Dept Colloquium Schedule for 2005/2006

Maintained by: Cezary Z. Janikow

Select:

Spring 2006
Fall 2005
Attendence form (graduate CS students admitted Fall 2005 and later)

Note: Standard venue is CCB302, standard day/time has to be decided.
Tools: Please specify in advance: e.g., overhead or PC projector, viewer, network, etc
Audience: General (faculty and students) or specific (as defined)

Schedule for Spring

Date	Place	Refreshments Time	Talk time	Speaker	Title
1/25 Wed	CCB302	2:30pm	3pm	Songtao Liu	Wavelet Multiscale Methods and Numerical Solutions of PDEs
2/15 Wed	CCB302	3:00pm	3:30pm	Martin Pelikan	Analysis of the Hierarchical Bayesian Optimization Algorithm (hBOA) on Classes of Random Problems
2/22 Wed	CCB302	3:00pm	3:30pm	Gualtiero Piccinini	The Physical Church-Turing Thesis: Modest or Bold?
3/10 Fri	CCB302	2:00pm	2:30pm	Yu-Ping Wang	Genetic image analysis with wavelets
3/20 Mon	CCB302	3:00pm	3:30pm	Weizhen Wang	Statistical Inferences in Orthogonal Saturated Designs
4/10 Mon	CCB302	3:00pm	3:30pm	Kang Chen	Financial Herding in a Local Interaction Model
4/19 Wed	CCB302	3:00pm	3:30pm	Sharlee Climer	A Formalization of the Use of Bounds with Applications in Biology and Engineering
5/1 Mon	CCB302	TBA	3pm	Nikolay Romanovsky	Free product decompositions in images of certain free products of groups

Schedule for Fall

Date Place Refreshments
Time Talk time Speaker Title

10/19
Wed
CCB302
3:30
4:00
Martin Butz The XCS Classifier System: From Theory to Applications

10/24
Mon
CCB302
3:30
4:00
Xin Yao The iterated prisoner's dilemma (IPD)

11/2
Wed
CCB302
3:00
3:30
Sanjiv Bhatia A Hierarchical Clustering Scheme for Image Databases

11/28 Mon
CCB302
3:30
4:00
Karen Chandler
The Fr\"oberg-Iarrobino conjecture on infinitesimal interpolation

Speaker: Sanjiv Bhatia
Title: A Hierarchical Clustering Scheme for Image Databases.
Abstract: The organization of an image database is one of the important issues in efficient storage and retrieval of images. Most of the existing image databasesare based on flat structures, with the possibility of an index into the databasethat can help in narrowing down the images to be searched. In this paper, I'll present a technique to create a hierarchical data structure based on the clustering approach such that a user can select or discard a number of images for subsequent operations. The presented technique is based on application of wavelet analysis to scale the images in hierarchy, and can take advantage of the structure of compressed images in the JPEG 2000 standard.
Bio: Sanjiv Bhatia received his Ph.D. from the University of Nebraska -- Lincoln in 1991. He is presently working as an Associate Professor in the Department of Mathematics & Computer Science in the University of Missouri -- St. Louis. His primary area of research is Image Databases, Digital Image Processing, and Computer Vision. He has been involved in digital terrain modeling for flight simulation and has recently been working on tracking objects in FLIR video streams. He has published several papers on image databases and the application of knowledge-based techniques to information retrieval. He is a member of ACM and AAAI.
Audience: General
Tools: Laptop+projector

Speaker: Uday Chakraborty
Title:
Abstract:
Bio:
Audience:
Tools:

Speaker: Wenjiv He
Title:
Abstract:
Bio:
Audience:
Tools:

Speaker: Cezary Z Janikow
Title:
Abstract:
Bio:
Audience:
Tools:

Speaker: Henry Kang (TBA)
Title: Interactive Sketch Generation
Abstract: In this talk, we propose an interactive system for generating artistic sketches from images, based on stylized multiresolution B-spline curve model and livewire contour tracing paradigm. Our multiresolution B-spline stroke model allows interactive and continuous control of style and shape of the stroke at any level of details. Especially, we introduce a novel mathematical paradigm called wavelet frame to provide essential properties for multiresolution stroke editing, such as feature point preservation, locality, time-efficiency, good approximation, etc. The livewire stroke map construction leads the user-guided stroke to automatically lock on to the target contour, allowing fast and accurate sketch drawing. We classify the target contours as outlines and interior flow, and develop two respective livewire techniques based on extended graph formulation and vector flow field. Experimental results show that the proposed system facilitates quick and easy generation of artistic sketches of various styles.
Audience: General
Tools: Projector

