Learning with Histogram Intersection Kernel
Jianxin Wu
Assistant Professor
School of Computer Engineering, Nanyang Technological University, Singapore.
Abstract :
Histogram Intersection Kernel (HIK) was proposed as a way to compare the similarity of two histograms, originally in the computer vision field. HIK was then proved to be a valid Mercer kernel when the histograms are composed of non-negative integer values. In this talk I will introduce recent advances in learning with HIK, in terms of both effectiveness and efficiency. I will present methods that make learning with HIK (almost) as fast as linear methods (e.g. linear SVM). These methods include a super fast HIK SVM solver. On several real-world, large scale computer vision problems, HIK has achieved superior performances. I will generalize HIK SVM learning to real-valued, non-histogram feature vectors.
Bio:
Jianxin Wu received the BS degree and MSc degree in computer science, both from Nanjing University, China. He obtained a PhD in Computer Science in Georgia Institute of Technology under the supervision of Dr. James M. Rehg. He is currently an assistant professor in the School of Computer Engineering, Nanyang Technological University, Singapore. His research interests are computer vision, machine learning, and robotics.