Please mark this date and time on your calendars.

This is joint work with Bin Yu.

If you are not able to make this meeting, but have comments to share, please contact Ellis Johnson.

For group-censored data and continuos data, it is verified that all the implicit functions are well defined, and the asymptotic distributions of the SMLE for model parameters and lifetime distributions are obtained.

A real life example with HIV data is presented to illustrate the application of SMLE method.

The meeting is open to all interested parties, and is free of charge. Please pass this information along to any associates whom you think might be interested.

The meeting will be in the Executive Classroom (Room 228) of the ISyE Main Entrance Bldg. (755 Ferst Drive, formerly the Dupree School of Management Building).

Here's a link to a map of the campus: http://www.isye.gatech.edu/visitors/maps/

There should be plenty of free parking in the area after 5:00 PM.

Kevin Geraghty (kevin@revenueresearch.com) is organizing an informal dinner after the meeting. Please contact him if you are interested in participating so he can make an appropriate reservation at a nearby restaurant.

We look forward to seeing you there. Please contact me if you have questions or suggestions, or if you wish to be added to the mailing list for future meetings.

The conference program covers a multitude of topics relevant to assistive technologies and universal accessibility and is structured around technical papers, poster sessions, demonstrations, panels, the doctoral consortium, and a host of social events.

ASSETS 2004 will be held at the Georgia Tech Global Learning Center at Technology Square, a state-of-the-art meeting and accommodation facility that combines all the amenities and modern technologies of tomorrow with a prime midtown Atlanta location. The hotel and conference center is fully accessible to all participants.

A reception will precede the meeting at 5:30. The meeting is open to all interested
parties, and is free of charge. Refreshments will be served, and there will
also be time to network with fellow OR professionals in the Atlanta community.
The meeting will be in the Executive Classroom (Room 228) of the ISyE Main
Entrance Bldg. (755 Ferst Drive, formerly the Dupree School of Management Building
- same location as our previous meetings this year.) There is plenty of free
parking next to the building after 5:00 PM.

Map

http://www.isye.gatech.edu/visitors/maps/

Informs Chapter Website

http://www.isye.gatech.edu/~informs/atl_chapter/index.html

In this talk we overview linear approximations
to Gamma-minimax actions and demonstrate
that in many decision theoretic problems
linear approximations are not substantially affecting
the risk of the decision maker.

The talk is a mixture of educational and research overviews,
and does not assume prior exposure to minimaxity and
Gamma-minimaxity.

Sequential resource allocation systems (RAS) constitute a pertinent modeling abstraction for the operational dynamics of a broad range of contemporary technological applications, including production systems, material handling and railway / monorail systems, e-commerce and other service-related processes, and even computational environments like those emerging in internet-based computing. In all these environments, a set of concurrently executing processes contest for the sequential exclusive acquisition of a finite set of re-usable resources that are necessary to support the execution of their various processing stages. The resulting resource allocation process must be controlled for (i) operational efficiency, a requirement giving rise to scheduling problems in the context of these environments, but also, for (ii) logical correctness and inherent consistency, a requirement addressed by an emerging logical control theory for these systems. The effective logical control of the aforementioned applications becomes an even more important problem as these environments migrate to extensively automated operational modes.

This talk will survey the state-of-the-art in RAS logical control. More specifically, the first part of the talk will provide a general description of the problem and a formal characterization of it in the Discrete Event Systems (Ramadge & Wonham

(Joint work with Liwei Bai (Georgia Tech), Liming Liu and Weixin Shang (Hong Kong Univ. of Sci. and Tech.)). Based on the paper, L. Bai, B. Fralix, L. Liu, and W. Shang. Inter-Departure Times in Base-Stock Inventory-Queues. Queueing Systems 47 (2004) 345-361.

