Differential privacy based on importance weighting

Ji, Zhanglong; Elkan, Charles

doi:10.1007/s10994-013-5396-x

Cited by 27 publications

(20 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In the Nissim's paper, the conception of local sensitivity is introduced, which can reduce the noise added to the real query answers. And there are papers focusing on privacy protection such as [3] and there are lots of papers focusing on privacypreserving data minting such as [1], [4], [5], [8], [10], [11]. All of these papers are based on noise drawn from Laplace distribution with mean 0.…”

Section: Pinq(privacy Integrated Queriesmentioning

confidence: 99%

Optimizing Query Times for Multiple Users Scenario of Differential Privacy

et al. 2019

View full text Add to dashboard Cite

Differential privacy is the state-of-the-art for preserving privacy and differential privacy mechanism based on Laplace distribution with mean 0 is common practice. However, privacy budget is exhausted so quick that the number of queries is not enough. In this paper, a differential privacy mechanism is proposed to optimize the number of queries for application scenario of multiple users. We isolate different users by assigning various noise distribution with non-zero mean to different users. First, in terms of privacy guarantee, the proposed mechanism is better than common practice. Second, for the utility aspect, the accuracy of proposed mechanism is analyzed from the view of data distribution's distortion and the view of noise's absolute value.INDEX TERMS Differential privacy, multiple users, non-zero mean.

show abstract

Section: Pinq(privacy Integrated Queriesmentioning

confidence: 99%

Optimizing Query Times for Multiple Users Scenario of Differential Privacy

et al. 2019

View full text Add to dashboard Cite

show abstract

“…In section 3.1, the supervised learning module and the feature extraction module of the proposed framework were evaluated using the subspace (ɑ3, ɑ4, ɑ11) of the NSL-KDD data set with binary classes (9,10) only. In a new experiment, the same subspace is again considered; however, the other classes (0,1), (0,5), (1,2), (1,9), (3,5), (3,9), and (6,8) are also studied. In addition, three other subspaces, (ɑ3, ɑ4, ɑ5), (ɑ3, ɑ4, ɑ7), and (ɑ4, ɑ7, ɑ10), are also included in the experiment to study the performance of the proposed analytical framework with DPLR.…”

Section: Multiple Subspace Analysismentioning

confidence: 99%

“…[6]. Since its introduction, a significant amount of studies have been conducted using this model for achieving privacy strength and prediction/classification accuracy [8,9,12,16]. This is a parametric approach and the selection of its privacy parameter ϵ is a challenging problem.…”

Section: Introductionmentioning

confidence: 99%

Characterization of differentially private logistic regression

Suthaharan

2018

Proceedings of the ACMSE 2018 Conference

View full text Add to dashboard Cite

The purpose of this paper is to present an approach that can help data owners select suitable values for the privacy parameter of a differentially private logistic regression (DPLR), whose main intention is to achieve a balance between privacy strength and classification accuracy. The proposed approach implements a supervised learning technique and a feature extraction technique to address this challenging problem and generate solutions. The supervised learning technique selects subspaces from a training data set and generates DPLR classifiers for a range of values of the privacy parameter. The feature extraction technique transforms an original subspace to a differentially private subspace by querying the original subspace multiple times using the DPLR model and the privacy parameter values that were selected by the supervised learning module. The proposed approach then employs a signal processing technique called signal-interference-ratio as a measure to quantify the privacy level of the differentially private subspaces; hence, allows data owner learn the privacy level that the DPLR models can provide for a given subspace and a given classification accuracy.

show abstract

“…Assuming there is labeled public data available, the importance weighting approach of Ji and Elkan (2013) can be used for efficient differentially private data publishing. Ji and Elkan (2013) report that the method can reach accurate results already with a small privacy budget, but their example has a much lower dimensionality than any genomic dataset and it is unclear how the method would scale to genomic data.…”

Section: Introductionmentioning

confidence: 99%

Representation transfer for differentially private drug sensitivity prediction

et al. 2019

View full text Add to dashboard Cite

Motivation Human genomic datasets often contain sensitive information that limits use and sharing of the data. In particular, simple anonymization strategies fail to provide sufficient level of protection for genomic data, because the data are inherently identifiable. Differentially private machine learning can help by guaranteeing that the published results do not leak too much information about any individual data point. Recent research has reached promising results on differentially private drug sensitivity prediction using gene expression data. Differentially private learning with genomic data is challenging because it is more difficult to guarantee privacy in high dimensions. Dimensionality reduction can help, but if the dimension reduction mapping is learned from the data, then it needs to be differentially private too, which can carry a significant privacy cost. Furthermore, the selection of any hyperparameters (such as the target dimensionality) needs to also avoid leaking private information. Results We study an approach that uses a large public dataset of similar type to learn a compact representation for differentially private learning. We compare three representation learning methods: variational autoencoders, principal component analysis and random projection. We solve two machine learning tasks on gene expression of cancer cell lines: cancer type classification, and drug sensitivity prediction. The experiments demonstrate significant benefit from all representation learning methods with variational autoencoders providing the most accurate predictions most often. Our results significantly improve over previous state-of-the-art in accuracy of differentially private drug sensitivity prediction. Availability and implementation Code used in the experiments is available at https://github.com/DPBayes/dp-representation-transfer.

show abstract

Differential privacy based on importance weighting

Cited by 27 publications

References 18 publications

Optimizing Query Times for Multiple Users Scenario of Differential Privacy

Optimizing Query Times for Multiple Users Scenario of Differential Privacy

Characterization of differentially private logistic regression

Representation transfer for differentially private drug sensitivity prediction

Contact Info

Product

Resources

About