IBM Research welcomes members of the research community to our seminars. To ensure compliance with IBM security guidelines, we request you to contact the seminar host in advance. When you arrive at the Research lab, please provide the host's name to the receptionist.
| The Use of Statistical Process Control in Health Care Monitoring and Public Health Surveillance | ||||
| William H. Woodall | On: | 26-Jun-2008 11:00 AM - 12:00 PM | ||
| Professor | At: | Watson Research Center (Yorktown), Room 20-043 | ||
| Department of Statistics, Virginia Tech, Blacksburg, VA | Host: | Emmanuel Yashchin | ||
Abstract: There are many applications of quality control charts in health care monitoring and in public health surveillance. In fact, statistical process control (SPC) forms the basis of many disease surveillance systems. In this overview presentation, these types of applications are introduced with discussion of some of the differences from the standard industrial monitoring environment. The advantages and disadvantages of some of the charting methods and performance metrics proposed in the health care and public health areas are considered. Public health data often include spatial information as well as temporal information, so the public health applications can be even more challenging than industrial applications. There are many research opportunities available in the health-related use of control charts. PC. Speaker biography: William Woodall is a professor of statistics in the Department of Statistics, Virginia Tech. He received a Ph.D. degree in Statistics from Virginia Tech in 1980. He was Russell Professor of Statistics in the Department of Management Science and Statistics at the University of Alabama (1995-2000), C&BA Board of Visitors Research Professor of Statistics (1992-1994), and Director of Applied Statistics Program (1993-1998). His research Interests include Statistical quality control and improvement, including all aspects of control charting; health-related monitoring and prospective public health surveillance; critiques of fuzzy logic. Among many awards he received are Shewhart Medal from American Society for Quality in 2002, Jack Youden Prizes for best expository papers in Technometrics in 1995 and 2003, Brumbaugh Awards from American Society for Quality in 2000 and 2006, Best Paper Award for IIE Transactions on Quality and Reliability Engineering in 1997, and Ellis Ott Foundation Award for best paper in quality control in 1987. He is Fellow of the American Statistical Association, Fellow of the American Society for Quality, and elected member of the International Statistical Institute. | ||||
| Bayesian Hierarchical Model for Integrating Multi-resolution Metrology Data | ||||
| Haifeng (Heidi) Xia | On: | 24-Jan-2008 11:00 AM - 12:00 PM | ||
| PhD Candidate | At: | Watson Research Center (Yorktown), Room 40-100 | ||
| Department of Industrial and Systems Engineering, Texas A&M University - College Station | Host: | Emmaunuel Yashchin | ||
Abstract: Nowadays, engineers and scientists have multiple sensor technologies available to make almost any measurement. However, different sensors give information at different scales and resolutions. To achieve both efficiency and accuracy, different types of sensors often have to work together for a single measuring task, which results in multi-resolution data. This talk will present a Bayesian hierarchical model to integrate multi-resolution metrology data for inspecting geometry quality of manufactured parts. The basic idea is to denoise low-resolution data first and then to connect the filtered low-resolution dataset with the high-resolution dataset. This talk will also address the problem of the misalignment between the low- and high-resolution datasets. The solution includes an optimization procedure to roughly align datasets of different resolutions and a neighborhood linkage model to link each high-resolution data point with multiple low-resolution data points. The proposed Bayesian hierarchical model is tested using two-resolution simulated data as well as the measurements of a milled sine-wave part from two coordinate measuring machines of different resolutions. The resulting analysis improves the prediction accuracies by more than 40% over methods that use only single-resolution data. Speaker biography: Haifeng (Heidi) is currently a Ph.D. candidate in Industrial and Systems Engineering at Texas A&M University. She received an M.S. degree in 2002 and a B.S. degree in 2000 from Tianjin University, China. Her research focuses on multi-scale/multi-resolution modeling and analysis as well as its applications in dimensional quality control, nanotechnology and remote sensing. She received the 2007 “Quality, Statistics and Reliability” Best Student Paper Award sponsored by INFORMS. Haifeng (Heidi) is a member of INFORMS (the Institute for Operations Research and the Management Sciences) , IIE (Institute of Industrial Engineers) and ASA(American Statistical Assoication). | ||||
