On December 12, CSEE professor Hillol Kargupta will receive the 10-year Highest-Impact Paper Award from the IEEE International Data Mining Conference (ICDM) in Brussels, Belgium.
The winning paper—“On the Privacy Preserving Properties of Random Data Perturbation Techniques”—discusses privacy-preserving data mining and it also received the 2003 ICDM Best Paper Award. It is co-authored by former UMBC PhD student Souptik Datta (CS '08) and Dr. Kargupta’s colleagues at Washington State University—Qi Wang and Professor Krishnamoorthy Sivakumar.
Privacy Preserving Data Mining (PPDM) is important in many domains where the data is privacy sensitive and exposing the data to a third party for mining is not an option. Researchers have come up with many PPDM algorithms that attempt to protect data privacy while allowing analysis of the data for detecting patterns. Many of these algorithms make use of randomized techniques. This paper offers a perspective on the structure of random noise using theories of random matrices and their spectral properties in order to analyze their role in preserving data privacy while still keeping data patterns intact for analysis. It points out that spectral properties of random matrices can be exploited to create attacks on many commonly used privacy-preserving data mining algorithms.
Kargupta and his associates point out is that you must be very careful when using random noise to protect data, since it can be easily filtered out. “Random noise is really not that unpredictable,” explains Kargupta, since it has a pattern of its own.
Out of all of the papers on data mining published within the last ten years, this year Dr. Kargupta’s paper was chosen by IEEE as the most impactful paper in its field.