Abstract
Relaxations for nonconvex quadratic optimization commonly use
the "Reformulation-Linearization Technique" (RLT) to replace bilinear and quadratic terms with new variables, adding constraints that are implied by upper and lower bounds on the original variables. For two original variables, RLT constraints and semidefiniteness give an exact convex reformulation for nonconvex box-constrained quadratic programming (QPB). In any dimension,projecting out the quadratic variables in an instance of QPB produces the Boolean Quadric Polytope (BQP) associated with quadratic optimization over Boolean variables. Polyhedral combinatorics of the BQP can therefore be used to strengthen convex relaxations for QPB and more general nonconvex quadratic optimization problems. We describe recent theoretical and computational results.

Bio
Kurt M. Anstreicher is the Tippie Research Professor of Management Sciences at the University of Iowa College of Business. Professor Anstreicher received his P.h.D from the Department of Operations Research at Stanford University in 1983, under the direction of George Dantzig. He was an assistant and associate professor of Operations Research at Yale University from 1982-1991, and went to the University
of Iowa as a professor of Management Sciences in 1991. He was appointed Daly Professor in 1998, and Tippie Research Professor in 2002.

Professor Anstreicher is an internationally recognized researcher in continuous optimization. He has authored approximately 50 journal articles, mainly in the area of interior-point algorithms for linear and nonlinear programming. In 2002 he was awarded (joint with N. Brixius,
J.P. Goux, and J. Linderoth) the SIAM Activity Group on Optimization (SIAG/OPT) Prize for the best paper in optimization in 2000-2002. He was an associate editor of the SIAM Journal on Optimization from 1991-1997, and was named co-editor of Mathematical Programming Series A in 1999. He was a member of the ORSA (now INFORMS) Lanchester Prize committee in 1991, and the Mathematical Programming Society (MPS) Tucker Prize committee in 1997 and 2000. He served on the elected Council of the MPS from 1997 to 2000.

Professor Anstreicher received the MBA Faculty Member of the Year award at the University of Iowa School of Management in 1993 and 1996, and the MBA Core Faculty Member of the Year award in 2000. In 2002 he received the G.R.E.A.T. Teaching Award from the graduating Executive MBA class. He was cited as an outstanding faculty member at the University of Iowa School of Management in the fourth (1995), fifth (1997) and seventh (2001) editions of The Business Week Guide to the Best Business Schools.

Abstract
It is well-known that when job size variability is high, one needs to prevent short jobs from getting stuck behind long jobs. In a server farm setting, one way to achieve this goal is to allocate short jobs their own server (or set of servers). This is the theory behind the popular Size Interval Task Assignment policy (SITA) for server farms, which assigns each server a unique size range, so that short jobs are given isolation from long ones. The SITA policy is prevalent throughout compute server farms and manufacturing systems, whenever job size variability is high. The higher the job size variability, the more important it is to provide short jobs some isolation from long ones, via a SITA policy, or some variation thereof.

This talk questions the above common wisdom. To understand what's going on, we study the performance of task assignment policies, in the limit, as the variability of job sizes (service demands) approaches infinity. Results in this limiting regime reveal that the SITA policy can be far inferior to much simpler greedy policies, like Least-Work-Left (LWL), for many common job size distributions, including a range of Pareto distributions. Regimes are also defined where SITA's performance is good, and here simple closed-form bounds are proved on its performance. Towards the end of the talk we will also consider the performance of SITA variants/hybrids.

Parts of this work appeared in ACM SIGMETRICS 2009.

Joint work with: Alan Scheller-Wolf and Andrew Young

Bio
Mor Harchol-Balter is Associate Department Head of the Computer Science Department at Carnegie Mellon University. She received her doctorate from the Computer Science department at the University of California at Berkeley under the direction of Manuel Blum. She is a recipient of the McCandless Chair, the NSF CAREER award, the NSF Postdoctoral Fellowship in the Mathematical Sciences, multiple best paper awards, and several teaching awards, including the Herbert A. Simon Award for Teaching Excellence. She is heavily involved in the ACM SIGMETRICS research community, and recently served as Technical Program Chair for SIGMETRICS. Mor's work focuses on designing new resource allocation policies (load balancing policies, power management policies, and scheduling policies) for server farms and distributed systems in general. Her work spans both queueing analysis and systems implementation, and emphasizes integrating measured workload distributions into the problem solution.

Learn more, visit http://www2.isye.gatech.edu/humlog09/.

For more information, visit:
http://www.scl.gatech.edu/professional-education/scef/

The H. Milton Stewart School of Industrial and Systems Engineering

Speaker: Shiyu Zhou, Associate Professor, University of Wisconsin - Madison

Abstract:

The performance of a system is generally influenced by many factors, among which some are uncontrollable factors, also called risk factors. The impact of these factors must be taken into account when we examine a system for performance improvement. On the other hand, with the rapid development of information technology, abundant data on both performance and influential factors are available in many systems. The data rich environment enables us to develop new methodologies to account for risk factors in studying system performances. This talk presents the recently developed methodologies for risk-adjusted performance measuring and monitoring in manufacturing systems and healthcare delivery systems based on Graphical Models and Bayesian inference.

Bio:

Dr. Shiyu Zhou is an associate professor in the Department of Industrial and Systems Engineering at the University of Wisconsin-Madison. He got his B.S. and M.S. in Mechanical Engineering at University of Science and Technology of China in 1993 and 1996 respectively, and got his Master in Industrial Engineering and Ph.D. in Mechanical Engineering at the University of Michigan in 2000. Dr. Zhou's research focuses on the fault management of complex engineering systems by integrating statistics, system and control theory, and engineering knowledge. His research is sponsored by National Science Foundation, Department of Energy, NIST-ATP, and industries. He is a recipient of the CAREER Award from the National Science Foundation and the best application paper award from IIE Transactions. Dr. Zhou is a member of IIE, INFORMS, ASME, and SME.

SPEAKER: Professor Victor de la Pena

ABSTRACT:

Self-normalized processes arise naturally in statistical applications. Being unit free, they are not affected by scale changes. Moreover self-normalization often eliminates or weakens moment assumptions. In this talk, I will present several exponential and moment inequalities for self-normalized processes as well as their connection to maximum likelihood estimators.

SPEAKER: Utku Yildirim

ABSTRACT:

Although accurate demand forecasting is one of the most critical component of any revenue management application, it has not received enough attention. Utku Yildirim will give a presentation that focuses on the commonly used demand forecasting techniques and compares their performances based on real-world hotel booking data.

Bio

Utku Yildirim focuses on development of pricing and revenue management solutions at Prorize LLC. Prior to joining Prorize LLC, he was a member of the Revenue and Profit Management department at Disney Parks and Resorts, which is responsible for the development and deployment of revenue management solutions.

Utku received his Ph.D. degree in Operations Research and Industrial Engineering at the University of Texas at Austin. During his graduate studies, Utku worked on stability and pricing of queuing networks as well as hotel and airline revenue management problems.

SPEAKER: Gillian R. Hayes

ABSTRACT:

Improvements in modern healthcare enabled many formerly fatal conditions to become survivable but chronic. At the same time, new technologies are enabling more care to move into homes and away from doctors' offices and hospitals. In this talk, I will use my research in technologies for chronic care across multiple domain problems to illustrate how ubiquitous and collaborative computing systems and applications may enable new forms of patient care and powerment. This work has included research focused on enabling improved care and record-keeping for extremely low birth-weight children, children with autism, children and adults with cancer, personal and electronic health records across a lifetime, and the multitude of surveillance and privacy concerns these novel technologies engender.

