On this page:
SPES Session Descriptions
SPES Session Schedule
SPES Roundtable Discussions
Continuing Education
Poster Sessions
Remember Christmas time when you wanted to sneak a look at the presents under the tree? Well, now you can take a peek at what SPES will be bringing to the Joint Statistical Meetings in Baltimore, August 8-12. I believe we have a really great program this year and I hope you concur. We have 5 Invited sessions, 4 Topic (used to be called Special) Contributed sessions, and 7 Regular Contributed sessions. This amounts to 84 individual talks!
The invited sessions cover a wide breadth of statistical topics; reliability theory, spatial statistics, chemometrics, Bayesian methods, and multiple response experimentation. The session Statistical Modeling of Growth and Degradation Processes was organized by Joanne Wendelberger and is concerned with processes which age or change as a function of time or some other aging variable. The session is intended to motivate the development of statistical theory to aid in the modeling of growth and degradation processes. The session Spatial Methodology for Minefield Detection was organized by Noel Cressie and is concerned with the serious problem of minefield detection. The use of mines in regional conflicts is a cheap, but effective way to provide defense of land or sea. Unfortunately, long after the conflict has stopped, the mines can remain an indiscriminate danger to all who happen into uncleared minefields. Clearing of minefields is thus a major concern for U.S. defense forces. To do this, one has to find the mines and due to the variety of types of mines, both in shape and composition of materials, this problem is a difficult one. Come and learn what people are doing in this important area of applied spatial statistics. The Chemometrics session was organized by Kwan Lee, titled Chemometrics for Pharmaceutical Research and Development, is concerned with the application of statistical methods to chemistry to enable the extraction of chemically relevant information produced in chemical experiments. Those wishing to know more about the applications of techniques such as PCA, Partial Least Squares, and Projection to latent structure (PLS) should attend this session. Bayesian Methods in Physical Measurement was organized by Mark Vangel. and is concerned with Bayesian contributions in three area of metrology science: uncertainty analysis, calibration, and combining of information. It is my prior belief that this session will hold interest for Bayesian's as well as frequentists. The final invited session is Topics in Multiple Response Experimentation organized by Geoff Vining. Most engineering experiments involve several characteristics of interest. Classical approaches treat each response independently, which is simplistic, but easy. This session deals with ways of handling such data, including experimental design, optimization, and a commercial application using multiple responses. The invited sessions are all very interesting and I would like to personally thank the organizers for putting them together.
There are 4 Topic Contributed sessions: Statistical Challenges in the Geosciences, organized by Doug Nychka, Current Issues in Statistical Analysis and Modeling of Automotive Emissions, Parts I and II, both organized by Tim Coburn, and Statistics Application in Army R&D, organized by Barry Bodt. A nice mix of topics on problems and areas of great concern.
The Regular Contributed sessions cover a wide breadth of statistical areas from general applications, design of experiments, response surface and regression methods, reliability, estimation-testing-classification, time series analysis, and applications to national and international security.
I am very happy with the breadth and depth of this years SPES program and I am indebted to many people for helping me put this together. There is one unfortunate situation that you can observe from the time table below – there are some overlaps of SPES sessions. This is something that is hard to avoid and I apologize in advance to those who would like to attend two sessions that are overlapped. Finally, I would like to thank John Peterson, my colleaque at SmithKline Beecham, for stepping in for me while I was unable to attend planning meetings and for his help with many other administrative tasks, Janet Buckingham, SPES Chair, for her help and also to Alicia Carriquiry, JSM Chair, for keeping me on track – she has done a tremendous job.
Monday |
Tuesday |
Wednesday |
Thursday |
|
August 8, 1999 |
August 9, 1999 |
August 10, 1999 |
August 11, 1999 |
August 12, 1999 |
| 2:00-3:50 Applications of Statistics to the Physical and Engineering Sciences (Contributed) |
8:30-10:20 Advances and Applications in Design of Experiments (Contributed) |
8:30-10:20 Current Issues in Statistical Analysis and Modeling of Automotive Emissions, Part 1 (Topic) |
8:30-10:20 Chemometrics for Pharmaceutical Research and Development (Invited) |
8:30-10:20 Bayesian Methods in Physical Measurement (Invited) |
| 4:00-5:50 Statistical Challenges in the Geosciences (Topic) |
10:30-12:20 Statistical Modeling of Growth and Degradation Processes (Invited) |
10:30-12:20 Current Issues in Statistical Analysis and Modeling of Automotive Emissions, Part 2 (Topic) |
10:30-12:20 Applications of Statistics to National and International Security Issues (Contributed) |
8:30-10:20 Estimation, Testing, and Classification (Contributed) |
| 10:30-12:20 Recent Advances in Response Surface and Regression Methods (Contributed) |
10:30-12:20 Statistics Application in Army R&D (Topic) |
10:30-12:20 Time Series and Related Topics (Contributed) |
||
| 2:00-3:50 Spatial Methodology for Minefield Detection (Invited) |
2:00-3:50 Recent Advances in Reliability Inference and Applications (Contributed) |
10:30-12:20 Topics in Multiple Response Experimentation (Invited) |
Darryl J. Downing, 1999 JSM Program Chair
Robert N. Rodriguez , 1998 JSM Program Chair Elect
The SPES Roundtable Discussion topics for 1999 will offer an outstanding menu of food for thought. A quick tour of the kitchen reveals seven master chefs, each preparing his own haute cuisine:
For patrons who have never tried wavelets, Bhavik Bakshi (Ohio State University) will lead a discussion of what they offer the industrial statistician as a solution for process control and related problems. Measurement data often contain features at multiple scales, but most analysis methods are best for features at a single scale. Wavelet-based methods can exploit the multiscale nature of such data. You are also encouraged to join this discussion if you have experience applying wavelets in SPC, data reconciliation, fault diagnosis, and regression.