Motivated by these remarks, we consider the problem of performance modeling, analysis and control of capacitated, flexibly automated re-entrant lines. Specifically, we develop an analytical framework for the modeling, analysis and control of capacitated re-entrant lines, which is based on Generalized Stochastic Petri nets (GSPN) framework. Furthermore, the underlying scheduling problem is transformed to a Markov Decision Process (MDP) problem and finally, we suggest a systematic, efficient and scalable approximating scheme, which is based on the Neuro-Dynamic Programming (NDP) theory, for the optimal scheduling policy characterized in the GSPN / MDP framework. The quality of the obtained approximations is experimentally assessed by

A simple strategy for solving chance constrained problems is to generate N samples according to the distribution P and then impose the constraints f(x,Y_i) < 0, for all i = 1, ..., N (*)
Since Y_i are random samples, there is no hope that decision
variables x that are feasible for (*) are all feasible for the chance
constraint with probability 1. Therefore, one has to allow a probability of error delta. Question: How large should N be as a function of eps and
delta ?

Recently, Calafiore and Campi showed that when f(x,Y) is a convex function of x for every fixed Y we only need N =O(n/delta log(1/eps)). Nemirovski and Shapiro have shown that if the function f(x,Y) is bi-affine and the distribution P has a "concentration-of-measure" property the number of samples required drops to N = O(n log(1/(eps*delta))).

In many applications of chance constrained problems the distribution P is not completely known, i.e. the distribution is ambiguous. The natural
constraint to impose in this setting is the ambiguous chance constraint:
max_{P in M} {P(f(x,Y) > 0)} < eps, where M is an uncertainty set of distributions. In this talk we discuss how to extend many results known for chance constrained problems to this more general setting. Robust deterministic optimization naturally arises
in these extensions by way of the Strassen-Dudley representation theorem.

(Joint work with Emre Erdogan)

Brief Bio: Garud Iyengar received his PhD in Electrical Engineering from Stanford University in 1998. Since then he has been with the Industrial Engineering and Operations Research (IEOR) department at Columbia
University where he is currently an Associate Professor.

We assume that there are finite environment states and the changing environment is modeled as a general stochastic process which takes discrete values. At each state of the environment, the network operates as a queueing network where each server may serve multiple classes of customers. In this study, we establish a framework to search for asymptotically optimal scheduling policies for such queueing networks. We first show that open queueing networks in a slowly changing environment can be approximated by their fluid analog, stochastic fluid models, when the network speed increases. Given a solution of the stochastic fluid model, we provide a method to derive suitable scheduling policies for the original queueing networks. We further show that the queueing networks operating under the derived policies converge to the corresponding stochastic fluid model . This result implies that the derived scheduling policies are asymptotically optimal if the given stochastic fluid model solution is optimal. We also study a stochastic fluid model to investigate the optimal resource allocation policies of Web servers serving heterogeneous classes where the Web servers may be overloaded and operate under Quality of Service contracts.

This event is FREE FOR ISyE Ph.D. STUDENTS, FACULTY, AND STAFF. We encourage people to bring their families and close friends, as one of our goals is to have a "family-oriented" event. However, to cover expenses, there is a nominal $3 fee for all adult guests. There is no cost for children.

The menu will include, but not be limited to: hamburgers, veggie burgers, beef hot dogs, grilled chicken, vegetarian baked beans, and potato salad.
Water and soda will be provided as well. ALCOHOL IS PROHIBITED.

We are also planning to have a volleyball setup and, if enough interested faculty attend, a faculty vs student volleyball match.

If you would like to attend this event, please respond to this email no later than Thursday, September 30, at 5 P.M. with the following
information:

Name:
Year in Program (2nd, 3rd, 4th, etc.):
Area of Concentration or Specialty (EDA, Optimization, Statistics, etc):
Number of Adult Guests:
Number of Children Guests:
Names of Adult Guests:
Names of Children Guests:

We also could use some volunteers to help with the following:

1. Setup of tables and chairs at approximately 3:00 P.M.
2. Volleyball setup at approximately 3:30 P.M.
3. Clean-up from approximately 7:30-9:00 P.M.

If you are available and are willing to volunteer for one or more of these, please indicate that in your reply as well. Your help would be greatly appreciated.