BIO
Gillian R. Hayes is an Assistant Professor in the Donald Bren School of Information and Computer Sciences at the University of California at Irvine where she directs the Social and Technological Action Research Group. Her research interests are in human-computer interaction, specifically in the areas of ubiquitous computing, computer supported cooperative work, and health informatics. At UCI, Dr. Hayes is affiliated with the Center for Biomedical Informatics, the Center for Ethnography, California Institute for Telecommunications and Information Technology, and the Laboratory for Ubiquitous Computing and Interaction. She received her B.S. in mathematics and computer science from Vanderbilt University in 1999 and her Ph.D. in computer science, with an emphasis on human-computer interaction at the Georgia Institute of Technology in 2007. More information can be found at http://www.gillianhayes.com.

SPEAKER: Xinwei Deng

ABSTRACT:

This talk has two parts. The first part is active learning via sequential design with applications to detection of money laundering. Money laundering is a process to conceal the true origin of funds that were originally derived from illegal activities. However, detecting money laundering is not an easy job because of the huge number of transactions that take place each day. The usual approach adopted by financial institutions is to extract some summary statistics from the transaction history and do a thorough and time-consuming investigation on those suspicious accounts. In this work, we propose an active learning via sequential design method for prioritization to improve the process of money laundering detection. The method uses a combination of stochastic approximation and D-optimal designs to judiciously select the accounts for investigation. The sequential nature of the method helps to decide the optimal prioritization criterion with minimal time and effort. A case study with real banking data is used to demonstrate the performance of the proposed method.

The second part is Gaussian covariance matrix estimation with Markov structures. A fact overlooked in covariance matrix estimation is that the random variables are often observed with certain temporal or spatial structures. Effectively accounting for such structures not only results in more accurate estimation but also leads to models that are more interpretable. In this work, we proposed shrinkage estimators of the covariance matrix specifically to address this issue. The proposed methods exploit sparsity in the inverse covariance matrix in a systematic fashion so that the estimate conforms to models of Markov structure and is amenable for subsequent stochastic modeling. The present approach complements the existing work in this direction that deals exclusively with temporal orders and provides a more general and flexible alternative to explore potential Markov properties. We show that the estimation procedure can be formulated as a semi-definite program and efficiently computed. The merits of these methods are illustrated through simulation and the analysis of a real data example.

SPEAKER: Dr. Kwok Tsui

ABSTRACT:

Due to various outbreaks of influenza and continuing bioterrorism threat, research efforts on healthcare and public health surveillance have become very important worldwide. In this talk we will explain and classify the various types of health surveillance problems. We review the latest research in surveillance systems, monitoring methods, and performance measures. We also discuss the research challenges and illustrate them with various problems and examples. In particular, we will compare the performance of MCUSUM and MEWMA charts for spatiotemporal surveillance.

SPEAKER: John Gregory
CPLEX Product Manager ILOG

ABSTRACT:

CPLEX has introduced a variety of features in the new century. Rather than focus on versions and release numbers, this talk will provide a unified view covering topics that include performance features, infeasibility analysis, solution pools, and numerical issues. The aim is to ensure that the full power of CPLEX is available when our most demanding users need it.

SPEAKER: Professor Martin Wainwright

ABSTRACT:

Undirected graphical models or Markov random fields (MRF) provide a framework for capturing statistical dependencies among large collections of random variables, and are used in many application domains (e.g., spatial statistics, statistical image analysis, social network analysis). The problem of graphical model selection is to recover the edge structure of the graph based on a set of samples from the unknown model. In general, it is a challenging problem due to the exponential explosion in the number of possible graphs.

We analyze different methods for graphical model selection under high-dimensional scaling, in which the graph size $p$, maximum degree $d$ and sample size $n$ are all allowed to scale. For discrete graphical models over binary variables, we present a polynomial-time algorithm based on $ell_1$-regularized logistic regression, and show that under mild conditions, it can reliably recover the graph structure with $n = Omega(d3 log p)$ samples. We also analyze the information-theoretic limits of the problem, and prove that no method can succeed with fewer than $n = O(d log p)$ samples.

Based on joint work with John Lafferty, Pradeep Ravikumar, and Prasad Santhanam.

SPEAKER: Dr. Ton Dieker

ABSTRACT:

Multi-dimensional reflected Brownian motion is a Markov process which plays an important role in applications. It is particularly widely used to approximate the behavior of heavily loaded queueing networks. Of special interest is the long-term behavior of the process, i.e., its stationary distribution.

Although the stationary distribution of one-dimensional reflected Brownian motion with drift is exponential, in a multidimensional setting this is only true under a so-called skew symmetry condition on the reflection directions.

Not much is known on the stationary measure (or its density) when the skew symmetry condition fails to hold. For special multidimensional reflected Brownian motions, an intriguingly simple formula arises from connections with the reflection principle. In this formula, the stationary density is represented as a finite sum of exponential terms.

This talk reports an attempt to marry this formula with the literature on two-dimensional reflected Brownian motion in a wedge.

Joint work with J. Moriarty, University of Manchester, UK.

SPEAKER: Dr. Arkadi Nemirovski

ABSTRACT:

Compressed Sensing is a rapidly developing novel area in Signal Processing aimed at recovering sparse high-dimensional signals from their low-dimensional linear images, or, which is the same, recovering sparse solutions to heavily underdetermined systems of linear equations. The standard recovery algorithm in this context is L1 minimization, where we estimate the true signal by choosing among the solutions to the system the one with the minimal L1 norm. Theory says that when the matrix of the system is picked at random, such a procedure, with overwhelming probability, recovers true sparse signals in a surprisingly wide range of the sparsity parameter. At the same time, until very recently no computationally tractable sufficient conditions for a matrix to allow for good L1 recovery of signals of given sparsity were known. In the talk, based on joint research with A. Judistky (Grenoble University) and F. Kilinc Karzan (ISyE), we present and discuss conditions of this type along with their relations with the standard difficult to verify conditions, like Restricted Isometry Property.

SPEAKER: Professor Yanfeng Ouyang

ABSTRACT:

While planning facility locations in a supply chain to serve spatially distributed customers, we consider the case where facilities are subject to probabilistic failure (due to reasons such as natural or man-made disasters). If a facility fails, customers may have to be reassigned to
other facilities and incur excessive transportation cost.

Mathematical models are developed to determine optimal facility locations as well as customer assignment strategies, allowing facilities to fail under location-dependent probabilities. The goal is to minimize the sum of initial facility construction costs and expected customer transportation costs under normal and failure scenarios. This talk will focus on a continuum approximation (CA) model that accurately predicts the total logistics cost for large-scale systems. This model not only serves as a very efficient heuristic method to find near-optimum discrete solutions, but also helps provide useful managerial insights (i.e., solution robustness to input data errors, and advantage of having diversity in design parameters).

We also develop a discrete mixed-integer program (MIP) model and a customized Lagrangian relaxation algorithm. Extensive numerical experiments are conducted for both the discrete model and the continuous model. It is shown that the CA model provides a very good alternative especially for large-scale problem instances, when the MIP model has difficulty solving them in a reasonable amount of time.

