【学术报告】Incremental Semi-supervised Classification with Adaptive Metric Learning: Toward Interactive Classification of Multimedia Data

发稿时间:2010-03-11浏览次数:2328

题目: Incremental Semi-supervised Classification with Adaptive Metric Learning: Toward Interactive Classification of Multimedia Data

报告人:冈田将吾  日本京都大学

报告时间:3月15日9:00—10:30

报告地点:404会议室

摘要:
Goal of our reseach is to acheive a classification system of multimedia data through long term Human Computer Interaction. Toward this object, Our group proposes a novel incremental semi-supervised learning algorithm which is fundamental mechanism on the system. In this talk, I introduce the detail of proposed  algorithm and explain the experiment of the algorithm by using image data. It  is a  hybrid algorithm based on Self-organizing Incremental Neural Network , Graph based transactive learning algoritm and  Adaptive Metric Learning algorithm. Advantage of the algorithm  is that  incremental clustering and  learning of metric of high dimensional input data is performed simultaneously. Good distance metric for classification is calculated by using a few label data and a large amount of unlabeld data. Experimental results on real-world datasets show that the proposed algorithm is significantly better than graph based Semi-supervised learning algorithms in terms of classification accuracy.
 
个人简历
Shogo Okada received the B.E. degree in engineering from the Yokohama National University, Japan, and the M.E. degree and the PhD degree in engineering from the Tokyo Institute of Technology, Japan, in 2003, 2005 and 2008, respectively. He worked as an assistant professor at Kyoto University from 2008. His research interests include Pattern Recognition, Machine Learning, Neural Network, and it's application to Human Robot Interaction and Autonomous Development Robot.