Thank you, and we look forward to enjoying your fellowship on Friday!!! :)

(1) The objective function can only be evaluated with some error,
for example by and at high computational cost.

b) The error can be decreased with more computational effort.
c) The higher order derivatives of the objective function are unavailable.
Though our work is motivated from problems arising in stochastic simulation
optimization (Eg.Revenue Management), such optimization problems also commonly arise in
engineering design (Eg. helicopter rotor blade design).
Due to the high cost of evaluating the objective function, our aim is
to develop convergent algorithms that can solve such problems
while requiring the fewest possible number of evaluations of the
objective function.

When the objective function and its gradient can be evaluated easily
and exactly, a typical trust region algorithm works as follows.
At each iteration, we first construct a polynomial model function, typically a
truncated Taylor series expansion of the objective at the
current iterate, that approximates the objective function
in a certain neighborhood of the current iterate called
the trust region. We then optimize this model function within the trust
region. Depending on whether the resulting
optimal solution has a lower objective function value or not, we
either set this as the next iterate or conversely, alter the trust
region size and/or the model function appropriately and try again.

Since in our case, the objective function and its higher order
derivatives cannot be evaluated exactly, we cannot use a Taylor series
based model function. Accordingly, we propose
alternative linear or quadratic polynomial
model functions that are constructed by linear regression using only
sample average approximations of the objective, evaluated at points
in the neighborhood of the current iterate. We describe the various
changes that have to be made to the traditional trust region framework
in order to successfully construct and control the accuracy of such regression
based model functions and present the convergence theory for the
resulting modified trust region algorithm. Finally, we provide
computational results for such an algorithm run on selected problems
from the CUTE test set.

Comparisons with non-wavelet and wavelet-non-Bayesian
methods are discussed.

This is a joint work with Marianna Pensky from UCF.

Authors: Sigrun Andradottir, Hayriye Ayhan, and Douglas G. Down

The modeling and the methodological part of this talk are motivated by the following three scientific questions:

(a) How does one "statistically quantify" the unknown amount of gene in a "sample" using PCR amplification methods?
(b) What are the factors affecting the amplification rate of a PCR experiment?
(c) How does one efficiently design experiments to study the mutations induced by PCR experiments?

Statistical quantification can lead to an accurate estimate of the HIV-1 viral load in HIV-1 infected patients. This not only helps with disease diagnosis but also in the disease prognosis. Answers to questions (b) and (c) will facilitate a better understanding of the dynamics of a PCR process.

Once again, our festival will feature a Mexican Fajita & Taco Bar.

Assume the networks satisfy a so-called resource pooling condition. We prove a maximum pressure policy asymptotically minimizes the workload processes in heavy traffic. A key to the proof is to show the network processes exhibit state space collapse.
(joint with Jim Dai)

For those of you who came to our ISyE tailgate party this past October, this event will be very similar. We will have a picnic-style setup again will GRILLED MEATS, CHIPS, AND DRINKS. However, this time, we will have FRISBEES AND A VOLLEYBALL SET-UP in addition to other games and activities for your enjoyment.

With regard to food, there will be NON-MEAT OPTIONS available for vegetarians. In addition, there will be NO PORK among the meat products.

PLEASE RSVP NO LATER THAN THURSDAY, APRIL 15 USING THE FORM BELOW if you plan to attend. Like last time, THERE IS A $2 CHARGE FOR ALL ADULT GUESTS, BUT NO CHARGE FOR CHILDREN. So, for those of you with families and/or significant others, we encourage you to bring them, if possible.

Name:
Years in ISyE:
Number of Guests:
Names of Guests (Kids in Parentheses):
If Date is Moved to 4/30 Due To Rain, Can You Attend?

(The rest of the form is for Ph.D. students only.)

Concentration:
Advisor:

In case of rain, the alternate date for the event has been set for Friday, April 30.

This event is being sponsored by the Georgia Tech Chapter of INFORMS, with approval from its faculty advisor Dr. Dave Goldsman.

Thank you.