Richard De Veaux (Williams College) will offer his guests a discussion of the uses of neural networks for industrial process modeling and improvement. Guests joining Kwan Lee (SmithKline Beecham Pharmaceuticals) will learn from each other’s experience how chemometrics, which has made significant contributions in analytical chemistry, can be leveraged to impact other areas of industry.
For nouvelle cuisine, participants can join David Olwell (Naval Postgraduate School) to discuss successful and potential applications of self-starting control charts which support management decisions in a variety of non-manufacturing arenas ranging from sexual harassment to network security. Those who prefer meat and potatoes can dine with José Ramírez (W. L. Gore & Associates) to decide whether split-plot designs are the exception or the rule in a wide variety of industrial settings.
Weight watchers who require lean sample sizes should join Keith Muller (University of North Carolina) to discuss anxiety-reducing strategies, such as appropriate tables and plots, for defensibly incorporating and explaining highly speculative values in power calculations. Questions for discussion include the following: How can one elicit plausible ranges for the effect of interest? What are the ethical responsibilities of the statistician in terms of time and effort for a difficult task considered trivial by the client? And how can one create approximate or exact confidence intervals about sample size and power values?
Finally, for diners concerned with multiplicity issues in industrial screening experiments, Stan Young (Glaxo-Wellcome) will offer a discussion of the pros and cons of controlling familywise Type I error, along with practical methods and alternatives. Join us at Café SPES!
Nancy McMillan , SPES Education Chair
ABSTRACT
The need for solving multivariate optimization problems is pervasive in statistics and other fields. Unlike optimization problems seen in introductory calculus, most real-world optimization problems do not yield a closed-form solution. Hence, iterative (recursive) algorithms play a prominent role in finding the solution to most practical optimization problems. There has recently been much interest in iterative optimization algorithms that rely only on measurements of the objective function to be optimized, not on direct measurements or calculation of the gradient of the objective function. Such algorithms eliminate the need for detailed modeling information describing the relationship between the parameters to be optimized and the objective function.
In the above spirit, this half-day course will discuss the "simultaneous perturbation stochastic approximation (SPSA)" algorithm for optimization of multivariate systems. SPSA has recently attracted considerable attention in areas such as statistical parameter estimation, pattern recognition, nonlinear regression, neural network training, adaptive feedback control, and experimental design. The essential features of SPSA are its efficiency for multivariate problems and its relative ease of implementation for practitioners; these features result largely from the underlying simultaneous perturbation gradient approximation that only requires two objective function measurements independent of the number of parameters being optimized (this is in contrast to standard finite-difference techniques, which require a number of function evaluations equal to twice the number parameters being optimized). The audience will be introduced to the fundamental issues in stochastic optimization with special emphasis on cases where the classical gradient-based methods (Robbins-Monro SA, steepest descent, Fisher scoring, neural network backpropagation, etc.) do not apply. These cases include many important practical problems. For purposes of contrast, brief discussion will also be included on other modern approaches such as simulated annealing and genetic algorithms.
The aim of this course is to introduce statistical practitioners and researchers to a relatively new optimization approach (SPSA) that has attracted considerable international attention in a variety of problems. SPSA has not yet received wide publicity among statisticians although it has direct applications in many problems with a significant statistical component (the genesis of the approach has been in engineering, although the applications span virtually all areas of the physical and social sciences as well). It is expected that the class participants will come into the class with a working knowledge of probability and statistics at the undergraduate level and knowledge of multivariate calculus (at a level of understanding the basic principles in deriving and using gradient vectors).
INSTRUCTOR
Dr. James C. Spall joined The Johns Hopkins University, Applied Physics Laboratory (JHU/APL) in 1983 and was appointed to the Principal Professional Staff in 1991. He also teaches in the Johns Hopkins University School of Engineering, Applied Mathematics Program and has taught in-house professional development courses for JHU/APL staff members on neural networks and optimization. He has given workshops or short courses at conferences such as the Joint Statistical Meetings, the Summer Computer Simulation Conference, the IEEE International Symposium on Intelligent Control, the U.S. Army Artificial Intelligence Symposium. He has worked in diverse areas within the fields of statistics and control systems, and he has worked in applications areas such as defense systems and transportation systems. Dr. Spall has published over 70 papers in the areas of statistics and control systems, including articles on subjects such as adaptive control, small-sample data analysis, performance evaluation and testing, time series, optimization, parameter estimation, and neural networks
Cory Atwood, Associate Newsletter Editor
The poster session at the Joint Statistical Meetings in Baltimore this August ought to be a great one! This year all JSM posters will be grouped by topic (e.g. experimental design, quality control, spatial statistics, linear models, teaching statistics, web applications, etc.) rather than by sponsoring society or section. Please see the JSM99 Poster Page for more details. Of course, SPES members will have their share of interesting posters to contribute to the JSM poster session. There will be several nice methods-applications to the physical and engineering sciences, as well as a poster on the "Engineering Statistics Internet Handbook" by Paul Tobias. The interactive nature of posters should make for a relaxing and enjoyable learning session with your fellow statisticians.