Bio

Yanfeng Ouyang received his Ph.D. in civil engineering from the University of California at Berkeley in 2005. Since then, he has been on the faculty of the Department of Civil and Environmental Engineering at the University of Illinois at Urbana-Champaign (UIUC). He currently serves on the editorial advisory board for the journal Transportation Research Part B, and is a member of the Transportation Research Board's Network Modeling Committee (ADB30). His research interests focus primarily on improving stability and efficiency of transportation, logistics, and supply chain systems.

Dr. Ouyang received the "Faculty Early Career Development (CAREER) Award" from the National Science Foundation in April 2008. He has been on UIUC's "List of Teachers Ranked as Excellent by Their Students" four times. While a student at Berkeley, he received the Gordon F. Newell Award in 2005, and the "Outstanding Graduate Student Instructor Award" in 2004.

SPEAKER: Dr. Edward Frazelle

ABSTRACT:

The Georgia Tech Supply Chain & Logistics Institute is hosting a free Warehousing Self-Assessment webinar on Friday, February 6 from 11 a.m. to 12 p.m. The webinar will consist of an online self-assessment presentation, conducted by Edward H. Frazelle, Ph.D.

Dr. Frazelle will lead participants through an assessment of their warehouse operation, comparing their warehouse to world-class standards and revealing opportunities for cost, labor and space savings.

Focusing on world-class warehousing and material handling, a variety of professionals would benefit form this course: executives and managers of supply chain, logistics, and distribution; warehouse and distribution anagers/directors; material handling and material management personnel; industrial engineers; operations and facility managers; key personnel committed to improving order picking; and systems analysts, among others.

To register for this event, visit: www.scl.gatech.edu/webinar

SPEAKER: Dr. Edward Frazelle

ABSTRACT:

Dr. Frazelle will lead participants through an assessment of their warehouse operation, comparing their warehouse to world-class standards and revealing opportunities for cost, labor and space savings.

Focusing on world-class warehousing and material handling, a variety of professionals would benefit form this course: executives and managers of supply chain, logistics, and distribution; warehouse and distribution anagers/directors; material handling and material management personnel; industrial engineers; operations and facility managers; key personnel committed to improving order picking; and systems analysts, among others.

To register for this event, visit: www.scl.gatech.edu/webinar

SPEAKER: Dr. Gideon Weiss

ABSTRACT:

Recently I introduced the concept of infinite virtual queues, which are simply queues which never run out of work. In a standard queueing network the input source can be regarded as an infinite virtual queue. In a muti-class queueing network IVQs provide a new paradigm for networks in balanced heavy traffic: In a MCQN with IVQs we can have full utilization of servers (rho=1) and no congestion. I will describe a network which illustrates this phenomena. I will then show how we can use MCQN with IVQs for optimal control of a MCQN over a finite time horizon.

Joint work with students: Anat Kopzon, Yoni Nazarathy.

SPEAKER: Dr. Maria Vlasiou

ABSTRACT:

We investigate the tail behaviour of the steady state distribution of a stochastic recursion that generalises Lindley's recursion. This recursion arises in queuing systems with dependent interarrival and service times, and includes alternating service systems and carousel storage systems as special cases. We obtain precise tail symptotics in three qualitatively different cases, and compare these with existing results for Lindley's recursion and for alternating service systems.

SPEAKER: Dr. Andrew Nobel

ABSTRACT:

Exploratory analysis of high dimensional data often begins with independent clustering of samples and variables, yielding a partition of the data matrix into disjoint row-column blocks (submatrices). Of particular interest in practice are submatrices whose entries are large on average. In conjunction with clinical and functional annotation, large average submatrices are frequently the starting point for subsequent analyses, such as the identification of genetic pathways and new disease subtypes.

This talk describes a simple algorithm, belonging to the general category of biclustering methods, for identifying large average submatrices in high dimensional data. Like other biclustering methods, the algorithm improves on independent sample variable clustering in several respects: the submatrices it identifies can overlap and they need not cover the entire data matrix (features that better reflect underlying biology), and the inclusion of samples and variables in a submatrix does not depend on their expression values outside the submatrix. The algorithm seeks to maximize a simple measure of statistical significance, which also provides an objective basis for comparing and selecting among submatrices of different sizes and average intensities. I will discuss the applications of the algorithm to a recent gene-expression based cancer study, and will provide a detailed comparison of its performance with several other biclustering method, including its application to semi-supervised classification.

Joint work with Andrey Shabalin, Vic Weigman, and Charles Perou.

SPEAKER: Ashlea Bennett

ABSTRACT:

Effective utilization of nurses is crucial in meeting the demand for home care, which is expected to double by 2030, just as the nursing supply shortage increases to 400,000. We attempt to improve nurse utilization by developing automated solution procedures for the dynamic scheduling of home health nurses.

The routing and scheduling problems encountered in the home health industry are dynamic and periodic variants of routing problems that typically include fixed appointment times. Because this is a component not often addressed in the routing literature, we investigate the impact of fixed appointment times on resulting schedules. A solution approach for the dynamic periodic problem variant is developed that uses insertion criteria specific to the fixed appointment time setting. The approach is compared to a more traditional routing heuristic on a set of test problems that have a variety of patient location distributions and arrival frequencies. Computational results are presented.

SPEAKER: Dr. Gideon Weiss

ABSTRACT:

I will motivate continuous linear programs by describing a problem of finite horizon transient control of a multi-class queueing network. I will give the formulation of a separated continuous linear program with linear objective and right hand side, and its symmetric reverse time dual. I will describe the structure of the optimal solutions, in the case that measurable bounded solutions exist, and illustrate the simplex algorithm for its solution. Finally, I will discuss full duality, when impulse controls at time 0 are allowed.

SPEAKER: Prof. Dr. Ir. Hendrik Van Landeghem
Veronique Lim

SPEAKER: Prof. Jianjun (Jan) Shi

ABSTRACT:

A multistage system refers to a system consisting of multiple units, stations, or operations to finish the final product or service. Multistage system is very common in practice such as various modern manufacturing processes and service systems. In most cases, the final product or service performance of a multistage system is determined by complex interactions among multiple stages - the performance characteristics of one stage are not only influenced by the local variations at that stage, but also by the propagated variations from upstream stages. Multistage systems present significant challenges, yet also opportunities for system informatics and control research. The purpose of this presentation is to provide a brief review of the emerging methodologies for tackling various issues in multistage systems including modeling, analysis, monitoring, diagnosis, control, inspection, and design optimization. The focus will be given to the problem motivation, formulation, existing research results, and discussions on open topics for future interdisciplinary research opportunities.

Bio: Dr. Jianjun Shi s a professor and holds the Carolyn J. Stewart Chair Professorship at H. Milton Stewart School of Industrial and Systems Engineering, Georgia Institute of Technology. Before joined Georgia Tech in 2008, he was the G. Lawton and Louise G. Johnson Professor of Engineering at the University of Michigan. He got his B.S. and M.S. in Electrical Engineering at the Beijing Institute of Technology in 1984 and 1987 respectively, and his Ph.D. in Mechanical Engineering at the University of Michigan in 1992.

Professor Shi's research interests focus on the fusion of advanced statistics, signal processing, control theory, and domain knowledge to develop methodologies for modeling, monitoring, diagnosis, and control for complex systems in a data rich environment. Professor Shi is the founding chairperson of the Quality, Statistics and Reliability (QSR) Subdivision at INFORMS. He is also the founding co-Director of the Quality Science Center at the Chinese Academy of Science in China. He serves as the Focus Issue Editor of IIE Transactions on Quality and Reliability Engineering. He is a Fellow of the Institute of Industrial Engineering (IIE), a Fellow of American Society of Mechanical Engineering (ASME), and a Fellow of Institute of Operations Research and Management Science (INFORMS), and also a member ASQ, SME, and ASA.