W. Brad Jones

_________________________________
_________________________________
W. Brad Jones, Ph.D. Student
Georgia Institute of Technology
School of Industrial and Systems Engineering
310 ISyE Main Building
Atlanta, GA 30332

This talk will survey the state-of-the-art in RAS logical control. More specifically, the first part of the talk will provide a general description of the problem and a formal characterization of it in the Discrete Event Systems (Ramadge & Wonham

Modeling and simulation is widely regarded as one of the breakthrough technologies that will accelerate progress in addressing the grand challenges facing manufacturing in the next decade. In this seminar, new research challenges arising from the increased criticality of inter-factory dependencies in today

For a large electronic equipment leasing company, optimal management of assets supported by good logistics decisions is crucial and may provide a significant competitive advantage. The leasing company tries to maximize operating profits through key decisions associated with the length of leases, efficient utilization of logistics facilities for material flow to and from customer sites, and equipment reuse, refurbishment and disposal actions.

In this research, a mixed integer linear programming (MILP) model is developed to facilitate better decisions from the perspective of an electronic equipment leasing company. A case study with representative industry data validates the approach and demonstrates the utility of the model in answering key research questions. Next, important problem uncertainties are identified and prioritized. The effects of these key uncertainties on optimal lease length and product flow decisions are examined in detail via an extended case study. It is also shown how the leasing company can make near-robust leasing decisions in the face of these uncertainties.

The computational research results also have implications for policy formulation on electronic waste. The important insights include an understanding of the potential impacts and expected effectiveness of alternative environmental legislation in different geographic areas, and the imposition of negative externalities on other policy realms as a result of this non-uniform approach. Therefore, this research contributes new models and understanding to the intersection of the fields of reverse logistics and equipment replacement, and provides valuable insights to both business asset managers and environmental policy makers.

We apply fully sequential procedures due to Kim and Nelson (2004) to
Bernoulli data for terminating simulations, employing common random numbers.
We find that significant savings in total observations can be realized for
two to five systems when we wish to detect small differences between
competing systems. We also study the multinomial selection problem. We offer
a Monte Carlo simulation of the Bechhofer and Kulkarni (1984) MBK
multinomial procedure and provide extended tables of results. In addition,
we introduce a multi-factor extension of the MBK procedure. This procedure
allows for multiple independent factors of interest to be tested
simultaneously from one data source (e.g., one person will answer multiple
independent surveys) with significant savings in total observations compared
to the factors being tested in independent experiments (each survey is run
with separate focus groups and results are combined after the experiment).
Another multi-factor multinomial procedure is also introduced, which is an
extension to the MBG procedure due to Bechhofer and Goldsman (1985, 1986).
This procedure performs better that any other procedure to date for the
multi-factor multinomial selection problem and should always be used
whenever table values for the truncation point are available.

References Bala MV, Zarkin GA. Pharmacogenomics and the evolution of healthcare : is it time for cost-effectiveness analysis at the individual level? Pharmacoeconomics. 2004;22(8):495-8. Flowers CR, Veenstra D. The role of cost-effectiveness analysis in the era of pharmacogenomics. Pharmacoeconomics. 2004;22(8):481-93. Danzon P, Towse A. The economics of gene therapy and of pharmacogenetics. Value Health. 2002 Jan-Feb;5(1):5-13.

Dr. Christopher Flowers, MD, is an Assistant Professor in Hematology and Oncology and the Clinical Director for Oncology Informatics Program at the Winship Cancer Institute, Emory University School of Medicine

Besides low procurement costs, important considerations in
supplier selection are responsiveness and the reliability of the
suppliers in meeting demand. Hence, manufacturers face the
pressure for quoting short and reliable lead-times. We cover
several aspects of the manufacturer's problem, such as quoting
reliable due-dates based on the workload status in the system,
maximizing profit considering the lateness cost incurred due to
late deliveries, and deciding on the level of inventory to
increase responsiveness. We employ a model where demand arrival
and manufacturing processes are stochastic, and obtain insights on
the optimal due-date policy and on the optimal inventory level.

In this talk, it will first be shown that even in simple networks, commonly used operational policies like first-in-first-out sequencing may perform badly, failing to achieve even "throughput optimality." We then present two families policies, called discrete-proportional-processor-sharing and maximum pressure policies, that are always throughput optimal, regardless of the processing network's topology or parameter values. These policies have other attractive features as well, including distributed implementation that uses only local or semi-local congestion information. A simulation study has been undertaken that evaluates these policies in a wafer fabrication setting, using SEMATECH data sets. The results of that study will be discussed, along with other attractive theoretical properties of the two policy families.

Founded in 1990, Citadel Investment Group is a world leader in alternative investments, with a team of over 800 people in five offices worldwide. Our research philosophy is to apply a systematic process driven approach
to investing in order to advance the reliability and repeatability of high risk adjusted returns.

Speaker Bio:

Dr. Byrns is the Director of Energy Research at Citadel Investment Group in Chicago, IL. He received his PhD in Engineering, (GA Tech 1991), as
well as a MS in Economics (GA Tech 1991), a MSE in Aerospace Engineering (GA Tech 1988) and a BSE in Mechanical and Aerospace Engineering (Princeton 1985). He has been actively involved in commodity research for almost 8 years, holding various staff and management positions at Williams
Energy, Merchant Energy Group of the Americas, and Koch Industries. Prior to entering the commodity field, he worked as a consultant in Washington
D.C.

We propose a novel simulation approach for pricing American-style
contingent claims. We develop an adaptive policy search algorithm
for obtaining the optimal policy in exercising an American-style
option. The option price is first obtained by estimating the
optimal option exercising policy and then evaluating the option
with the estimated policy through simulation. Both high-biased and
low-biased estimators of the option price are obtained. We show
that the proposed algorithm leads to convergence to the true
optimal policy with probability one. This policy search algorithm
requires little knowledge about the structure of the optimal
policy and can be naturally implemented using parallel computing
methods. As illustrative examples, computational results on
pricing regular American options and American-Asian options are
reported and they indicate that our algorithm is faster than
certain alternative American option pricing algorithms reported in
the literature.

Secondly, we investigate arbitrage opportunities arising from
continuous monitoring of the price difference of highly correlated
assets. By differentiating between two assets, we can separate
common macroeconomic factors that influence the asset price
movements from an idiosyncratic condition that can be monitored
very closely by itself. Since price movements are in line with
macroeconomic conditions such as interest rates and economic
cycles, we can easily see out of the normal behaviors on the price
changes. We apply a statistical process control approach for
monitoring time series with the serially correlated data. We use
various variance estimators to set up and establish trading
strategy thresholds.

The second half of the thesis presents multiple testing in
large-scale contingency tables and its application to pattern
recognition of protein structures. One of the most common test
procedures using two-way contingency tables is the test of
independence between two categorizations. Current significant
tests such as $chi^2$ or likelihood ratio tests provide overall
independency but bring limited information about the nature of the
association in the contingency tables. The main purpose of this
study is to develop a follow-up method to $chi^2$ or likelihood
ratio tests that identifies the significantly associated
individual cells in the contingency table. We propose a framework
of multiple testing procedures for testing independence of the
cell categories in contingency tables. In the simulation study, we
compare the power, type I error, and false discovery rate of five
different testing procedures in the contingency table. We observe
that no single procedure is superior for every scenario examined.
In addition, we record the relationships among the proportion of
true null hypotheses, power, type I error, and false discovery
rate. Finally, we employ the proposed method to identify the
patterns of pair-wise associations between amino acids involved in
$beta$-sheet bridges of proteins. We identify a number of amino
acid pairs that exhibit either strong or weak association. These
patterns provide useful information for algorithms that predict
secondary and tertiary structures of proteins.

The multivariate quadratic loss function (MQL) is a popular multivariate
technique in static RD and has occasionally been applied to multivariate
SPC. In both areas we integrate the contents of the MQL into specially
constructed principal components called loss-scaled principal components
(LSPC). We examine how well a subset of these LSPC approximate the
expected value of MQL and apply them to a RD problem featuring six
responses and eight predictor variables. We also show when
LSPCs can quickly detect and accurately diagnose shifts in location and
dispersion in multivariate SPC.