【论文笔记15】Boosting and Differential Privacy, 对查询集进行boosting以达到对数据的差分隐私, IEEE ASFCS 2010

原文链接
我的论文笔记频道

在一篇综述里，是这样介绍本文的工作

Dwork et al. [36] designed a queryboosting algorithm, which aims to convert a weak and sometimes­accurate learner into a strong and accurate learner with differential privacy. It considers the input database as a training dataset, each row in the database as exactly a sample and almost does not compromise the accuracy. To achieve privacy preserving, it gradually changes the weight as a function of how accurate the answer is, rather than using a sharp threshold between the accurate and inaccurate answers, due to the fact that the change of each row in the database can affect the answers to all the queries and thus influence the distribution of queries

不是他没读懂 Boosting and Differential Privacy 就是我没读懂。。。

1 Abstract

2 Introduction

2.1 Summary of Results

Principle Result: a technique for generating privacy-preserving synopses for any set of low-sensitivity queries (不限于计数查询). This is achieved by a novel use of boosting, together with the construction of an appropriate base synopsis generator.

2.2 Boosting for Queries

2.3 Base Synopsis Generators for Arbitrary Low-Sensitivity Queries and for Counting Queries

2.4 Bounding Expected Privacy Loss and Composition Theorems

Reference

[1] Dwork, Cynthia, Guy N. Rothblum, and Salil Vadhan. “Boosting and differential privacy.” 2010 IEEE 51st Annual Symposium on Foundations of Computer Science. IEEE, 2010.