other distinctive data, such as birth date and ZIP code, often combine uniquely and can be linked to publicly available information to re-identify individuals. Producing anonymous data that remains specific enough to be useful is often a very difficult task and practice today tends to either incorrectly believe confidentiality is maintained when it is not or produces data that are practically useless. The goal of the work presented in this book is to explore computational techniques for releasing useful information in such a way that the identity of any individual or entity contained in data cannot be recognized while the data remain practically useful. I begin by demonstrating ways to learn information about entities from publicly available information. I then provide a formal framework for reasoning about disclosure control and the ability to infer the identities of entities contained within the data. I formally define and present null-map, k-map and wrong-map as models of protection. Each model provides protection by ensuring that released information maps to no, k or incorrect entities, respectively. The book ends by examining four computational systems that attempt to maintain privacy while releasing electronic information. These systems are: (1) my Scrub System, which locates personally-identifying information in letters between doctors and notes written by clinicians; (2) my Datafly II System, which generalizes and suppresses values in field-structured data sets; (3) Statistics Netherlands' m-Argus System, which is becoming a European standard for producing public-use data; and, (4) my k-Similar algorithm, which finds optimal solutions such that data are minimally distorted while still providing adequate protection. By introducing anonymity and quality metrics, I show that Datafly II can overprotect data, Scrub and m-Argus can fail to provide adequate protection, but k-similar finds optimal results.

The contingency table is a work horse of official statistics, the format of reported data for the US Census, Bureau of Labor Statistics, and the Internal Revenue Service. In many settings such as these privacy is not only ethically mandated, but frequently legally as well. Consequently there is an extensive and diverse literature dedicated to the problems of statistical disclosure control in contingency table release. However, all current techniques for reporting contingency tables fall short on at least one of privacy, accuracy, and consistency (among multiple released tables). We propose a solution that provides strong guarantees for all three desiderata simultaneously. Our approach can be viewed as a special case of a more general approach for producing synthetic data: Any privacypreserving mechanism for contingency table release begins with raw data and produces a (possibly inconsistent) privacypreserving set of marginals. From these tables alone – and hence without weakening privacy – we will find and output the “nearest ” consistent set of marginals. Interestingly, this set is no farther than the tables of the raw data, and consequently the additional error introduced by the imposition of consistency is no more than the error introduced by the privacy mechanism itself. The privacy mechanism of [20] gives the strongest known privacy guarantees, with very little error. Combined with the techniques of the current paper, we therefore obtain excellent privacy, accuracy, and consistency among the tables. Moreover, our techniques are surprisingly efficient. Our techniques apply equally well to the logical cousin of the contingency table, the OLAP cube.

Abstract. k-anonymization techniques have been the focus of intense research in the last few years. An important requirement for such techniques is to ensure anonymization of data while at the same time minimizing the information loss resulting from data modifications. In this paper we propose an approach that uses the idea of clustering to minimize information loss and thus ensure good data quality. The key observation here is that data records that are naturally similar to each other should be part of the same equivalence class. We thus formulate a specific clustering problem, referred to as k-member clustering problem. We prove that this problem is NP-hard and present a greedy heuristic, the complexity of which is in O(n 2). As part of our approach we develop a suitable metric to estimate the information loss introduced by generalizations, which works for both numeric and categorical data. 1