The Institute of Statistics and Computational Intelligence (ISCI)

Recent Activities

Seminar: Hierarchical Segmentation of Multimodality MRI Human Brain Tumors - Towards a Comprehensive Characterization of Tumors via Structural and Diffusion Tensor Magnetic Resonance Imaging: Speaker: Dr. Cai Hong-Min; Time: 2008. 05.21, 16:00-17:00; Abstract: The talk aims at creating a multi-modal profile of tissue components that will not only help in delineating tumor and edema from healthy tissue, distinguishing between enhancing and non-enhancing tumors, but also produce a probabilistic characterization of tissue around the tumor to determine abnormal regions that may have a tendency to convert to tumor in the future. The multimodality profile is generated by a combination of five structural MR images, FLAIR, T1, Gadolium enhanced T1 (GAD), DWI and B0, and two scalar maps, including Fractional Anisotropy(FA) and Apparant Diffusion Coefficient (ADC) computed from diffusion tensor images (DTI), creating a seven-dimensional intensity feature vector for each voxel. The tumor-grade- specific ground truth identified by doctors, are trained through a newly proposed hierarchical Support Vector Machine (SVM) based on spatial and texture features, the system achieves near-perfect characterization of tumor components (enhancing and non-enhancing), edema and healthy tissue with a 90 – 95 % classification rate together with almost 0 false positive rate. In addition to this hard tissue segmentation, the framework also provides probability profile for tissue, indicating unhealthy regions that may have a tendency to convert to tumorous tissue, hence providing a better characterization of the resection margin. The classifiers, trained on a dataset of 22 patients, can be applied to a new dataset with a success rate of 80% for classification, as has been tested using a leave-one-out paradigm on these 22 datasets. The multimodality processing pipeline that we have designed is general and is applicable to any study that has multi-modal data acquisition; About Speaker:
Dr. CAI Hongmin received his B.S and M.S degrees both from Harbin Institute of Technology, China. He obtained his PhD degree in applied mathematics from the University of Hong Kong in 2007. His research interest includes biomedical image analysis and biometric recognition.

Seminar: Two-level Multiple Discrete – Continuous Model: Speaker: Dr. Jiu-Kun Li; Time: 3:00-4:00pm,Wednesday, May 14th, 2008; Absract: This paper developed a two-level multiple discrete-continuous model. Details of the model and its estimation method are explained and illustrated by an example. The model may be used to investigate activity choices and time allocations in activity-travel behavior modeling; product and service choices and money allocation in marketing.; About Speaker:
Dr. Li Jiu Kun is currently a visiting scholar in UIC. She obtained her PhD degree from The University of Hong Kong. Her research interests are statistics and its applications, optimization, logistics and supply chain management, and transportation modeling.

Seminar: Camera Calibration with Spheres: Linear Approaches: Speaker: Dr. Amy Zhang; Time: 3:00-4:00pm,Wednesday, Apr. 16th, 2008; Abstract: This paper addresses the problem of camera calibration from spheres. By studying the relationship between the dual images of spheres and that of the absolute conic, a linear solution has been derived from a recently proposed non-linear semi-definite approach. However, experiments show that this approach is quite sensitive to noise. In order to overcome this problem, a second approach has been proposed, where the orthogonal calibration relationship is obtained by regarding any two spheres as a surface of revolution. This allows a camera to be fully calibrated from an image of three spheres. Besides, a conic homography is derived from the imaged spheres, and from its eigenvectors the orthogonal invariants can be computed directly. Experiments on synthetic and real data show the practicality of such an approach.

Seminar: Modeling of lunar Helium-3 distribution with remote sensing data from Chang’E-1: Speaker: Prof. Tsang Kang Too; Time: 2008.04.02, 16:00-17:00; Abstract: Sometime around 2050, due to shortage of fossil fuels and their green house effects, human society has to make the transition to other energy sources. Nuclear energy can provide a temporary solution but short of a permanent one because of the limited supply of fissible uranium and the proliferation problem that comes with it. Other forms of alternative energy like solar, wind or biofuels are either too expensive or diffuse in power density that they can only play a supplementary role. The only viable energy source that can assume a dominant role is nuclear fusion, if current international R&D effort in fusion technology is carried out as planned and achieving its goal successfully.[Ref1]

