A workflow for differentially-private graph synthesis

Proserpio, Davide; Goldberg, Sharon; McSherry, Frank

doi:10.1145/2342549.2342553

Cited by 47 publications

(40 citation statements)

References 12 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Most research in this direction [21,20,24] projects an input graph to dKseries and ensures differential privacy on dK-series statistics. These private statistics are then either fed into graph generators or used by MCMC to generate a fitting synthetic graph.…”

Section: Related Workmentioning

confidence: 99%

Differentially private network data release via structural inference

Xiao

Chen

Tan

2014

Proceedings of the 20th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining

128

104

View full text Add to dashboard Cite

Information networks, such as social media and email networks, often contain sensitive information. Releasing such network data could seriously jeopardize individual privacy. Therefore, we need to sanitize network data before the release. In this paper, we present a novel data sanitization solution that infers a network's structure in a differentially private manner. We observe that, by estimating the connection probabilities between vertices instead of considering the observed edges directly, the noise scale enforced by differential privacy can be greatly reduced. Our proposed method infers the network structure by using a statistical hierarchical random graph (HRG) model. The guarantee of differential privacy is achieved by sampling possible HRG structures in the model space via Markov chain Monte Carlo (MCMC). We theoretically prove that the sensitivity of such inference is only O(log n), where n is the number of vertices in a network. This bound implies less noise to be injected than those of existing works. We experimentally evaluate our approach on four real-life network datasets and show that our solution effectively preserves essential network structural properties like degree distribution, shortest path length distribution and influential nodes.

show abstract

Section: Related Workmentioning

confidence: 99%

Differentially private network data release via structural inference

Xiao

Chen

Tan

2014

Proceedings of the 20th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining

128

104

View full text Add to dashboard Cite

show abstract

“…Grouping has been recently used for differentially-private publication of graph topologies [22,23]. These solutions divide a dataset into disjoint groups, which is different from our interleaved grouping technique.…”

Section: Related Workmentioning

confidence: 99%

Efficient Privacy-Preserving Stream Aggregation in Mobile Sensing with Low Aggregation Error

Cao

2013

Privacy Enhancing Technologies

View full text Add to dashboard Cite

Abstract. Aggregate statistics computed from time-series data contributed by individual mobile nodes can be very useful for many mobile sensing applications. Since the data from individual node may be privacy-sensitive, the aggregator should only learn the desired statistics without compromising the privacy of each node. To provide strong privacy guarantee, existing approaches add noise to each node's data and allow the aggregator to get a noisy sum aggregate. However, these approaches either have high computation cost, high communication overhead when nodes join and leave, or accumulate a large noise in the sum aggregate which means high aggregation error. In this paper, we propose a scheme for privacy-preserving aggregation of time-series data in presence of untrusted aggregator, which provides differential privacy for the sum aggregate. It leverages a novel ring-based interleaved grouping technique to efficiently deal with dynamic joins and leaves and achieve low aggregation error. Specifically, when a node joins or leaves, only a small number of nodes need to update their cryptographic keys. Also, the nodes only collectively add a small noise to the sum to ensure differential privacy, which is O(1) with respect to the number of nodes. Based on symmetric-key cryptography, our scheme is very efficient in computation.

show abstract

“…End-users need to process sanitized data to extract query answers and build predictive models while taking advantage of probabilistic knowledge and constraints that are known to hold (e.g., [22,23,46,7,24,31,38,49,36,3]). Since the ability of a sanitizing algorithm to preserve information should (according to the axioms) be measured as the expected error of a Bayesian decision maker, it stands to reason that this Bayesian methodology should play a more prominent role in the analysis of sanitized data…”

Section: Processing Sanitized Datamentioning

confidence: 99%

Information preservation in statistical privacy and bayesian estimation of unattributed histograms

Lin

Kifer

2013

Proceedings of the 2013 ACM SIGMOD International Conference on Management of Data

View full text Add to dashboard Cite

In statistical privacy, utility refers to two concepts: information preservation -how much statistical information is retained by a sanitizing algorithm, and usability -how (and with how much difficulty) does one extract this information to build statistical models, answer queries, etc. Some scenarios incentivize a separation between information preservation and usability, so that the data owner first chooses a sanitizing algorithm to maximize a measure of information preservation and, afterward, the data consumers process the sanitized output according to their needs [22,46].We analyze a variety of utility measures and show that the average (over possible outputs of the sanitizer) error of Bayesian decision makers forms the unique class of utility measures that satisfy three axioms related to information preservation. The axioms are agnostic to Bayesian concepts such as subjective probabilities and hence strengthen support for Bayesian views in privacy research. In particular, this result connects information preservation to aspects of usability -if the information preservation of a sanitizing algorithm should be measured as the average error of a Bayesian decision maker, shouldn't Bayesian decision theory be a good choice when it comes to using the sanitized outputs for various purposes? We put this idea to the test in the unattributed histogram problem where our decisiontheoretic post-processing algorithm empirically outperforms previously proposed approaches.

show abstract

A workflow for differentially-private graph synthesis

Cited by 47 publications

References 12 publications

Differentially private network data release via structural inference

Differentially private network data release via structural inference

Efficient Privacy-Preserving Stream Aggregation in Mobile Sensing with Low Aggregation Error

Information preservation in statistical privacy and bayesian estimation of unattributed histograms

Contact Info

Product

Resources

About