Understanding the Sparse Vector Technique for Differential Privacy

Lyu, Min; Su, Dong; Li, Ninghui

doi:10.48550/arxiv.1603.01699

Cited by 14 publications

(25 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…, 𝑚 𝑘 (𝑥))) satisfies 𝑖 ∈ [𝑘 ] 𝜖 𝑖 -DP 2.1.3 Randomness Alignment. A DP mechanism can be so complicated (e.g., the sparse vector technique described in the following section) that the wrong proofs appear [25]. Then, many verification tools using a proof syntax [2,32,34] are proposed to verify the correctness of proof for DP.…”

Section: Compositionmentioning

confidence: 99%

“…[14]. Then, successors proposed many applications [25]. Recently, Ding et al found an accuracy-improved version of the sparse vector technique [10].…”

Section: Counting Querymentioning

confidence: 99%

“…Top-𝑘 Query. For evaluating the asymmetric sparse vector technique and the asymmetric report noisy max algorithm, we use two real-world datasets from [25] generator from the IBM Almaden Quest research group: BMS-POS, Kosarak, and T40I10D100K. These datasets are collections of clickstream, and each record is a stream of items.…”

Section: Datasets and A Policymentioning

confidence: 99%

“…Then, we show that ADP allows the Euclidean distance error 1/𝜖 due to the asymmetricity. Second, this paper considers two representative DP mechanisms using the Laplace mechanism: report noisy arg-max and sparse vector technique [10,25]. The algorithms aim to make the histogram of top-𝑘 counts (i.e., set of counting query).…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Asymmetric Differential Privacy

Takagi¹,

Cao²,

Yoshikawa³

2021

Preprint

View full text Add to dashboard Cite

Recently, differential privacy (DP) is getting attention as a privacy definition when publishing statistics of a dataset. However, when answering a decision problem with a DP mechanism, it causes a two-sided error. This characteristic of DP is not desirable when publishing risk information such as concerning COVID-19. This paper proposes relaxing DP to mitigate the limitation and improve the utility of published information. First, we define a policy that separates information into sensitive and non-sensitive. Then, we define asymmetric differential privacy (ADP) that provides the same privacy guarantee as DP to sensitive information. This partial protection induces asymmetricity in privacy protection to improve utility and allow a one-sided error mechanism. Following ADP, we propose two mechanisms for two tasks based on counting query with utilizing these characteristics: top-𝑘 query and publishing risk information of viruses with an accuracy guarantee. Finally, we conducted experiments to evaluate proposed algorithms using real-world datasets and show their practicality and improvement of the utility, comparing state-of-the-art algorithms.

show abstract

Section: Compositionmentioning

confidence: 99%

“…[14]. Then, successors proposed many applications [25]. Recently, Ding et al found an accuracy-improved version of the sparse vector technique [10].…”

Section: Counting Querymentioning

confidence: 99%

Section: Datasets and A Policymentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Asymmetric Differential Privacy

Takagi¹,

Cao²,

Yoshikawa³

2021

Preprint

View full text Add to dashboard Cite

show abstract

“…We propose to further accelerate the execution and reduce privacy cost by augmenting the sparse vector technique using sub-sampling without replacement [2]. However, sparse vector technique is known for its poor utility [17], which raises concern about the accuracy of sieve-and-examine. Actually, there exist two types of errors in sieve-and-examine.…”

Section: Introductionmentioning

confidence: 99%

Towards practical differentially private causal graph discovery

Wang¹,

Pang²,

Song³

2020

Preprint

View full text Add to dashboard Cite

Causal graph discovery refers to the process of discovering causal relation graphs from purely observational data. Like other statistical data, a causal graph might leak sensitive information about participants in the dataset. In this paper, we present a differentially private causal graph discovery algorithm, Priv-PC, which improves both utility and running time compared to the state-of-the-art. The design of Priv-PC follows a novel paradigm called sieve-and-examine which uses a small amount of privacy budget to filter out "insignificant" queries, and leverages the remaining budget to obtain highly accurate answers for the "significant" queries. We also conducted the first sensitivity analysis for conditional independence tests including conditional Kendall's τ and conditional Spearman's ρ. We evaluated Priv-PC on 4 public datasets and compared with the state-ofthe-art. The results show that Priv-PC achieves 10.61 to 32.85 times speedup and better utility.

show abstract