Chris M. Schlegel, Risk Analysis Services Manager at Southern Company, will be our guest speaker for the meeting. The presentation topic is "Financial Modeling and the Financial Crisis of 2008". Mr. Schlegel will discuss the financial crisis and the impact on financial risk
modeling, including capital markets and structured notes. A brief bio for Mr. Schlegel is below.

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. Please pass this 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). This is the same location as previous recent meetings.

Here's a link to a map of the campus: http://www.isye.gatech.edu/visitors/maps/. There is plenty of (no longer free) parking next to the ISyE building.
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.
Thanks,
John M. Harris, Ph.D.
President
INFORMS - Atlanta
404.506.1198

Presentation Topic & Speaker Bio
Financial Modeling and the Financial Crisis of 2008

Bio
Chris M. Schlegel received his BA (cum laude) and MA in economics at Simon Fraser University, British Columbia, where he worked as graduate research assistant and also taught Advanced Macroeconomics, Advanced Statistics and International Trade. He received recognition for excellence in teaching for all his classes.

Following his graduate work he was Research Economist at The Fraser Institute, and from 2001 served as expert witness for the cost of capital and market analysis for the staff of the New Hampshire Public Utilities Commission in Concord, NH.

Chris subsequently worked as Senior Analyst and later Manager in the Risk Control group on the trading floor of Southern Company, where he developed power and gas trading risk metrics and portfolio valuation tools. He is presently Manager of Risk Analysis Services, a group at
Southern Company headquarters in Atlanta focused on quantitative risk management.

Chris is President of the U.S. Association for Energy Economics in Atlanta, invited lecturer at the Georgia Institute of Technology, and was recently named Associate at the James C. Bonbright Center for Public Utilities at the University of Georgia. He has been invited as speaker
to numerous professional conferences on quantitative risk modeling and utility economics, and has published in peer reviewed journals including Energy Risk and the European Journal of Political Economy. His professional interests include energy economics and enterprise risk management.

SPEAKER: Professor Runze Li

ABSTRACT:

Local polynomial regression is a useful nonparametric regression tool to explore fine data structures ad has been widely used in practice. In this paper, we propose a new nonparametric regression technique called local composite quantile (CQR) smoothing in order to further improve the local polynomial regression. Sampling properties of the proposed estimation procedure are studied. We derive the asymptotic bias, variance and normality of the proposed estimate. Asymptotic relative efficiency of the proposed estimate with respect to the local polynomial regression is investigated. It is shown that the proposed estimate can be much more efficient than the local polynomial regression estimate for various non-normal errors, while being almost as efficient as the local polynomial regression estimate for normal errors. Simulation is conducted to examine the performance of the proposed estimates. The simulation results are consistent with our theoretic findings. A real data example is used to illustrate the proposed method.

SPEAKER: Sangho Shim

ABSTRACT:

omory gives a subadditive characterization of facets for group problems which contains redundant constraints. We discuss how to remove redundant constraints by substitution and discover minimal subadditive characterization of facets for the group problems on cyclic groups of prime order.

Joint work with Professor Ellis Johnson.

SPEAKER: Professor Yixin Fang

ABSTRACT:

In genetic epidemiology, the first and foremost task is testing for familial aggregation; if no familial aggregation is found, it is unnecessary to conduct further genetic analysis. For multivariate traits, we propose a test statistic, which is actually the leading principal component of heritability. The p-value can be obtained by a permutation procedure or a theorem of Johnstone and Forrester (2004). Furthermore, if the test is significant, it is desired to get rid of those redundant traits to avoid overfitting. To this aim, a cross-validation approach and its approximation are proposed. The talk concludes with some discussion on testing for familial aggregation of functional traits and an application to the cholesterol data from Framingham Heart Study.

SPEAKER: Mike Hewitt

ABSTRACT:

Less-than-truckload carriers face an environment with increased competition and customers who demand faster and more reliable service. To respond to these challenges, carriers today seek competitive advantage by reducing costs through improved service network planning. One new direction is to use predictable daily freight volume variations in the planning process to build plans that vary by weekday. Another important new approach is to integrate planning of loaded and empty moves, which have traditionally been conducted sequentially, to plan better utilization of backhaul lanes. Both extensions result in very difficult optimization problems.

In our research, we develop a new heuristic solution technique for these hard problems, which generates day-differentiated service network plans while simultaneously deciding on loaded and empty trailer moves. The method integrates exact optimization into heuristic search by solving an integer program at each iteration that replans the service network for freight destined for a limited set of terminals while holding fixed freight destined for other terminals. Since the method relies on repeated solution of IPs, we also develop classes of valid inequalities and various techniques for reducing their solution time. Computationsl results indicate the methodology uncovers service network plan changes that have the potential to yield reductions of 5% in linehaul costs.

Joint work with Dr.'s Alan Erera, George Nemhauser and Martin Savelsbergh.

This is a joint SCL/DOS Seminar.

SPEAKER: Jung-Kyung Kim

ABSTRACT:

This talk focuses on the tail characteristics of queueing networks with subexponential service times. Our objective is to analyze the tail asymptotics of various performance measures such as cycle times (system times) and waiting times on a variety of queueing models which may arise in telecommunications. First, we consider a general class of closed feedforward fork and join queueing networks with subexponential service time distributions. Our purpose is to provide the tail asymptotics of transient and stationary cycle times and waiting times. The second system is an open tandem queue with a communication or manufacturing blocking scheme. We are interested in the tail asymptotics of transient and stationary response times and waiting times.

SPEAKER: Professor Kjell Doksum

ABSTRACT:

I will give an overview of semiparametric models for two-sample and regression experiments. General frameworks and specific examples such as the Cox regression model will be considered. I will explore a model where the hazard rates of the treatment and control groups start out equal at the time the treatment is introduced and then diverges continuously as time increases. A semiparametric model with bounded random variables is also considered and it is shown that in this model empirical maximum likelihood estimates converge at a rate of n rather than usual rate root(n).

Various likelihoods in use for semiparametric models will be discussed and compared. This is joint work with Aki Ozeki.

SPEAKER: Dr. Vishv Jeet

ABSTRACT:

This paper models, analyzes and develops solution techniques for a network design and inventory stocking problem. The proposed model captures important features of a real service part logistics system, namely time-based service level requirements, and stochastic demands satisfied by facilities operating with one-for-one replenishment policy. In essence, along with usual decisions of location and allocation, the model considers stock levels and fill rates as decisions, varying across facilities to achieve system-wide target service levels. A variable substitution scheme is used to develop an equivalent convex model for an originally non-convex problem. An outer-approximation scheme is used to linearize the convex model. Exact solution schemes based on the linearized model are proposed and computationally less demanding lower and upper bounding techniques for the problem are devised. Results from extensive computational experiments on a variety of problem instances based on a real-life industrial data show the effectiveness of the overall approach.