Workshop for Statistics Students’ Career Development 2008: Date: 30/3/2008 (Sunday); Time: 2-4pm; 　; Invited speakers:; (I) Professor Minggao Gu, Department of Statistics, The Chinese University of Hong Kong; (II) Iris and Timothy, graduate from Math department, HKBU, currently working in standard chartered bank and DBS bank; (III) Prof. Cheng, Visiting Professor, United International College, Zhuhai, China and Academia Sinica, Taipei, Taiwan; 　; Title:; (I) Kelly Formula for Investment; (II) Sharing as a university graduate in Statistics; (III) The Statistical Market in Taiwan; 　; Abstract:; (I) In this talk I will introduce the concept of probability, expectation and the optimal amount of investment formula in an advantaged game known as the Kelly formula. Examples of real life random events such as Mark 64 and horse racing will be introduced.; (II) We will share stories of us and our former classmates after we graduate from Mathematics Department, HKBU.; (III) The market for statisticians has always been very good in Taiwan, even in the recent years of low economic growth in the early twenty-first century. There are several types of job in the Taiwan market for statistics graduates that keep on consistent demands, and a few new types emerge with growing demands.

Seminar：Query Result Ranking over E-commerce Web Databases: Speaker: Dr. Su Wei-Feng; Time: 2008. 03.19, 16:00-17:00; Abstract: Query Result Ranking over E-commerce Web Databases Abstract: To deal with the problem of too many results returned from an E-commerce Web database in response to a user query, this paper proposes a novel approach to rank the query results. Based on the user query, we speculate how much the user cares about each attribute and assign a corresponding weight to it. Then, for each tuple in the query result, each attribute value is assigned a score according to its “desirableness” to the user. These attribute value scores are combined according to the attribute weights to get a final ranking score for each tuple. Tuples with the top ranking scores are presented to the user first. Our ranking method is domain independent and requires no user feedback. Experimental results demonstrate that this ranking method can effectively capture a user’s preferences.

Seminar: A learning-based approach to algorithm selection: Speaker: Dr. Guo Haipeng; Time: 3:00-4:00pm,Wednesday, Mar.5th, 2008

Seminar: Some Ideas on Uniform Design of Mixture Experiments: Speaker: Mr. Ning Jian-Hui; Time: 2:00-3:00 p.m., Wednesday, Dec. 05, 2007

Four-Points Innovation Laboratory Collaborates with local Enterprise: The Four-Points Innovation Laboratory (FPIL) under the Institute of Statistics and Computational Intelligence (ISCI) was set up in September 2007 by Prof. Fang Kai Tai. In mid-September and early October, the FPIL had their first collaboration program with a famous Zhuhai enterprise, the Zhuhai Jiuzhou Port Administration Group Co. Led by Dr. Peng Xiaoling and Dr. Dennis, Zhang. On Nov. 1, executives from the Jiuzhou Port Group attended the project report meeting at UIC. Mr. Mo Nenglin, Chairman of the JIuzhou Port Group commented that the survey was conducted carefully and professionally. He believed that the data collected was helpful to the Jiuzhou Port Group in various aspects. They were expecting to have further collaboration with the FPIL.; more details: UIC News Zhuhai TV News (video)

Seminar: Reliability and Statistics: Speaker: Prof. Y. L. Tong, Georgia Institute of Technology; Time: 2:00-3:00 p.m., Wednesday, Nov. 21, 2007; [poster] [photos]

Seminar: Star Matching and Distance Two Labelling: Speaker: Dr. Wen-Song Lin; Time:2:00-3:00 pm, Wednesday, Nov 7, 2007

Seminar: Modeling of Covariance Structures in Generalized Estimating Equations for Longitudinal: Speaker: Dr. Hua-Jun Ye; Time:2:00-3:00 pm, Wednesday, Oct 31, 2007

Seminar: Inverse Optimization Theory, Methods and Applications: Speaker: Prof. J.Z. Zhang; Time: 2:00-3:00 pm, Wednesday, Oct 24, 2007
Abstract: In this talk, the speaker shall first introduce the concept of inverse optimization and formulate such problems mathematically. Some applications shall be followed to justify usefulness of the study. Then some general methods shall be discussed. Furthermore, in order to achieve high efficiency, special methods will be given to deal with each type of particular inverse optimization problems. Some extensions and further development, such as system improvement problems and partial inverse optimization problems, shall be mentioned to conclude this talk.

