Speaker: Dragos Trinca
Day:  Wednesday, 9/27/2006
Room: ITEB 201
Time: 2:30pm

Title: Data Distortion Techniques for Privacy-preserving Data Mining

Abstract: Privacy preservation has recently become one of the top priorities in the design of various data mining applications. A critical component in the design of privacy-preserving data mining algorithms is data distortion.  In this lecture, we will discuss several data distortion techniques: (1) Adding Uniformly Distributed Noise, (2) Adding Normally Distributed Noise, (3) Distorting a Matrix using its Singular Value Decomposition (SVD), and (4) Distorting a Matrix using its Sparsified SVD. Some data distortion measures assessing the degree of the distorted data will be considered. The utility of the distorted data can be measured using Support Vector Machines classifications. Experimental results will also be presented.