| Bayesian Functional Data Analysis for Computer Model Validation | ||||
| Fei Liu | On: | 20-Nov-2007 11:00 AM - 12:00 PM | ||
| Assistent Professor | At: | Watson Research Center (Yorktown), Room 40-100 | ||
| Department of Statistics, University of Missouri at Columbia | Host: | Alejandro Veen | ||
Abstract: Functional data analysis (FDA) -- inference on curves or functions -- has wide application in statistics. An example of considerable recent interest arises when considering computer models of processes; the output of such models is a function over the space of inputs of the computer model. The output is functional data in many contexts, such as when the output is a function of time, a surface, etc. A nonparametric Bayesian statistics approach, utilizing separable Gaussian Stochastic Process as the prior distribution for functions, is a natural choice for smooth functions in a manageable (time) dimension. However, direct use of separable Gaussian stochastic processes is inadequate for irregular functions, and can be computationally infeasible in high dimensional cases. In this talk, we will develop and extend several Bayesian FDA approaches for high dimensional irregular functions in the context of computer model validation, tailored to interdisciplinary problems in engineering. Speaker biography: Fei Liu is an assistant professor at University of Missouri-Columbia. She graduated from Duke University in 2007, under the supervision of professor James Berger. Her primary research interest is the development of Bayesian methodology, with emphasis on interdisciplinary collaborations. | ||||
| Information Theory Today | ||||
| Sergio Verdu | On: | 16-Nov-2007 01:00 PM - 02:30 PM | ||
| Professor | At: | Watson Research Center (Yorktown), Room 20-043 | ||
| Princeton University | Host: | Luis Lastras | ||
Abstract: Founded by Claude Shannon in 1948, information theory has taken on renewed vibrancy with technological advances that pave the way for attaining the fundamental limits of communication channels and information sources. Increasingly playing a role as a design driver, information theory is becoming more closely integrated with associated fields such as coding, signal processing and networks. In this talk, Prof. Verdu will review current research trends in the field as well as some of its longstanding open problems. Speaker biography: Sergio Verdú is a professor of Electrical Engineering at Princeton University where he teaches and conducts research on information theory in the Information Sciences and Systems Group. He is also affiliated with the Program in Applied and Computational Mathematics. A native of Barcelona, Sergio Verdú received a Telecommunications Engineering degree from Polytechnic University of Catalonia, Barcelona, Spain, in 1980 and a Ph.D. degree in Electrical Engineering from the University of Illinois at Urbana-Champaign in 1984. Conducted at the Coordinated Science Laboratory of the University of Illinois, his doctoral research pioneered the field of Multiuser Detection. Sergio Verdú was elected Fellow of the IEEE in 1992 and member of the U. S. National Academy of Engineering in 2007. He received the 2000 Frederick E. Terman Award from the American Society for Engineering Education, and the IEEE Third Millennium Medal in 2000. In 2005, he received a Doctorate Honoris Causa from the Polytechnic University of Catalonia. He is the recipient of the 2007 Claude E. Shannon Award. In 1998, Cambridge University Press published his book ``Multiuser Detection.'' His papers have received several awards: the 1992 IEEE Donald Fink Paper Award, the 1998 Information Theory Outstanding Paper Award, a IEEE Information Theory Golden Jubilee Paper Award, the 2000 Paper Award from the Japan Telecommunications Advancement Foundation, the 2002 Leonard G. Abraham Prize Award in the field of Communications Systems and the 2007 IEEE Joint Communications/Information Theory Paper Award. Sergio Verdú has served as associate editor of the IEEE Trans. on Automatic Control, and as associate editor for Shannon Theory of the IEEE Transactions on Information Theory. He served as president of the IEEE Information Theory Society in 1997. He is currently editor-in-chief of Foundations and Trends in Communications and Information Theory. He has held visiting appointments at the Australian National University, the Technion-Israel Institute of Technology, and the University of Tokyo. In 1998, he was a visiting professor at the Electrical Engineering and Computer Science Department of the University of California, Berkeley, and in 2002 he held the Hewlett-Packard Visiting Research Professorship at the Mathematical Sciences Research Institute, Berkeley. He is a member of the Scientific Advisory Board of Telefónica I+D. | ||||
| Subject-Adaptive, Real-Time Sleep Stage Classification | ||||
| Wanli Min | On: | 18-Oct-2007 11:00 AM - 12:00 PM | ||
| Research Staff Member | At: | Watson Research Center (Yorktown), Room 20-059 | ||