Seminar: Testing Hypotheses with Contingency Tables Analysis: Speaker: Philip E. Cheng, Institute of Statistical Science, Academia Sinica; Time: 2:00-3:00 pm, Wednesday, Oct 17, 2007
Abstract: A recent study of log-likelihood identity in terms of mutual information yields useful applications in statistical inference. For testing association in a 2 x 2 table (Pearson, 1904; Fisher, 1934), it establishes power analysis using likelihood ratio (LR) test that can not be achieved by other existing methods. An extended identity provides power evaluations for testing inhomogeneous odds ratios of three-way tables. A problem of the celebrated CMH test (1954, 1959) is examined by the geometry of an information identity, and resolved by using an omnibus LR test together with a family of two-step LR tests. In contrast to the hierarchical log-linear models, information identities lead to developing a natural family of linear information models (LIM). Empirical studies of two-way and high-way contingency tables are used to illustrate the new statistical inference at college textbook level.; [photos]

Seminar: Should it be Our Survey Software for the Four Points Innovation Lab (FPIL)?: Speaker: Ms Dwight Thé, Division Science & Technology, UIC; Time: 2:00pm, Wednesday, Sept 19, 2007
Abstract: Survey research, broadly conceived as the practice of collecting sample records in order to learn something about an entire population, is likely millennia old. However, the formal systematization of survey research did not begin until shortly after World War II; the effort was spearheaded by three groups: University of Michigan; Columbia University; and the U.S. Census Bureau. Today, survey research constitutes a truly inter-disciplinary endeavor that is fraught with numerous potential complicating factors. Some key issues include how to best design the instrument, conduct the survey, analyze the data, and report the results based on the research questions, the population surveyed, and the type of survey used. As the world's "global economy" (of which China is a major contributor) continues to move towards a "market-based economy", the overall importance of survey-centered, market research is likely to increase. Given the diversity of the survey method landscape and the complexity of the questions asked, it is very important to select "the right tool for the job". In this regard, The Survey System (TSS) 9.5 has been called "the market researcher's surveying package." The question at hand is whether or not TSS 9.5 lives up to its name and more specifically "Should it be our survey software for the FPIL?" The seminar will be organized into three parts: (i) overview of the software; (ii) demonstration of some key features; and (iii) discussion of the strengths and weaknesses of the product.

SUMMER SCHOOL 2007 FOR BUSINESS STATISTICS -------A CERTIFICATE PROGRAM: Objective: The program provides a comprehensive education and training for students in business statistics, market analysis, data mining and decision-making. Specifically it is designed to train students how to effectively and skillfully apply statistics to a number of areas in business, such as data centered management, financial management, market research, bank credit management, customer relationship management and risk management, etc.; Period of Study: 01-31 July, 2007; Applicants: All undergraduates in UIC, irrespective of major, are welcome.; More Details: Summer School [poster]

Seminar: Bayesian Statistics and Computation: An Advanced Tool for Quantitative Decision Making: Speaker: Xiao-Li Meng, Department of Statistics, Harvard University; Time: 4:00-6:00 p.m., Thursday, July 12, 2007
Abstract: In business, finance, and other endeavors that require constant decision making, effectively combining quantitative information with subjective judgment is critical in achieving success and maintaining competitiveness. Bayesian methodology is ideally suited for this task because it provides a coherent and rigorous probability-based framework for statistical analysis with explicit subjective input via prior specification. The first part of this tutorial demonstrates the modeling aspects of Bayesian Statistics via a couple of examples on airline stocks and insurance mortality. Aided by movies, the second part introduces Markov chain Monte Carlo (MCMC), a general class of simulation methods that have revolutionized the Bayesian Statistics since 1990s because they made it possible to fit many realistic Bayesian models that defeat traditional computational methods.; [poster]

Seminar: A Path to UC, UCLA and Biostatistics: Speaker: Gang Li, Ph.D., Professor of Biostatistics, University of California at Los Angeles; Time: 2:00-4:00 p.m., Thursday, July 5, 2007
Abstract: The University of California (UC), home to more than 209,000 students, is "the heart and soul of California, and its future" (Robert Dynes, UC President). In this talk I will give a general description of the higher education system in California, with more detailed information on UC, UCLA, and how to get there. The second part of the talk will give you a peek into the field of biostatistics through the eyes of a biostatistician.; [poster]

Seminar: Variable Selection in Semiparametric Regression Modeling: Speaker: Runze Li, Pennsylvania State University; Time: 3:00-5:00 Monday, May 14, 2007
Abstract: In this talk, I will introduce how to select significant variables in semiparametric modeling. Variable selection for semiparametric regression models consists of two components: model selection for nonparametric components and selection of significant variables for the parametric portion. Thus, semiparametric variable selection is much more challenging than parametric variable selection (e.g., linear and generalized linear models) because traditional variable selection procedures including stepwise regression and the best subset selection now require separate model selection for the nonparametric components for each submodel. This leads to very heavy computational burden. In this paper, we propose a class of variable selection procedures for semiparametric regression models using nonconcave penalized likelihood. We establish the rate of convergence of the resulting estimate. With proper choices of penalty functions and regularization parameters, we show the asymptotic normality of the resulting estimate, and further demonstrate that the proposed procedures perform as well as an oracle procedure. A semiparametric generalized likelihood ratio test is proposed to select significant variables in the nonparametric component. We investigate the asymptotic behavior of the proposed test and demonstrate that its limiting null distribution follows a chi-squared distribution, which is independent of the nuisance parameters. Extensive Monte Carlo simulation studies are conducted to examine the finite sample performance of the proposed variable selection procedures.; [poster]

