题目：Trends and New Directions in Data Clustering

讲座人：Ming Dong, Wayne State University, USA

讲座时间：7月10号（星期四）下午3:30-5:30

地点：海韵行政楼C505

摘要：

Today digital data are accumulated at the faster than ever speed in science, engineering, biomedicine, and real-world sensing. Clustering has been widely used in data mining to discover the interesting patterns and gain insights from large amounts of data. Usually, real-world data are highly complex, sharing one or several following prominent characteristics: they are tremendous in size with millions of objects; they come unstructured with heterogeneous features; knowledge is often embedded in large amounts of noisy, even confusing data. The complexity of the acquired digital data overwhelms the useful information and makes it extremely difficult to derive true understanding from it. This talk presents our recent efforts in addressing some of these challenges: (1) How to handle relational and heterogeneous data? We proposed a novel graph theoretic approach to perform pairwise and high-order heterogeneous data co-clustering; (2) How to incorporate prior or background knowledge to improve the quality of clustering?  We developed a non-negative matrix factorization framework for semi-supervised clustering, in which user is able to provide pairwise constraints on a few data objects specifying whether they “must" or “cannot" be clustered together; (3) How to efficiently mine and visualize extremely large-scale data? We proposed an exemplar-based clustering and visualization technique, which provides high efficiency and high interpretability through the use of exemplars. Applications of data clustering in biomedical imaging, bioinformatics, text mining and web analysis will also be discussed.

Short Bio

Ming Dong received his B. S. degrees in electrical engineering and industrial management engineering from Shanghai Jiao Tong University, Shanghai, China, in 1995. He received his Ph. D degree in electrical engineering from University of Cincinnati in 2001. He joined the faculty of Wayne State University in 2002 and is currently an Professor in the Department Computer Science and the director of Machine Vision and Pattern Recognition Laboratory. Dr. Dong's areas of research include pattern recognition, data mining, and multimedia content analysis. His research is funded by National Science Foundation, State of Michigan, and Industries. He has published over 90 technical articles in premium journals and conferences in related fields, e.g., TPAMI, TKDE, TNN, TVCG, CVPR, ACM MM and WWW, and received over 1,000 citations until now. He was as an associate editor of IEEE Transactions on Neural Networks (2008-2011) and Pattern Analysis and Applications (2007-2010), and served in many conference program committees and US National Science Foundation panels. He also served as senior research consultant in Baidu Inc. and Ford Motor Company, and has given over twenty invited talks in various institutes.