| We are pleased to invite you to the coming PREMIA's Annual General Meeting (AGM) 2010 to be held on 19 August at NUS Shaw Foundation Alumni House. |
| 2010 Annual General Meeting (AGM) with a talk on Graph Construction and Manipulation Towards Informativeness and Scalability |
| Dr. Yan Shuicheng |
| |
| Date: 19 August 2010 |
| Time: 6:30 pm |
| Venue: Seminar Rooms 2 and 3, Level 2, Shaw Foundation Alumni House |
| |
Registration: Please register online via http://www.premia-sg.org/index.php?option=com_events&event_id=31&task=register
(Registration is closed now). |
Synopsis:
Graph is a general and popular representation, which can describe complex high order relations through pairwise relations (edges). This talk covers two aspects of graph. First, graph construction, a fundamental problem for many pattern recognition tasks, is restudied beyond traditional k-NN graph and epsilon-ball graph. Sparse coding is introduced for the construction of the so-called L1-graph to characterize more informative and robust one-to-many relations, whose scalability is further augmented by Hashing trick. The positive semidefinite property of the weight matrix of the enhanced collective L1-graph construction is theoretically proved based on low-rank objective. Several successful applications of L1-graph are introduced from both computer vision and multimedia areas. Then, by taking dense subgraph detection as an example, we introduce how to manipulate large graph for certain tasks. More specifically, we define graph modes, which are the local maxima of graph density functions, to represent such dense subgraphs. We propose the graph shift procedure, which starts from every vertex, iteratively shifts the local small subgraph towards the nearest graph mode along a certain trajectory. Both theoretic analyses and experiments show that graph shift procedure is very efficient (only working on quite small subgraph for each step) and robust, especially when there exists large amount of noises and outliers. |
Biography:
Dr. Yan Shuicheng is currently an Assistant Professor in the Department of Electrical and Computer Engineering at National University of Singapore, and the founding leader of the Learning and Vision Research Group (http://www.lv-nus.org). Dr. Yan's research areas include computer vision, multimedia and machine learning, and he has authored or co-authored about 180 technical papers over a wide range of research topics. He has close collaborations with industry companies, such as Microsoft, Panasonic, Kodak, Google, and Omron. He is an associate editor of IEEE Transactions on Circuits and Systems for Video Technology, and has been serving as the guest editor of the special issue for Computer Vision and Image Understanding. He received the Best Paper Awards from ICME10 and ICIMCS'09. |
