Event(s) on July 2006

• 3/7/2006

  題目:	High-Order Spectral Difference Method for Conservation Laws on Triangular Grids
  講員:	Dr. Zhi Jian Wang, Department of Aerospace Engineering, Iowa State University, USA
  時間/地點:	11:30  -  12:30 FSC 1217
  摘要:	An efficient, high-order, conservative method named the spectral difference method has been developed recently for conservation laws on unstructured grids. It combines the best features of structured and unstructured grid methods to achieve high computational efficiency and geometric flexibility. In this talk, I will present how to extend the method to nonlinear systems of conservation laws, the Euler equations. Accuracy studies are performed to numerically verify the order of accuracy. Other benchmark test results will be shown to demonstrate its performance.

  
  

• 4/7/2006

  題目:	Finding Global Minimizers of Segmentation and Denoising Functionals
  講員:	Prof. Selim Esedoglu, Department of Mathematics, University of Michigan, USA
  時間/地點:	14:30  -  15:30 FSC 1217
  摘要:	Segmentation is a fundamental procedure in computer vision. It forms an important preliminary step whenever useful information is to be extracted from images automatically. Given an image depicting a scene with several objects in it, its goal is to determine which regions of the image contain distinct objects. Variational segmentation models, such as the Mumford-Shah functional and its variants, pose segmentation as finding the minimizer of an energy. The resulting optimization problems are often non-convex, and may have local minima that are not global minima, complicating their solution. We will show that certain simplified versions of the Mumford-Shah model can be given equivalent convex formulations, allowing us to find global minimizers of these non-convex problems via convex minimization techniques. In particular, we will show that a recent convex duality based algorithm due to A. Chambolle, which was originally developed for Rudin, Osher, and Fatemi's total variation denoising model, can be adapted to the segmentation problem.

  
  

• 11/7/2006

  題目:	A Unified Architecture for Searching, Extracting and Mining Large Biomedical Literature Databases
  講員:	Prof. Xiaohua (Tony) Hu, College of Information Science and Technology, Drexel University, USA
  時間/地點:	14:30  -  16:30 FSC 1217
  摘要:	Despite the rapid electronic dissemination of research results and experimental data in bioinformatics research, most bioinformatics knowledge and experimental results still exist only in text formats. Retrieving and processing this information is made difficult due to the large volumes and the lack of formal structure in the natural-language narrative in those documents. It is very important to develop efficient and effective technologies that automatically search large collections of biomedical literature, extract and mine the important biological relationships such as protein-protein interaction, functionalities of the genes, etc, so that domain experts can analyze this information to form new hypotheses, conduct new experiments and enable new discoveries. In this talk, I will discuss our work in the efficiency and effectiveness of information retrieval procedures; of pattern learning methods for information extraction; and information overload in text mining simultaneously in a coherent and unified framework for biomedical literature data mining. A novel and unified architecture, the Bio-SET-DM (Biomedical Search, Extraction and Text Data Mining), will take ffective document retrieval, (2) automatic pattern generation and evaluation based on mutual bootstrapping for robust and portable information Extraction, and (3) ontology-enhanced Text Data Mining in biomedical literature.

  
  

• 12/7/2006

  題目:	Network Analysis of Protein-Protein Interactions
  講員:	Prof Xiaohua (Tony) Hu, College of Information Science and Technology, Drexel University, USA
  時間/地點:	14:30  -  16:30 FSC1217
  摘要:	This talk consists of two parts. First we report a comprehensive evaluation of the topological structure of protein-protein interaction (PPI) networks by mining and analyzing graphs constructed from the publicly available popular data sets to the bioinformatics research community. We compare the topology of these networks across different species, at different confidence levels, and from different experimental systems. Our results confirm the well-accepted claim that the degree distribution follows a power law. However, further statistical analysis shows that the residues are not independent on the fit values, indicating that the power law model may be inadequate. Our results also show that the dependence of the average clustering coefficient on the vertices degree is far from a power law, contradicting many published results. For the first time, we report that the average vertex density exhibits a strong power law dependence on the vertices degree for all the networks studied, regardless of species, confidence levels, and experimental systems. In the second part, we present a novel and efficient approach for the detection of a community in a protein-protein interaction network. Our approach exploits the underlying topological structure of the interaction network and automatically discovers a community by growing it from a given seed protein. Experiment results show strong structural and functional relationships among member proteins within each of the communities identified by our approach, as verified by MIPS functional and complex catalogue database.