Speaker: Martin Pelikan
Title: Analysis of the Hierarchical Bayesian Optimization Algorithm (hBOA) on Classes of Random Problems
Abstract: The hierarchical Bayesian optimization algorithm (hBOA) derives inspiration from evolutionary computation and machine learning. hBOA solves optimization problems by building and sampling a probabilistic model of promising solutions in the form of a Bayesian network with local structures. It has been theoretically and empirically shown that hBOA can solve nearly decomposable and hierarchical problems of bounded difficulty in low-order polynomial time. In this talk, I will outline hBOA and analyze its performance on two classes of random problems: (1) the problem of finding ground states of Ising spin glasses and (2) optimization of random decomposable problems. I will also discuss preliminary results for hBOA on minimum vertex cover, which represents yet another important class of difficult optimization problems.
Bio: Martin Pelikan has been a researcher in genetic and evolutionary computation since 1995. He worked at the Slovak University of Technology at Bratislava, the German National Center for Information Technology at Sankt Augustin, the Illinois Genetic Algorithms Laboratory (IlliGAL) at the University of Illinois at Urbana-Champaign, and the Swiss Federal Institute of Technology (ETH) at Zurich. Currently, Pelikan is an assistant professor at the Dept. of Mathematics and Computer Science at the University of Missouri at St. Louis. Pelikan's most important contributions to genetic and evolutionary computation are the Bayesian optimization algorithm (BOA), the hierarchical BOA (hBOA), and the scalability theory for BOA and hBOA. BOA and hBOA combine machine learning with genetic and evolutionary algorithms to create optimizers that can solve broad classes of optimization problems in a robust and scalable manner with few or no parameters. BOA and hBOA are among the most advanced and powerful genetic and evolutionary algorithms.
Audience: General
Tools: PC Projector

Speaker: Martin Butz
Title: The XCS Classifier System: From Theory to Applications
Abstract: The XCS classifier system maybe the currently most successful and most promising representative of the family of rule-based evolutionary online learning systems, often referred to as Michigan-style learning classifier systems (LCSs). As all LCSs, XCS combines the strength of reinforcement learning with the generalization and search capabilities of genetic algorithms resulting in a flexible, online-learning and generalizing predictive learning system. This talk focuses on the questions how and when XCS works and, derived from these questions, how XCS can be designed and enhanced to solve diverse online reinforcement, control, or general predictive problems. A facetwise approach is proposed that partitions the learning biases of the system and analyzes the components separately respecting their possible interactions. The insights directly lead to a comprehensive application manual for XCS that outlines the most promising design (of XCS modules) for the problem at hand.
Bio: http://www-illigal.ge.uiuc.edu/~butz/ResumeMartinButz.html
Audience: General
Tools: PC Tools

Speaker: Xin Yao
Title: The iterated prisoner's dilemma (IPD)
Abstract: The iterated prisoner's dilemma (IPD) game has been used extensively in
modelling various real-world situations. This talk is concerned with the
evolutionary approach to the IPD game. Firstly, we generalise the game from
the classical 2 player case to N (N>2) players and investigate the impact of
the group size on the evolution. Secondly, we study a more realistic IPD game
where more than two levels of cooperations are allowed. Surprisingly, more
choices appear to discourage cooperation among players. Possible reasons for
this are discussed. Lastly, we introduce reputation into the IPD game and study
its impact on the evolution of cooperation. It turns out that the reputation
of a player is an important factor in encouraging cooperative behaviours.
Bio: Xin Yao is a professor of computer science from the University of Birmingham, UK. He is also a Distinguished Visiting Professor of the University of Science and Technology of China (USTC), P. R. China, and a visiting professor of three other universities. He is a Fellow of the IEEE, Editor-in-Chief of IEEE Transactions on Evolutionary Computation, an associate editor or an editorial board member of several other journals.
He is also the editor of the World Scientific book series on "Advances in Natural Computation". He was the winner of 2001 IEEE Donald G. Fink prize paper award and several other best paper awards. His research interests include evolutionary computation, neural network ensembles, global optimization, data mining and computational time complexity of evolutionary algorithms. He has more than 200 publications in those areas. He is currently the Director of the Centre of Excellence for Research in Computational Intelligence and Applications (CERCIA) (www.cercia.com), which is focused on applied research and knowledge transfer to the industry. He obtained his BSc in 1982 from USTC, MSc in
1985 from the North China Institute of Computing Technology (Beijing) and PhD in 1990 from USTC, all in computer science. He was a postdoctoral research fellow at the Australian National University in Canberra and CSIRO in Melbourne in 1990-92, and a lecturer, senior lecturer and associate professor at the Australian Defence Force Academy, University College, the University of New South Wales, in Canberra in 1992-99. He moved to Birmingham as a Professor of Computer Science in 1999.
Audience: General
Tools: PC Projector

