Differentially Private Data Releasing for Smooth Queries
Authors: Ziteng Wang, Chi Jin, Kai Fan, Jiaqi Zhang, Junliang Huang, Yiqiao Zhong, Liwei Wang
JMLR 2016 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We also develop practically efficient variants of the mechanisms with promising experimental results. Finally we conduct experiments on the efficient variant of the mechanism which outputs synthetic database as it may be more useful in practice. Experimental results demonstrate that the algorithm achieves good accuracy and are practically efficient on datasets of various sizes and of a number of attributes. Section 5.2.1 Experimental Results |
| Researcher Affiliation | Academia | Key Laboratory of Machine Perception (MOE), School of EECS Peking University Beijing, 100871, China; Department of Computer Science University of California Berkeley, CA 94720-1776, USA; Computational Biology & Bioinformactics Duke University Durham, NC 27708, USA; School of Mathematical Sciences Peking University Beijing, 100871, China |
| Pseudocode | Yes | Algorithm 1 Outputting the summary Algorithm 2 Answering a query Algorithm 3 Private Synthetic DB for Smooth Queries Algorithm 4 Private Subspace Iteration (Hardt, 2013) |
| Open Source Code | Yes | 1. Source codes available at http://www.cis.pku.edu.cn/faculty/vision/wangliwei/software.html |
| Open Datasets | Yes | We adopt three datasets all from the UCI repository. A summary of the size and the number of attributes of these datasets is given in Table 3. |
| Dataset Splits | Yes | We randomly partition each dataset into two subsets of equal size. One subset is used as training dataset, the other as test dataset. |
| Hardware Specification | Yes | The computer used in all the experiments is a workstation with 2 Intel Xeon X5650 processors of 2.67GHz and 32GB RAM. |
| Software Dependencies | No | The paper mentions learning an SVM classifier and numerical integration methods, but does not provide specific software names with version numbers for reproducibility. |
| Experiment Setup | Yes | Detailed parameter setting of the query functions is as follows. We consider P J j=1 αj exp x xj 2 . In all experiments, we set J = 10; αj is randomly chosen from [0, 1], and xj is randomly chosen from [ 1, 1]d. We test various values of σ and various ϵ to see how the smoothness of the query function and privacy parameter affect the performance of the algorithm (see below for detailed results). We test three values of ϵ, and the performances are given in Table 5. |