| Department of Mathematical Sciences, IBM Watson Research Center | Host: | Ta-Hsin Li | ||
Abstract: Sleep stager performs one of the most important steps in sleep analysis: classifying sleep recordings into different sleep stages. We introduce an approach to online sleep stage classification using electroencephalogram (EEG) signal. Using sleep recordings from human beings and birds, we show that the average classification accuracy of our sleep stager almost approaches the theoretical limit and outperforms existing systems. The approach is readily applied to new subject with limited EEG signal in a Bayesian flavor. Joint work with Gang Luo, IBM T. J. Watson Research Center. Speaker biography: Wanli Min is a Research Staff Member in the Department of Mathematical Sciences at the IBM T. J. Watson Research Center. He joined IBM in 2005 after receiving a PhD degree in statistics from the University of Chicago. | ||||
| The Minimal Belief Principle: A New Method for Parametric Inference | ||||
| Chuanhai Liu | On: | 20-Sep-2007 11:00 AM - 12:00 PM | ||
| Professor | At: | Watson Research Center (Yorktown), Room 20-059 | ||
| Department of Statistics, Purdue University | Host: | Ta-Hsin Li | ||
Abstract: It is recognized that given a postulated sampling model, reasoning for statistical inference (about a particular realization of random variables) should be different from reasoning for data generation. The classical belief in distributional invariance of pivotal variables does not distinguish these two types of reasoning processes and is thus often too strong to be believable. A ``minimal belief'' (MB)-based method is considered for parametric inference. The intuitive idea is that beliefs with higher believability can be obtained from the classical belief by making it weaker. The MB principle serves as a general guidance rather than a precisely defined mathematical term. It may take different mathematical forms for sampling models of different data structures. Technically, the proposed method is built on the Dempster-Shafer (DS) theory, which is known as the successor of Fisher's fiducial argument. The MB posteriors for general single-parameter distributions and certain multiparameter distributions are obtained in closed form. The method is illustrated with a variety of examples, including the simple test of significance, the Behren-Fisher problem, the multinomial model, and a normal model involving a large number of unknown location parameters. The last example offers an MB perspective of often-crude Bayesian and related shrinkage techniques, which have been considered necessary in the last half a century. It is also shown that Markov chain Monte Carlo methods, which have made the Bayesian methodology computationally attractive, can be developed for MB-based analysis. Speaker biography: Chuanhai Liu is Professor of Statistics at Purdue University. He received his MS in Probability and Statistics from Wuhan University in 1987 and Ph.D. in Statistics from Harvard in 1994. He was Member of Technical Staff at Bell Laboratories from 1995 to 2005. His research contributions are in statistical modeling and computation, including Expectation-Maximization type of algorithms and Markov chain Monte Carlo methods. He received the 2000 American Statistical Association award for the most outstanding statistical application (with A. Gelman and G. King) and the 2000 Frank Wilcoxon award for the best practical application paper published in Technometrics (with D. Sun). He is an elected member of International Statistical Institute and has been recently named Fellow of the American Statistical Association. | ||||
| Computer Experiments: An overview and some design strategies | ||||
| William Notz | On: | 13-Sep-2007 11:00 AM - 12:00 PM | ||
| Professor | At: | Watson Research Center (Yorktown), Room 26-245 | ||
| Department of Statistics, Ohio State University | Host: | Ta-Hsin Li | ||
Abstract: In this talk I will begin with an overview of computer experiments and the models that have become popular in their analysis. I will then discuss three ongoing research projects involving experimental design. The first involves sequential designs for deciding where to observe the computer code in order to produce a statistical predictor with good overall fit to the code. The second involves sequential designs for deciding where to observe the computer code for purposes of calibration (deciding what values of certain tuning parameters produce the best agreement between the code and actual data from the physical process the code simulates). The third involves sequential designs for estimating quantiles of the induced distribution on the output of the computer code given distributions for the inputs to the code. Speaker biography: Dr. Notz is a Professor of Statistics at the Ohio State University. He received his PhD degree from Cornell University in 1978. He was on the faculty of the Purdue University from 1978 to 1984, and has been with the Ohio State University since 1984. He served as acting Chair of the Statistics Department, Associate Dean of the College of Mathematical and Physical Sciences, and Director of the department's Statistical Consulting Service. He also served on the editorial board of many journals including the Journal of the American Statistical Association and Technometrics. He is a Fellow of the American Statistical Association. Dr. Notz's research interests include optimal experimental design, robust procedures, and computer experiments. | ||||