Speaker: Keren Anne Chandler
Title: The Fr\"oberg-Iarrobino conjecture on infinitesimal interpolation
Abstract: Suppose that we wish to find the simplest approximation to a function f
in n variables, using experimental data on f and its partial derivatives
of order up to k on a collection of points in n-space.
Namely, we seek a polynomial P of minimal degree that fits the data set.
More generally, when there is no such polynomial in some degree m, we count
(dimensionwise, in the vector space of polynomials of degree at most m)
the number of obstructions to finding such a polynomial.
I shall describe geometric and algebraic viewpoints on this problem,
along with my recent verification of a conjecture of R. Froberg and
A. Iarrobino on counting obstructions for a random collection of points.
Bio: I obtained my Ph.D. at Harvard, with
advisor Joe Harris. My postdoctorate position was at the University of Chicago.
My main research interests are in algebraic geometry and complex geometry.
My work involves developing techniques on determining geometric
invariants of an algebraic manifold and of relative invariants of an embedding and
zero-dimensional schemes, as tools for analysis and as natural vehicles for fundamental
problems.
Audience:
Tools:

Speaker: Songtao Liu
Title: Wavelet Multiscale Methods and Numerical Solutions of PDEs
Abstract: In the talk we present our recent research achievements on wavelet multiscale methods for numerical solutions of PDEs. We first introduce a general approach to the construction of stable wavelets on irregular meshes in certain function spaces such as Sobolev spaces, which are crucial for many applications. Adaptive approximation by wavelet methods will be addressed, too. As an important application, we next discuss numerical solutions of singular perturbation problems and their connection with wavelet multiscale methods. We will present our results on optimal order approximation on our specially graded meshes. New challenges arising from numerical solutions of this type of problems by wavelet multiscale methods and possible solutions will also be discussed.
Bio: Songtao Liu is a P. T. Church postdoct fellow with the Department of Mathematics at Syracuse University. He obtained his Ph.D. degree in Mathematics from University of Alberta, Canada in 2004. His present research areas include:
Multiscale methods by wavelet approach and numerical solutions of singular perturbation problems on graded meshes.
Audience: General
Tools:

Speaker: Gualtiero Piccinini
Title: The Physical Church-Turing Thesis: Modest or Bold?
Abstract: The Church-Turing thesis (CT) may be stated as the thesis that the functions that are computable in the intuitive sense are computable by Turing Machines (TMs). To put it in Alan Turing's terms, CT pertains to what may be “naturally regarded as computable” (Turing 1936-7, p. 135). This formulation of CT should be uncontroversial. Unfortunately, here as elsewhere, people disagree on what is ‘intuitive’ or ‘natural’. As soon as we try to be more precise about the intuitive sense in which functions are computable, agreement ends. There is no consensus on how to discuss CT productively, and many pertinent issues remain unclear. This paper calls for more action on how to properly understand and evaluate CT and offers some suggestions on how progress might be possible.
Bio: I work primarily in philosophy of mind, with an eye to psychology, neuroscience, and computer science. My main current interests include computation, computational theories of mind, the relation between psychology and neuroscience, consciousness, and intentionality.In 2003, I graduated from the department of History and Philosophy of Science at the University of Pittsburgh. Between 2003 and 2005, I was a James S. McDonnell Post Doctoral Research Fellow in the Philosophy-Neuroscience-Psychology Program at Washington University in St. Louis. Since 2005, I am a member of the Philosophy Department at the University of Missouri – St. Louis.
Audience: General
Tools: Notebook and PC Projector

