Ensemble Boosting Algorithms Based on Data Sorting for Semi-Supervised Classification
主 讲 人：王建忠
In the presentation, several novel ensemble boosting algorithms based on data sorting for semi-supervised learning (SSL) are introduced. In such an algorithm, we first sort the data into several strings with randomly chosen heads, and construct a pre-classifier as a parallel ensemble of several weak classifiers that are constructed on individual data strings. Then we employ the pre-classifiers in a recursive boosting algorithm to enlarge the labeled data set or to learn the metric on data. We construct the final classifier according to the enlarged labeled set or the learned metric. We display three different boosting algorithms in the talk: Boosting the labeled set at random, boosting the labeled set by clusters, and boosting the labeled set by update metric. The validity and effectiveness of the algorithms are confirmed by the experiments on data sets of different types, such as handwritten digits and hyperspectral images. Comparing to several other popular SSL methods, the proposed method produces very promising results.
王建忠教授现任美国得克萨斯州山姆休斯顿大学教授。1967年毕业于北京大学数学力学系，1981年获浙江大学应用数学系硕士学位。历任武汉大学教授，香港中文大学客座教授，Ohio State University at Columbus客座教授，Approximation Center in Department of Mathematics, Texas A&M University高级访问研究员，University of North Carolina at Charlotte客座教授。主要研究方向包括：样条和逼近理论、小波分析、图像处理以及高维数据分析等。曾主持多项美国自然科学基金及其他基金课题研究。发表学术论文70余篇，专利1项。王教授获得过国内外多项奖项：The 2007 Excellence in Research Award of Sam Houston State University, 2007；The Third Class Award of Advanced Research in Science and Technology, Chinese National Education Committee, 1992 (joint with Professor Jianke Lu at Wuhan University)；Award of Special Allowance of The State Council of China (the second batch)；New Outstanding Talent of Wuhan City, 1985；Awards of “Outstanding Research Papers” issued by Hubei Province and Wuhan City, 1985, 1986, 1987, 1988, 1989, and 1991。