主       题：A Structured Graph Optimization Approach for Effective and Parameter-free Clustering 

报 告 人：聂飞平  教授
主 持 人：黄哲学 所长
日      期：2015年12月23日
时      间：下午 2：30
地      点：计软学院A623会议室

In this talk, I will introduce an effective clustering method called Clustering with Adaptive Neighbors (CAN). CAN learns the data similarity matrix and clustering structure simultaneously. The data similarity matrix is learned by assigning the adaptive and optimal neighbors for each data point based on the local distances. Meanwhile, a rank constraint is imposed to the Laplacian matrix of the data similarity matrix, such that the connected components in the similarity matrix are exactly equal to the cluster number. Thanks to this constraint, the learned similarity matrix enjoys block diagonal structure with proper permutation, and thus the data clustering can be directly conducted through the similarity matrix. A very simple and efficient algorithm is derived to optimize this challenging constrained problem. Theoretical analysis shows CAN is closely connected with K-means clustering and spectral clustering. CAN can be further extended to the Projected CAN (PCAN) to handle the high-dimensional data.
 

BIOGRAPHY
聂飞平，西北工业大学“光学影像分析与学习中心”教授、 博士生导师。2009年于清华大学自动化系获博士学位。主要兴趣为模式识别与机器学 习中的理论和方法设计等方面的研究工作，并已将所设计的 方法成功应用于图像分割与标注、多媒体信息检索、生物信 息学等领域的实际问题中。已在PAMI、IJCV、Bioinformatics、 ICML、NIPS、IJCAI、AAAI、SIGKDD、ICCV、CVPR等国 际顶尖期刊和会议上共发表学术论文百余篇，其中在中国计 算机学会（CCF）推荐的A类期刊和会议上发表论文70余篇。 据Google Scholar统计，论文总引用为3900余次，H指数为35。 常年担任相关领域顶级期刊和会议的审稿人或程序委员，同 时担任IEEE Transactions on Neural Networks and Learning Systems、Information Science等多个一流SCI期刊的编委。