Speaker: Nikolay Romanovsky
Title: Free product decompositions in images of certain free products of groups
Abstract: In 1978 the author proved the following result:
Let G be a group which has a presentation with n generators and m relators, where m < n. Then some set of n-m generators freely generates a free group. The history of this result dates back to 1930, when Magnus published his Freiheitssatz, which is essentially the case of our statement in which m = 1. In 2004 J.S.Wilson generalized our result by proving a similar statement for any generating set for G. The proof was indirect, relying on another result of the author.
Here we (with J.S.Wilson) give a direct proof of a considerably more general result. Roughly speaking, the improvement consists of the replacement of the generating elements with subgroups.
Bio: Dr. Nikolay Romanovskiy is a lead researcher at Institute of Mathematics of Russian academy of Sciences in Novosibirsk. Also, he is a professor of Novosibirsk University.
His research interests are in the infinite group theory: soluble groups, groups presented in varieties by generators and relations, algorithmic problems for soluble groups, pro-p-groups and pro-finite groups, automorphisms of groups, algebraic geometry over soluble groups. Romanovskiy has proved the following generalization of Magnus Freiheitssatz: if a group G is presented by n generators and m relations with n > m, then some n-m of given generators freely generate in G a free subgroup.
Audience: Some understanding of groiup theory
Tools: none

Speaker: Yu-Ping Wang
Title: Genetic image analysis with wavelets
Abstract: Genetic imaging is an interdisciplinary area, which combines image processing techniques with the use of biochemical probes for the detection of chromosomal or genomic aberrations responsible for cancers and genetic diseases. Recent years have witnessed parallel and significant progress in both image processing and genetics. On one hand, revolutionary multiscale wavelet techniques have been developed in signal processing and applied mathematics in the last two decades, providing powerful tools for genetic image analysis. On the other hand, reaping the fruit of genome sequencing, high resolution genetic probes have been developed to facilitate accurate detection of subtle and cryptic genetic aberrations.
These probes hold great promise in cancer diagnosis and prognosis, however, they bring about computational challenges for the analysis of vast amounts of genetic data. In this talk, I will discuss the fruitful interaction between wavelets and genetic imaging. I will show how wavelets offer a perfect tool in addressing a variety of genetic imaging problems such as enhancement, compression, registration and classification. In particular, I will reveal an interesting connection between the use of wavelets in mathematics and genetics; the same word "subband" has already been used in genetics to describe the multiresolution banding structure of chromosomes, even before its appearance in mathematics and communications!
Bio: Wang Yu-Ping is an Assistant Professor of Computer Science and Electrical Engineering at University of Missouri-Kansas City. He received his PhD in Communication and Electrical System and his BSc and MSc in Applied and Computational Mathematics. He had postdoctoral experience working on MRI medical imaging at Washington University Medical School in St. Louis. Before joining academia, he worked in bio-industry for various genomic imaging problems under the NIH Small Business Innovation support in collaboration with Baylor College of Medicine in Houston. His experiences and expertise are primarily in three areas:
the study of wavelet theory and algorithm development, image processing and medical imaging application, and the development of genetic imaging instrument and bioinformatics tools.
Audience: General
Tools: PC Projector

Speaker: Weizhen Wang
Title: Statistical Inferences in Orthogonal Saturated Designs
Abstract: In orthogonal saturated designs, the number of observations is equal
to the number of effects. The traditional variance estimator, MSE, is then not available. The sparsity of effects principle is assumed so that it is reasonable to use those effect estimators with smaller absolute value to estimate the variance. The primary interest here is to identify nonzero effects without knowing how many and which effects are nonzero. There are two important issues involved: i) controlling the experiment error rates when searching for nonzero effects; using data to decide the number of effects to estimate the variance (so-called adaptive procedures). Our goal is to derive confidence intervals, tests and step-wise tests that strongly control the experimental error rates AND use the data adaptively. In this talk, we will first provide a brief review on the statistical inference procedures in orthogonal saturated designs, and then introduce our work on this topic since 2000.
Bio: Weizhen Wang graduated from Peking Univerity in 1990 with a B.S. and a M.S. He received his Ph.D. at Cornell University in 1995 under the supervision of Prof. Gene J.T. Hwang on topics of bioequivalence. He joined Wright State University in 1996 and now is Associate Professor of Statistics. He likes constructive, elegant and useful work. His research interest includes bioequivalence, saturated designs, categorical data analysis, quality control, nonparametric tests and protein structures. His research is currently supported in part by a NSF grant.
Audience:
Tools:

