EAPSI: Creating Fast and Accurate Data Mining Algorithms that Preserve Privacy

Fish Benjamin S, Chicago IL

Investigators

Abstract

This project will use new techniques to create fast and accurate algorithms that preserves the privacy of people's data in big data mining, i.e. tasks that require large amounts of data. While big data mining has extraordinarily broad and important uses, releasing the output of data mining algorithms to data analysts and the public can result in leaking the private data of individuals whose data is being used. Even when data mining techniques are able to keep data private, they may be too slow or inaccurate to be used widely in practice. Previous approaches to big data mining tasks have typically either focused on a) keeping private data private or b) achieving fast and accurate algorithms, but not both. This research aims to find algorithms for big data mining tasks that are simultaneously privacy-preserving, fast, and accurate, to the degree that this is possible. This project will be conducted at The University of Melbourne with Benjamin Rubinstein, a leading expert in fast techniques for achieving privacy-preserving algorithms. Focus will be on finding such fast differentially-private techniques for the Frequent Itemset mining problem (FI), a fundamental question in knowledge discovery about extracting frequently occurring sets of items in a database, and related problems. Recent approaches have demonstrated fast and accurate approximation algorithms for these types of problems. Previous methods for solving this problem were not necessarily amenable for achieving privacy-preservation quickly and accurately. These new approaches, however, which involve sampling techniques and using under-utilized tools from the theory of machine learning, are especially helpful for creating privacy-preserving algorithms. This research will leverage this approach to create algorithms that are not only fast and accurate, but also keep individuals' data private. This research will also analyze the tradeoffs between speed, accuracy, and privacy in such solutions. This award under the East Asia and Pacific Summer Institutes program supports summer research by a U.S. graduate student and is jointly funded by NSF and the Australian Academy of Science.