Reference: Jeet, V., Kutanoglu, E., and Partani, A. (2009) Logistics Network Design with Inventory Stocking for Low-demand Parts: Modeling and Optimization. IIE Transactions, 41, 1

TITLE: The High Performance Supply Chain in a Volatile World

SPEAKER: Chloe Barzey

BIO:

Chloe is a Partner in Accenture

SPEAKER: Dr. S. Jack Hu

ABSTRACT:

Mass customization has been the mantra for today

SPEAKER: Professor Jack Hu

ABSTRACT:

Mass customization has been the mantra for today

SPEAKER: Dr. Jie Chen

ABSTRACT:

To study the genetic reasons of tumor growth, cancer formation, and genetic diseases, researchers can now use advanced bio-technologies to conduct DNA copy number experiments as it is believed that cancer development and genetic diseases are usually associated with DNA copy number changes in the genome. One of such technologies commonly used in DNA copy number experiments is called the array comparative genomic hybridization (aCGH), and the resulting data is called aCGH data. In aCGH data, gene positions, log base 2 ratio intensities along the biomarkers, and other genomic features, etc., are acquired. One of the important goals of analyzing the aCGH data is to indentify the loci at which there are abberrations of DNA copy numbers.

It turns out that identifying loci of DNA copy number changes or variations along the chromosome or genome can be viewed as a change point problem of detecting the changes presented in the sequence of log ratio intensities. In this talk, I will introduce the mean and variance change point model that is suitable for analyzing the aCGH data. Two approaches, namely the information criterion approach and a Bayesian approach, will be used to analyze the aCGH data in the proposed change-point model setting. Several applications of the proposed change point methods to the analysis of fibroblast cell line aCGH data, the breast tumor aCGH data, and the breast cancer aCGH data will be given.

SPEAKER: Eugene Feinberg

ABSTRACT:

This talk discusses whether an occupation measure for a given randomized stationary policy for a Markov Decision Process (MDPs) can be represented as a convex combination of occupation measures for simpler policies. In particular, the question is whether an occupation measure for a randomized stationary policy can be represented as a convex integral combination of occupation measures for deterministic policies. If this is possible, we say that the policy can be split.

We present general results on such representations and focus our attention on two particular situations: (i) the given subset of states is a singleton (splitting at a state), and (ii) the given subset is finite and the policy uses a finite number of actions at this set (finite splitting). For splitting at a state, we present new formulas and provide necessary and sufficient conditions when a policy can be split. For finite splitting, we describe the structure of the split and provide an algorithm for its computation. Important properties of the structure of the finite splitting are that each two consecutive policies in the split differ only at one state and the number of summands in the convex combination is relatively small.

We apply finite splitting to constrained discounted MDPs with countable state spaces and prove that an optimal mixed stationary policy can be selected in such a way that each two consecutive stationary policies in the split differ at one state. Though this talk discusses infinite state and action MDPs, the results on finite splitting are new for discounted MDPs with finite state and action sets. We shall also discuss splitting of state-action frequencies in unichain MDPs with average rewards per unit time.

This talk is based in part on a joint paper with Eric V. Denardo and Uriel G. Rothblum.

SPEAKER: Eugene A. Feinberg

ABSTRACT:

This talk discusses whether an occupation measure for a given randomized stationary policy for a Markov Decision Process (MDPs) can be represented as a convex combination of occupation measures for simpler policies. In particular, the question is whether an occupation measure for a randomized stationary policy can be represented as a convex integral combination of occupation measures for deterministic policies. If this is possible, we say that the policy can be split.

We present general results on such representations and focus our attention on two particular situations: (i) the given subset of states is a singleton (splitting at a state), and (ii) the given subset is finite and the policy uses a finite number of actions at this set (finite splitting). For splitting at a state, we present new formulas and provide necessary and sufficient conditions when a policy can be split. For finite splitting, we describe the structure of the split and provide an algorithm for its computation. Important properties of the structure of the finite splitting are that each two consecutive policies in the split differ only at one state and the number of summands in the convex combination is relatively small.

We apply finite splitting to constrained discounted MDPs with countable state spaces and prove that an optimal mixed stationary policy can be selected in such a way that each two consecutive stationary policies in the split differ at one state. Though this talk discusses infinite state and action MDPs, the results on finite splitting are new for discounted MDPs with finite state and action sets. We shall also discuss splitting of state-action frequencies in unichain MDPs with average rewards per unit time.

This talk is based in part on a joint paper with Eric V. Denardo and Uriel G. Rothblum.

SPEAKER: Professor Olympia Hadjiliadis

ABSTRACT:

We consider the problem of sequential detection of a change in the drift of independent Brownian motions and the mean of discrete-time exponential family observations received in parallel at the sensors of decentralized systems. We examine the performance of one shot schemes in decentralized detection in the case of many sensors with respect to appropriate criteria. One shot schemes are schemes in which the sensors communicate with the fusion center only once; when they must signal a detection. The communication is clearly asynchronous and we consider the case that the fusion center employs one of two strategies, the minimal and the maximal. According to the former strategy an alarm is issued at the fusion center the moment in which the first one of the sensors issues an alarm, whereas according to the latter strategy an alarm is issued when both sensors have reported a detection. In this work we derive closed form expressions for the expected delay of both the minimal and the maximal strategies in the case that CUSUM stopping rules are employed by the sensors and for the specific value of a 0 correlation across sensors. We prove asymptotic optimality of the above strategies in the case of across-sensor independence and specify the optimal threshold selection at the sensors. We also point out the special interpretation of the extreme case of a correlation of -1.

Moreover, we address the problem of quickest detection in coupled systems in models that display more general dependencies in the observations captured by general It processes. We set-up appropriate stochastic optimization problems with respect to Kullback-Leibler divergence and prove the asymptotic optimality of the N-CUSUM stopping rule in this case. We discuss applications of this work in the detection of structural damages.

The first part of the talk is joint work with H. Zhang and H.V. Poor and the second part of this work is joint work with Tobias Schaefer and H.V. Poor.

SPEAKER: Professor Brad Carlin

ABSTRACT:

Hierarchical modeling of spatial point process data has historically been plagued by computational difficulties. Likelihoods feature intractable integrals that are themselves nested within a Markov chain Monte Carlo (MCMC) algorithm. We extend customary spatial point pattern
analysis in the context of a log-Gaussian Cox process model to accommodate spatially referenced covariates, individual-level risk factors, and individual-level covariates of interest that mark the process. We also use multivariate process realizations to capture dependence among the intensity surfaces across the marks. We illustrate using a collection of breast cancer case locations collected over the mostly rural northern part of the state of Minnesota that are marked by their treatment selection, mastectomy or breast conserving surgery (``lumpectomy'). The key substantive covariate (driving distance to the nearest radiation treatment facility) is spatially referenced, but other important covariates (notably age and stage) are not. Our approach facilitates mapping of marginal log-relative intensity surfaces for the two treatment options, and resolves the issue of whether women who face long driving distances are significantly more likely to opt for mastectomy while still accounting for all sources of spatial and nonspatial variability in the data. We also briefly discuss methods for statistical boundary analysis ("wombling") in such settings.

SPEAKER: Shuangchi He

ABSTRACT:

We study G/G/n+GI queues in which customer patience times are independent, identically distributed following a general distribution. When a customer's waiting time in queue exceeds his patience time, the customer abandons the system without service. For the performance of such a system, we focus on the abandonment-count process and the queue length process. Assume the system is operated in many-server heavy traffic. We prove that, under some conditions, a deterministic relationship among the two stochastic possesses holds asymptotically under the diffusion-scaling when the number of server n to infty. The key assumption is that the diffusion-scaled queue length process is stochastically bounded. We also establish a comparison result that allows one to check the stochastic boundedness by studying a corresponding system without customer abandonment.