| L-Moments: Inference for Distributions and Data Using Linear Combinations of Order Statistics | ||||
| Jonathan R. M. Hosking | On: | 16-Aug-2007 01:30 PM - 02:30 PM | ||
| Research Staff Member | At: | Watson Research Center (Yorktown), Room 40-200 | ||
| Department of Mathematical Sciences, IBM Watson Research Center | Host: | Ta-Hsin Li | ||
Abstract: L-moments are expectations of certain linear combinations of order statistics. They form the basis of a general theory which covers the summarization and description of theoretical probability distributions, the summarization and description of observed data samples, estimation of parameters and quantiles of probability distributions, and hypothesis tests for probability distributions. L-moments are in analogous to the conventional moments, but are more robust to outliers in the data and enable more secure inferences to be made from small samples about an underlying probability distribution. They can be used for estimation of parametric distributions, and can sometimes yield more efficient parameter estimates than the maximum-likelihood estimates. This talk gives a general summary of L-moment theory and methods, describes some applications ranging from environmental data analysis to financial risk management, and indicates some recent developments on nonparametric quantile estimation, "trimmed" L-moments for very heavy-tailed distributions, and L-moments for multivariate distributions. Speaker biography: Jonathan R. M. Hosking has been with the IBM Research Division, Yorktown Heights, N.Y., since 1986. He is a Research Staff Member in the Statistics group of the Mathematical Sciences Department. Previously he was with the Institute of Hydrology, Wallingford, England (1979-86). He holds an M.A. in Mathematics from Cambridge University and a Ph.D. in Statistics (time series analysis) from Southampton University (1979). He is the author of one book and over 50 research papers, covering such subjects as feature selection and ranking in classification problems, statistics for summarizing data samples (L-moments), "long-memory" time-series models -- useful for modelling series with complicated structure related to Mandelbrot's "fractals" -- and estimating the frequency of extreme environmental events. He has experience in applying statistical methods in financial modelling, business forecasting, and civil and environmental engineering. | ||||
| Hilbert-Huang Transform and Its Application | ||||
| Hee-Seok Oh | On: | 8-Aug-2007 11:00 AM - 12:00 PM | ||
| Associate Professor | At: | Watson Research Center (Yorktown), Room 40-200 | ||
| Department of Statistics, Seoul National University, Korea | Host: | Ta-Hsin Li | ||
Abstract: A signal in real world usually reflects complex phenomena. One may have difficulty in extracting and interpreting information embedded in such a signal. A natural way to reduce the complexity is to decompose the original signal into several components of simple form in a sense, and then to analyze period or frequency of each component. Utilizing the periodicity has been a popular approach to analyze a signal in many areas such as economics, engineering and physics. For example, sun-spot data is fluctuated over about 11 year and 85 year. Huang et al. (1998, 1999) proposed a data-adaptive decomposition method called empirical mode decomposition (EMD) and then applied Hilbert spectral analysis to decomposed signals alled intrinsic mode function (IMF). Huang et al. (1998, 1999) named these two step procedures as Hilbert-Huang transform (HHT). Due to its robustness to presence of non-linearity and non-stationarity, HHT has been applied to various fields. In this talk, we introduce the HHT and demonstrate the promising capability of HHT for various statistical problems. Speaker biography: Dr. Oh received his PhD degree in statistics from Texas A&M University at College Station in 1999. He was a Post-doctoral Fellow at the University of Bristol, UK, from 1999 to 2000 and a Visiting Scientist at the National Center for Atmospheric Research (NCAR), Boulder, CO, from 2000 to 2002. He was an Assistant Professor in the Department of Mathematics and Statistics at the University of Alberta, Canada, from 2002 to 2004. He is now an Associate Professor in the Statistics Department at the Seoul National University. Dr. Oh's current research interests include multiscale methods in statistics, functional estimation, time series analysis, spatial statistics, and statistical climatology. | ||||