Seminar: Online Evolutionary Algorithms

Speaker: Prof. Alfredo Milani, Dept. of Mathematics & Computer Science, University of Perugia

Time: 3:00-4:00pm, Monday, April 23, 2007

Abstract: The problem of providing services to a mass of anonymous users whose goals and needs evolve over the time in unpredictable way is common to many application domains of information technology, from web to mobile communication from news broadcasting to advertising. The basic idea of the proposed approach is to use the evolutionary scheme of genetic algorithms in order to dynamically adapt to the audience and service evolution while optimizing the global systems performance.

[poster]

Seminar: On Multifractal Property of Joint Distributions and Its Application to Bayesian Network Inference

Speaker: Dr. Hai-Peng Guo
Time: 2:00 pm - 4:00 pm, Oct. 12th(Thursday) , 2006

Abstract: Bayesian Belief Network (BBN) is the currently dominant method for reasoning under uncertainty in Artificial Intelligence. Inferences with BBNs are either optimization, or marginalization, or both on the joint probability space.

We demonstrate that the joint probability distribution of a BBN is a multifractal in its most general form - a random multinomial multifractal. With sufficient asymmetry in individual prior and conditional probability distributions, the joint distribution is not only highly skewed, but also stochastically self-similar and has clusters of high-robability instantiations at all scales. Inspired by the multifractal properties, a sampling-and-search algorithm for finding the Most Probable Explanation (MPE) in BBN is developed and tested. The experimental result shows that these multifractal properties provide good heuristic for solving the NP-hard MPE problem.

[poster]

Seminar: Modelling Nonlinear Dynamics of Time Series with Physiological Applications

Speaker: Dr. Yi Zhao
Time: 2:00 pm - 4:00 pm, Sep. 21th, 2006

Abstract: Over-fitting has long been recognized as a problem endemic to models with a large number of parameters. The usual method of avoiding this problem in neural networks is to avoid fitting the data too well. In our research project we propose an alternative, information theoretic criterion to determine the number of neurons in the optimal model. When applied to the time series prediction problem we find that models, which minimize the description length of the data, both generalize well and accurately capture the underlying dynamics.

The optimal neural network based on the estimation of the minimum description length and the surrogate data method are then combined to apply to human cardiac systems. The experimental results present that pulse waveform measure on the lateral arterial artery (wrist) is equivalent to pulse measure on the fingertip, and pseudo-periodic determinism exists in both ECG and pulse time series but human ECG and pulse data do not conform to the same deterministic process. The human ECG data might provide additional information of the human body than the pulse.

[poster]

Seminar: A Learning-Based Approach for Algorithm Selection

Speaker: Dr. Hai-peng Guo
Time: 2:00 pm - 4:00 pm, Sep. 14th, 2006

Abstract: The algorithm selection problem aims at selecting the best algorithm for a given computational problem instance according to some characteristics of the instance. In this talk we will look at a machine learning-based inductive approach using experimental algorithmic methods and machine learning techniques to solve the algorithm selection problem. Experimentally, we have applied the proposed methodology to algorithm selection for the MPE (Most Probable Explanation) problem and the results will be presented.

[poster]

Seminar: Data Mining in Mass Spectral Databases

Speaker: Dr. Ping He
Time: 2:00 pm - 4:00 pm, Sep. 12th, 2006
Abstract: with the growth of chemical measurement and modern information technology, more and more huge databases containing a large amount of chemical compounds information are established, for example, mass spectral database. How to discover knowledge hidden in huge collections is a big challenge. In this talk we will discuss data mining in mass spectral (MS) database. We will review the application of library searching and classification methods, traditional or modern, on MS database and propose new methods to analyze MS database.

[poster]

Seminar: A Condition Number for (LP), (SOCP), (SDP)

Speaker: Dr. Dennis Cheung
Time: 2:00 pm - 4:00 pm, Sep. 7th, 2006
Venue: A103, LiZe Building, Zhuhai Campus of Beijing Normal University

[poster]

Seminar: Uniform Experimental Design and its Recent Development

Speaker: Prof. Kai-Tai Fang
Time: 2:00 pm - 4:00 pm, Sep. 5th, 2006
Venue: A103, LiZe Building, Zhuhai Campus of Beijing Normal University

[poster]

Seminar: Distinguished Lecture Series in Statistics and Computer Science: Invited speakers from academia, industries and government will give public lectures at UIC.

　