Seminar: Recent Results in Addressing Embedded Implementation of Support Vector Machines (SVM)

Abstract

Support Vector Machines (SVM) are a family of machine learning algorithms widely used in classification and regression tasks. When dealing with large and structurally complex data sets, the classification or prediction by SVM can become memory intensive and requires significant computation resources, and especially so in non-sparse variants of the SVM such as the least-squares SVM (LS-SVM). This is of particular importance for implementing multi-classifiers in embedded systems where memory and computation capabilities are limited with added power penalty associated with network communication. In this talk we will discuss a decomposition method for SVM classification functions using polynomial approximation. The SVM decision function is expanded into a polynomial form and consolidated into classification function with a significantly lower memory footprint and computational cost (two orders of magnitude for each metric). We will also introduce a method to codify an “allowed confusion” between sets in the feature space.  This method allows the integration of expert knowledge that may be important in industrial and commercial applications of automated classification. The methods are validated with public and proprietary classification data sets.

About the Speaker

George Chiu is a Professor in the School of Mechanical Engineering with courtesy appointments in the School of Electrical and Computer Engineering and the Department of Psychological Sciences at Purdue University. He received the B.S. degree in Mechanical Engineering from the National Taiwan University in 1985 and the M.S. and Ph.D. degrees in Mechanical Engineering from the University of California at Berkeley, in 1990 and 1994, respectively. Before joining Purdue, he worked at the Hewlett-Packard company, designing printers and multifunction devices.  Dr. Chiu's current research interests are mechatronics and dynamic systems and control with applications to digital imaging, digital fabrications and functional printing, human motor control, motion and vibration perception and control. Between September 2011 and June 2014, he served as the Program Director for the Control Systems Program at National Science Foundation. He received the 2012 NSF Director’s Collaboration Award, the 2010 IEEE Transactions on Control System Technology Outstanding Paper Award, the Purdue University College of Engineering Faculty Engagement/Service Excellence Award in 2010, and Team Excellence Award in 2006. He served as the Editor for the Journal of Imaging Science and Technology and is the Editor-in-Chief Elect for the IEEE/ASME Transactions on Mechatronics. Dr. Chiu was the Chair of the Executive Committee of the ASME Dynamic Systems and Control Division (DSCD) between 2013 and 2014. He is a Fellow of ASME and a Fellow of the Society for Imaging Science and Technology (IS&T). 

Hosted by Professor Junmin Wang.