Feature Selection via Local Hyperplane Approximation

by Dr. Hongmin Cai, South China University of Technology

 :  01 Feb 2013 (Fri)
 :  11:00 am - 12:00 noon
Venue  :  Rm 4475, 4/F (Lift 25, 26), Academic Complex, HKUST


In this talk, we will report a new feature weighting algorithm through maximizing a margin via local hyperplane approximation.  The key idea is to estimate the feature weights through local approximation rather than global measurement, as used in previous methods. The proposed method can be categorized into the famous RELIEF framework.  In particularly, the feature weights can be further estimated by minimizing leave-one-out cross validation error of classifier of HKNN. Therefore, optimal solution can be explicitly approximated to give feature estimation.  The weights obtained by our method are more robust to degradation of noisy features, even when the number of dimensions is huge. Empirical study on both synthetic and real-world data sets demonstrate the superior performance of the feature selection for supervised learning, and the effectiveness of our algorithm.


Dr. Cai received bachelor and master’s degree from Harbin Institute of Technology in 2001 and 2003, respectively.  He got Ph.D from University of Hong Kong in 2007.  In 2005, he was a research fellow at the Center of Bioinformatics at Harvard University. In 2006, he was visiting Section for Biomedical Analysis of Prof. Christos Davatzikos at University of Pennsylvania.  From 2008 to 2012, he was assistant Professor at the School of Information and Technology,  The Sun Yat-Sen University.  In 2012, he was associate Professor of School of Computer Science and Technology, South China University of Technology.  He has principled and co-principled fundings from NSF, and Guangdong Province NSF. His research interests including biomedical image process and bioinformatics, in particular image segmentation and analysis, survival data mining.