Prof. Panagiotis Karras
November 28, 2013
13.00 – 14.30
Beijing – 1 Auditorium, China cluster (Skolkovo School of Management)
SEMINAR ABSTRACT:
Today there is a strong interest in publishing data about individuals in a privacy-preserving manner. Data anonymization strives to (i) ensure that an adversary cannot identify an individual’s record from published attributes with high probability and, at the same time, and (ii) provide high data utility. These goals can be formulated as an optimization problem with privacy as the constraint and utility as the objective function. Conventional research, using the k-anonymity model, has resorted to publishing data in groups in which each individual record has its attribute values generalized so as to be indistinguishable from all other group members. A recently proposed alternative does not create such cliques; instead, it recasts data values in a heterogeneous manner, represented by a bipartite graph, aiming for higher utility. Nevertheless, such works either built roundabout solutions, introducing superfluities, or adopt a monolithic fixed-form solutionapplied to all data irrespective of their characteristics; thus, the utility gains they achieve are limited. In this talk I will propose a methodology that achieves the full potential of heterogeneity and gains higher data utility. We formulate the problem of maximal-utility k-anonymization as a network flow problem. We develop an optimal solution therefor using Mixed Integer Programming. Observing the non-scalability of this solution, we develop more efficient heuristic that solve the problem by building a set of k perfect matchings from original to anonymized data. Our techniques can resist adversaries who may know the employed algorithms, thanks to a randomization scheme. Our experiments with real-world data verify that our schemes achieve near-optimal utility, much higher than previous work, while they can exploit a parallel processing environment, gaining even an efficiency advantage over simpler methods. At the end of the talk, I will give an overview of other research activities and discuss the future directions of my research work.
SPEAKER INTRODUCTION:
Panagiotis Karras is an Assistant Professor of Information Systems and Teaching Excellence Fellow at Rutgers Business School. He was an LKY Postdoctoral Fellow at the National University of Singapore from 2008 to 2011 and an Oberassistent at the University of Zurich from 2007 to 2008. He earned a Ph.D. in Computer Science from the University of Hong Kong in 2007 and an M.Eng. in Electrical and Computer Engineering from the National Technical University of Athens in 1999, while he has also worked and studied at the Technical University of Denmark and the Karlsruhe Institute of Technology.
Prof. Karras conducts research in the intersection of data management, data mining, and information security. His work has introduced algorithms and systems that allow for the satisfaction of efficiency, accuracy, privacy, integrity, and adaptability requirements in data representation and querying. He has developed methodsthat achieve optimal synopsis construction in low polynomial time, novel privacy models and algorithms for efficient data anonymization under those models, techniques for query authentication over data streams and query answering over encrypted data, robust and reliable adaptive indexing for modern main-memory column-stores, scalable similarity search over time series data bases, novel spatial join variants and techniques for in-memory spatial joins, and systems for scalable RDF data indexing. Last, he is working on parallel algorithms for graph management problems.
Prof. Karras has published over 35 research articles, advised 8 PhD students, and received over 700 citations. His research has been funded by the Lee Kuan Yew Endowment in Singapore.He received the 2008 Hong Kong Young Scientist Award in Physical/Mathematical Science. He is a member of ACM and IEEE, while he has served in the program committees of and refereed for major conferences and journals in the above areas.