SPEAKER: Alejandro Toriello

ABSTRACT:

We investigate the possibility of generating good solutions to multi-period inventory routing problems by solving single-period sub-problems linked by inventory levels. The inventory tailing-off effect is counterbalanced by an approximate value function obtained via sampling and data fitting. We also study the fundamental question of whether approximate value functions allow us to shorten planning horizons so that solving multi-period problems becomes computationally tractable.

This is joint work with George Nemhauser and Martin Savelsbergh.

This is a joint SCL/DOS Seminar.

SPEAKER: Derek Bingham

ABSTRACT:

Statistical emulators of computer simulators have proven to be useful in a variety of applications. However, a widely adopted model for building these emulators, using a Gaussian process distribution with strictly positive correlation function, can be computationally intractable when the number of evaluated input values is large. We propose a modification of this model that uses a combination of low-order regression terms and compactly supported correlation functions to recreate the desired predictive behavior of the simulator at a fraction of the computational cost. Following the usual approach of taking the correlation to be a product of correlations in each input dimension, we show how to impose restrictions on the range of each correlation, giving sparsity, while also allowing the ranges to trade off against one another in a data-adaptive way, thereby giving good predictive performance when the data are non-isotropic. We illustrate the method using data from a computer simulator of photometric red-shift.

SPEAKER: David Gamarnik

ABSTRACT:

A parallel server queueing system model is used in a variety of applications including computer networks, call centers and health care management. Understanding the behavior of this system in the heavy traffic setting when the number of servers is large is a very challenging problem. While a lot is known in the special case of exponentially distributed processing times, starting with
the classical work of Halfin and Whitt in 1980, far less is known in the non-exponential case. This is unfortunate since the real life data, for example the number of days spent by patients in a hospital, often suggests distributions far from exponential.

We will present a recent progress in understanding the steady state behavior of a parallel server queueing system when the processing time distribution is arbitrary. Specifically, we establish that the basic performance measures for this queueing model have the same scaling as for the special case of exponential processing times. Then we obtain a surprisingly simple and explicit upper bound on the limiting tail distribution of the queue length. In special cases we establish the tightness of this bound.

Joint work with Petar Momcilovic (University of Michigan) and David Goldberg (MIT).

SPEAKER: Dr. Yuri Nesterov

ABSTRACT:

We propose two simple upper bounds for the Joint Spectral Radius (JSR) of arbitrary sets of nonnegative matrices. These bounds, the Joint Column Radius, and the Joint Row Radius, can be computed in polynomial time as solutions to some convex optimization problems. We show that for general matrices they are within a factor ${1 over n}$ of the exact value, where $n$ is the size of the matrices. However, for the set of matrices with independent column (or row) uncertainties, the corresponding bounds coincide with JSR. As a byproduct of this result, we get a possibility to solve in polynomial time some boolean optimization problems related to spectral radius. We present also other economical and engineering applications of our results, which were never considered in computational practice in view of their intrinsic complexity.

SPEAKER: Dr. Yuri Nesterov

ABSTRACT:

In this talk we present a new approach for finding approximate solutions to different network problems related to multi-commodity flows. We consider simple subgradient schemes and schemes based on the smoothing technique. The fastest of our methods solves the maximal concurrent flow problem in $O({qm ln n over delta})$ iterations, where $delta$ is the related accuracy, $m$ and $n$ are the number of arcs/nodes in the graph, and $q$ is the number
of commodity sources. Each iteration of these schemes is very simple and does not require any sophisticated operations (e.g. shortest path computation). Its complexity is of the order $O(mq ln q)$ operations. The application of our approach needs a preliminary computational stage
consisting in finding all node-to-node maximal flows, which takes $O(n2 m ln n)$ operations.

SPEAKER: Professor Haesun Park

ABSTRACT:

In this talk, I will present our research, education, and community building activities as the NSF/DHS FODAVA-Lead (Foundations of Data and Visual Analytics) awardees since the summer of 2008 and the plans to establish the Georgia Tech Massive Data Analysis and Visual Analytics Center. Then I will briefly cover some of my research on dimension reduction of high dimensional data such as generalization of Linear Discriminant Analysis (LDA) for high dimensional under-sampled data sets and the Nonnegative Matrix Factorization (NMF). Some experimental results from text classification, facial recognition, fingerprint classification, and 2D visualization of clustered high dimensional data demonstrate the effectiveness of the LDA based approaches. I will also present one of the fastest algorithms for the NMF and various formulations of the NMF and their applications in data analysis.

Bio: Prof. Haesun Park received her B.S. degree in Mathematics from Seoul National University, Seoul Korea, in 1981 with summa cum laude and the University President's Medal for the top graduate, and her M.S. and Ph.D. degrees in Computer Science from Cornell University, Ithaca, NY, in 1985 and 1987, respectively. She was on the faculty of the Department of Computer Science and Engineering, University of Minnesota, Twin Cities, from 1987 to 2005. From 2003 to 2005, she served as a program director for the Computing and Communication Foundations Division at the National Science Foundation, Arlington, VA, U.S.A. Since July 2005, she has been a professor in the Computational Science and Engineering Division at the Georgia Institute of Technology, Atlanta, Georgia where she is currently the associate chair. Her research interests include numerical algorithms, data analysis, bioinformatics, and parallel computing. She has published over 120 refereed research papers in these areas. She is the director of the NSF/DHS FODAVA (Foundations of Data and Visual Analytics) project where the goal is to create mathematical and computational foundations for data and visual analytics which is a newly emerging discipline of science of analytical reasoning facilitated by data analysis and interactive visualization. Prof. Park has served on numerous conference committees including conference co-chair for SIAM International Conference for Data Mining in 2008 and 2009. Currently she is on the editorial board of BIT Numerical Mathematics, SIAM Journal on Matrix Analysis and Applications, Statistical Analysis and Data Mining, and International Journal of Bioinformatics Research and Applications.

SPEAKER: Dr. Hayriye Ayhan

ABSTRACT:

We consider Markovian tandem queues with finite intermediate buffers and flexible servers and study how the servers should be assigned dynamically to stations in order to obtain optimal long-run average throughput. We assume that each server can work on only one job at a time, that several servers can work together on a single job, and that the travel times between stations are negligible. Under various server collaboration schemes, we characterize the optimal server assignment policy for these systems.

Joint work with Sigrun Andradottir and Douglas G. Down.

SPEAKER: Dr. Zonghao Gu

ABSTRACT:

The talk will begin with a brief introduction to Gurobi. This will be followed by a description of the product design both in terms of accessibility and the underlying algorithmic framework. This framework has been built from the ground up to provide maximum flexibility in exploiting modern strategies for solving mixed-integer programs (MIP). We will discuss our approaches to handling the key MIP issues in deterministic parallel, MIP search, cutting planes, simplex based warm-start for solving node relaxations etc. Finally, we will present computational results for the Gurobi MIP solver.

This is joint work with Bob Bixby and Ed Rothberg

SPEAKER: Professor Martha Grover

ABSTRACT:

Kriging is an empirical modeling approach that has been widely applied in engineering for the approximation of deterministic functions, due its flexibility and ability to interpolate observed data. Despite its statistical properties, kriging has not been developed to approximate stochastic functions or to describe the dynamics of systems with multiple outputs. Our paper proposes a methodology to construct approximate models for multivariate stochastic dynamic simulations using kriging, by combining ideas from design of experiments and dynamic systems modeling. We then apply the methodology in the prediction of a dynamic size distribution during the synthesis of nanoparticles.