Speaker: Sharlee Climer
Title: A Formalization of the Use of Bounds with Applications in Biology and Engineering
Abstract: In this presentation, we reflect on the history of the use of bounds and observe that previous research has focused on bounds derived from relaxations of constraints. However, bounds can be derived by tightening constraints, adding or deleting variables, or modifying the objective function. A formalization of the use of bounds as a two-step procedure, called limit crossing, is presented.
We used limit crossing to produce an original search strategy called cut-and-solve. We compared a simple, generic implementation of cut-and-solve with state-of-the-art solvers for seven real-world Traveling Salesman Problem classes. Cut-and-solve was faster when solving large instances for five of the problem classes and able to solve larger (sometimes substantially larger) instances for four classes.
We are currently implementing cut-and-solve for the biological problem of inferring haplotypes. A haplotype is a set of nucleotide sites gathered from a stretch of DNA. Genetic association studies use haplotypes to identify correlations between diseases and genes. There are two components to the haplotype inferencing problem: determining a biologically sound model and devising an algorithm that is capable of solving this model within reasonable time allowances. Our model for haplotype inferencing is based on characteristics that can be expected from human populations and we are using cut-and-solve to reduce computation time.
Bio: Sharlee Climer is a doctoral candidate at Washington University in St. Louis. Her advisors are Weixiong Zhang, in the department of Computer Science, and Alan Templeton, in the department of Biology. She is an alum of UM-St. Louis, where she studied under Sanjiv Bhatia. She holds degrees in Physics (BA), Civil/Structural Engineering (BS), and Computer Science (BS and MS). Sharlee is a computational scientist with a focus on combinatorial optimization problems. She presented tutorials on her dissertation work at both the 19th International Joint Conference on Artificial Intelligence (IJCAI'05) and the 20th National Conference on Artificial Intelligence (AAAI'05). She has served on program committees for the 20th National Conference on Artificial Intelligence (AAAI'05) and the 2002, 2005, and 2006 Olin Conferences. Sharlee is the recipient of a National Defense Science and Engineering Graduat (NDSEG) Fellowship, an Olin Fellowship, and more than a dozen scholarships.
Audience: General
Tools: PC Projector

Speaker: Kang Chen
Title: Financial Herding in a Local Interaction Model
Abstract: Herd behavior in financial markets has been classified into three basic types of models: information-based herding and reputation and compensation-based herding. The first type of models is associated with information asymmetry between individual investors whereas the last two types are related to agency problems of fund managers or analysts. All three types of models are structured in a setting of global interactions in which individual decisions are revealed to all investors and aggregated information is publicly available. However, many social interactions of financial market players are local interactions. Informational cascades can be formed within a small group of investors who exchange information regularly, reputational effects and concerns for compensation can bring fund managers into social circles to make collective decisions, and emotional contagions are more likely to occur within a group of homogeneous agents. This paper presents a local interaction model which can be used to analyze the type of herd behavior originated from interactions within social circles. In this model, the formation of social groups follows a random cluster process which has time-varying connectivities. The group decision rule, which is a weighted combination of individual opinions and group opinions, is specified in a flexible manner to allow for various types of group interactions. Simulation results show that group herds can quickly develop into crashes or bubbles in the market if the weight assigned to group opinions is too high. When the weights are within the appropriate range such that the market itself can recover from booms and busts, the connectivity and activeness of agents are found to be important explanatory variables for the intensity of herding indicated by market excess demand.
Bio: Dr. Chen is a professor and the head of the division of economics at Nanyang Technological University, Singapore. He got his BS degree in mathematics from Xiamen University, China and Ph degree in economics from the University of Maryland.
Audience: Math/CS grad students, math, ecomimics, finance
Tools: laptop and PC projector

Speaker:
Title:
Abstract:
Bio:
Audience:
Tools: