Machine Learning with Adaptive Data Transformation

Dr. Carlotta Domeniconi
Assistant Professor
Department of Information and Software Engineering
George Mason University

Abstract

Much of machine learning attempts to extract, from a sample of data, a hypothesis which explains the nature of the environment from which that data was generated. Though long and widely used, machine learning techniques still face daunting challenges posed by real world problems. These problems include: handling the (often) high dimensionality of the environment, transforming the data such that the hypothesis is more parsimonious and general, and eliminating assumptions one must place on the environment in order to best apply a given machine learning technique.

I will explain common machine learning paradigms and will discuss challenges in the field such as the ones above. I will then introduce techniques I have developed which overcome these challenges. Many of these techniques are based on adaptive transformations of the data space into ones more appropriate for the machine learning task at hand. I will also discuss various machine learning problem domains to which I have applied my techniques, including microarray data analysis, text classification, and anomaly detection.

Speaker Bio

Carlotta Domeniconi received a Laurea Degree in computer science from the University of Milan, Italy, in 1992, an M.S. in information and communication technologies from the International Institute for Advanced Scientific Studies, Salerno, Italy, in 1997, and a Ph.D. in computer science from the University of California, Riverside, in 2002.

She is Assistant Professor in the Department of Information and Software Engineering at George Mason University. Her research interests include machine learning, pattern recognition, data mining, and feature relevance estimation, with applications in text mining and bioinformatics. Her research is in part supported by an NSF CAREER Award and a grant from the U.S. Army.