SPEAKER: Dr. Meinholf Sellmann

ABSTRACT:

It is well-known that constrained optimization and constraint satisfaction problems are closely related in that being able to solve one allows solving the other. For example, the classic two-phase simplex algorithm reduces the problem of finding a feasible solution to a linear optimization problem for which achieving feasibility is trivial. Conversely, the ellipsoid algorithm reduces the linear optimization problem to a linear feasibility problem.

We study two problems on the intersection of constrained optimization and constraint satisfaction. The first regards the idea to simplify optimization problems based on feasibility and optimality considerations. We devise an expected amortized sub-linear time algorithm for the simplification of Knapsack problems and present numerical results which show speed-ups of two orders of magnitude compared to the former state-of-the-art.

The second problem regards binary search which is frequently used to augment a feasibility solver to handle optimization problems. In this setting, we often observe that negative trials (i.e. showing that a certain solution quality cannot be achieved) is significantly harder than positive trials. We consider a simple cost-model where negative trials cost a constant factor more than positive trials and show how, in this setting, binary search can be biased optimally to achieve optimal worst-case and average-case performance.

Joint work with Serdar Kadioglu, Irit Katriel, Eli Upfal, and Pascal Van Hentenryck.

SPEAKER: Aleda Roth

ABSTRACT:

The presentation addresses quality risks in outsourcing and offshoring. Recent media attention of product recalls suggests that these key supply decisions may have significant implications for manufacturing prowess and consumer opinions. However, many of the product recall reports are anecdotal; and it is unclear whether the cause of the observed increased quality risk is simply a natural result of the large number of products being outsourced or produced in offshore locations. Outsourcing and offshoring decisions may have been more complicated outcomes than originally envisioned. Their independent influences on quality risks are often confounded in the extant literature because it difficult to assess whether and to what extent the reported quality issues arose due to either inappropriate controls, which are associated with either the outsourcing of production, or different skills and coordinative challenges associated with the "offshore" location of production. Arguably, the distinction between the effects of organization and location on manufacturing outcomes is of substantial interest to supply management academics and practitioners. This presentation discusses two related empirical studies of manufacturing plants in the U.S. drug industry; and both use a new plantlevel measure of quality risk derived from Food and Drug Administration (FDA) inspection data as the dependent variable. In the first study, a theoretical model is tested on a sample of 154 FDA regulated domestic plants; and the results demonstrate that the outsourcing of production to contract manufacturers presents a significant quality risk to buying firms, which is moderated by plant age. The second study employs 33 pairs of plants in the mainland U.S. that are matched with counterparts in Puerto Rico, in order to evaluate relative onshore versus offshore quality risks. Further, we show that offshoring poses a quality risk that is qualitatively larger than that posed by outsourcing. The empirical results from the two studies offer insights quality risks associated with "pure" outsourcing and "pure" offshoring, respectively, that are useful for future research and practice.

SPEAKER: Dr. Lisa Maillart

ABSTRACT:

We present a variety of sequential medical decision making problems, both from the perspective of an individual patient as well as from the perspective of a healthcare provider. From the individual's perspective, we analyze Markov models of organ accept/decline decisions and breast cancer screening decisions. More specifically, in the former setting, we provide an overview of results for individuals entertaining liver offers from cadaveric and/or living donors under various levels of waiting list transparency, including: when the patient should accept, when the patient should update her status, in which locations the patient should join the waiting list and how the performance of the optimal policy is impacted by the degree to which waiting list information is made public. In the latter setting, we formulate a partially observed Markov chain model that captures age-based dynamics not previously considered simultaneously, and use it to evaluate a broad range of policies resulting in a menu of "efficient" breast cancer screening policies.

Finally, we touch on the management of expedited placement livers both from the perspective of a UNOS procurement coordinator as well as the perspective of a transplant center. In particular, for the procurement coordinator, we consider the questions of: when and how many standard offers to extend, and when to initiate expedited placement. For a transplant center to which an expedited organ is offered, we consider the question of which (if any) of its patients should receive the organ.

BIO:
Lisa Maillart is an Assistant Professor in the Industrial Engineering Department at the University of Pittsburgh. Prior to joining the faculty at Pitt, she served on the faculty at Case Western Reserve University. She received her M.S. and B.S. in industrial and systems engineering from Virginia Tech, and her Ph.D. in industrial and operations engineering from the University of Michigan. Her primary research interest is in sequential decision making under uncertainty, with applications in medical decision making and maintenance optimization. She is a member of INFORMS, SMDM and IIE.

SPEAKER: Dr. Lisa Maillart

ABSTRACT:

We present a variety of sequential medical decision making problems, both from the perspective of an individual patient as well as from the perspective of a healthcare provider.

From the individual's perspective, we analyze Markov models of organ accept/decline decisions and breast cancer screening decisions. More specifically, in the former setting, we provide an overview of results for individuals entertaining liver offers from cadaveric and/or living donors under various levels of waiting list transparency, including: when the patient should accept, when the patient should update her status, in which locations the patient should join the waiting list and how the performance of the optimal policy is impacted by the degree to which waiting list information is made public. In the latter setting, we formulate a partially observed Markov chain model that captures age-based dynamics not previously considered simultaneously, and use it to evaluate a broad range of policies resulting in a menu of "efficient" breast cancer screening policies.

Finally, we touch on the management of expedited placement livers both from the perspective of a UNOS procurement coordinator as well as the perspective of a transplant center. In particular, for the procurement coordinator, we consider the questions of: when and how many standard offers to extend, and when to initiate expedited placement. For a transplant center to which an expedited organ is offered, we consider the question of which (if any) of its patients should receive the organ.

BIO:
Lisa Maillart is an Assistant Professor in the Industrial Engineering Department at the University of Pittsburgh. Prior to joining the faculty at Pitt, she served on the faculty at Case Western Reserve University. She received her M.S. and B.S. in industrial and systems engineering from Virginia Tech, and her Ph.D. in industrial and operations engineering from the University of Michigan. Her primary research interest is in sequential decision making under uncertainty, with applications in medical decision making and maintenance optimization. She is a member of INFORMS, SMDM and IIE.

SPEAKER: Timo Berthold

ABSTRACT:

Constraint integer programming (CIP) is a novel way to combine constraint programming (CP) and mixed integer programming (MIP) methodologies. It is a generalization of MIP that supports the notion of general constraints as in CP. We introduce the software SCIP which is a solver and framework for constraint integer programming that also features SAT solving techniques.

We describe the various types of plugins which turn the basic CIP framework into a fully-fledged MIP solver and explain their role in the solving process. A special focus is put on branching rules, which are in the core of any branch-and-bound algorithm. We present hybrid branching, which combines variable selection rules from CP, MIP and SAT and outperforms state-of-the-art branching rules for MIPs on common MIP benchmark sets.

SPEAKER: Astronaut Shane Kimbrough

ABSTRACT:

The College of Engineering, Aerospace Engineering, and Industrial Systems and Engineering invite you to hear Astronaut Shane Kimbrough speak about his recent space shuttle mission on Friday, June 26th at 10:30 a.m. in the Georgia Tech Student Center Theatre.

Kimbrough, who flew into orbit for the first time aboard the Space Shuttle Endeavor last November, is a Lieutenant Colonel in the U.S. Army and a 1998 graduate of the Georgia Tech Stewart School of Industrial and Systems Engineering with a master's degree in Operations Research.

Kimbrough is in town for the Hometown Heroes Campaign 2009, a program that brings astronauts to their home states to share the exciting message of continued space exploration and scientific research represented by the International Space Station.

The presentation will begin at 10:30am with an autograph session at the end of the presentation.

SPEAKER: Professor Robert L. Smith

ABSTRACT:

Industry is often accused of being short-sighted in its planning and implementation at both tactical and strategic levels, with one proverbial eye firmly fixed on next quarter's profit margin. However, planning tools themselves are at least partly to blame, almost always incorporating a somewhat arbitrary finite horizon, thus inviting the end-of-study distortions inherent in finite horizon look aheads. The few exceptions that incorporate an unbounded horizon often make the heroic assumption that the future will bring a world exactly like that we confront today (e.g. the repeated plant assumption of the classical engineering economics treatment of equipment replacement). In this talk, we discuss mathematical programming models and algorithms for a general time-varying problem that allow for the finite recursive computation of optimal decisions for an unlimited horizon. The key insight here is that although the decisions made near the end of the horizon maybe seriously distorted by end of study-of-study effects, the near term and in particular the first decision will be least affected. This decoupling of the present and future decisions may be driven by discounting, uncertainty, or artificially through a tie-breaking selection. Applications will be discussed in capacity expansion and equipment replacement. Implications for solving infinite dimensional mathematical programs will be addressed.

Opportunities for NSF funding in Operations Research will be discussed at the end of the talk.

Bio

Robert L. Smith is Director of the Operations Research Program at NSF. Dr. Smith received his Ph.D. in Engineering Science from the University of California at Berkeley where he held an NSF Fellowship. He holds a bachelors degree in Physics from Harvey Mudd College and an MBA from Berkeley. He is on leave while at NSF from the University of Michigan in Ann Arbor where he is the Altarum/ERIM Russell D. O'Neal Professor of Engineering.

He is the recipient of the first Altarum/ERIM Russell D. O'Neal Professorship of Engineering at the University of Michigan. He has also been honored with the Distinguished Faculty Achievement Award from the University of Michigan, the College of Engineering Research Excellence Award, the Industrial and Operations Engineering Award for Outstanding Accomplishment, an Outstanding Teacher Award from the Michigan Student Assembly, and a National Science Foundation Fellowship. He is a Fellow of the Institute for Operations Research and the Management Sciences.

Professor Smith teaches courses in dynamic programming and stochastic processes. He has supervised the doctoral research of over twenty-five students since 1984.

At the University of Michigan, he serves as Director of the Dynamic Systems Optimization Laboratory. The Laboratory research is directed toward the modeling and analysis of dynamical systems over time. Dr. Smith worked earlier at Bell Laboratories in the Network Planning Department where he developed models and algorithms for optimal routing of communications traffic. He is an Associate Editor of Operations Research and past Associate Editor of Management Science, and is the author of nearly one hundred peer reviewed publications.

SPEAKER: Professor Sidney Resnick
Cornell University

ABSTRACT:

Using various rules, the flow of packets past a sensor can be amalgamated into higher level entities called sessions. Statistical analysis of these sessions is complex: session duration (D) and size (S) are jointly heavy tailed but average transmission rate (R=S/D) is not heavy tailed and
arrival times of sessions is not Poisson. By segmenting sessions using a peak rate covariate, we find conditional on segment that within segment session initiations can be modeled as Poisson. For modeling the distribution of (D,S,R), the conditional extreme value (CEV) model is
useful. This model is an alternative to multivariate extreme value modeling and is applicable to modeling the distribution of a random vector if some component of the vector is not in a unidimensional domain of attraction. Combining these elements, an overall model of packet flows
emerges which is suitable for simulation.
(Joint work at various times with Jan Heffernan, Bikramjit Das, Luis Lopez-Oliveros.)

Bio:
Dr. Sid Resnick received Masters and PhD. degrees from Purdue University in 1968 and 1970, respectively and has since worked at the Technion (Haifa), Stanford University, Colorado State University and for the last 20 years at Cornell University. He has authored 4 books and coauthored 151 journal articles appearing in major international journals. Sid is also a fellow of the Institute of Mathematical Statistics. His research and scholarship areas cover heavy tails, statistical analysis of extremes, service systems and networks.

PLEASE SIGN UP IN ACADEMIC AREA OR EMAIL graduate.studies@isye.gatech.edu

YOU MUST ATTEND ONE OF THESE SESSIONS

MONDAY, AUGUST 10, 2 PM
TUESDAY, AUGUST 11, 2 PM
THURSDAY, AUGUST 13, 2 PM

ROOM 217
INSTRUCTIONAL CENTER (IC)

PIZZA LUNCH
FRIDAY, AUGUST 14, NOON
ISYE MAIL BUILDING ATRIUM, 2ND FLOOR

Friday, August 14, 2009

10:00 a.m.

Executive Classroom
228 Main Building

Pizza lunch at noon
ISYE Main Building Atrium

SPEAKER: Guodong (Gordon) Pang
Columbia University

ABSTRACT:

We prove FWLLN and FCLT limits for queue-length processes, including elapsed or remaining service times, for non-Markovian infinite-server queues with delayed feedback, where both the arrivals and the feedback are allowed to be time varying. We are motivated by call centers in which customers may not initially receive adequate service, and so often need to call back later, which adds to the system load and changes time-varying arrival patterns.

Joint work with Ward Whitt

SPEAKER: Sidney Resnick

ABSTRACT:

Using various rules, the flow of packets past a sensor can be amalgamated into higher level entities called sessions. Statistical analysis of these sessions is complex: session duration (D) and size (S) are jointly heavy tailed but average transmission rate (R=S/D) is not heavy tailed and
arrival times of sessions is not Poisson. By segmenting sessions using a peak rate covariate, we find conditional on segment that within segment session initiations can be modeled as Poisson. For modeling the distribution of (D,S,R), the conditional extreme value (CEV) model is
useful. This model is an alternative to multivariate extreme value modeling and is applicable to modeling the distribution of a random vector if some component of the vector is not in a unidimensional domain of attraction. Combining these elements, an overall model of packet flows
emerges which is suitable for simulation.
(Joint work at various times with Jan Heffernan, Bikramjit Das, Luis
Lopez-Oliveros.)

Bio:
Dr. Sid Resnick received Masters and PhD. degrees from Purdue University
in 1968 and 1970, respectively and has since worked at the Technion
(Haifa), Stanford University, Colorado State University and for the last
20 years at Cornell University. He has authored 4 books and coauthored 151
journal articles appearing in major international journals. Sid is also a
fellow of the Institute of Mathematical Statistics. His research and
scholarship areas cover heavy tails, statistical analysis of
extremes, service systems and networks.

SPEAKER: David Goldberg

ABSTRACT:

We study the rate of convergence to stationarity of the M/M/n queue in the Halfin-Whitt Regime. We prove that there is an interesting phase transition in the system's behavior, occurring when a critical parameter B reaches B* sim 1.85772. For B < B*, the exponential rate of convergence is B2/4; above B* it is the solution to an equation involving the Parabolic Cylinder functions. We also bound the prefactor governing the rate of convergence uniformly over n, for B < B*.

SPEAKER: Dr. Kai Huang

ABSTRACT:

In an Assemble-To-Order (ATO) system, when the components are allocated according to First-Come-First-Serve (FCFS) rules, they may still stay in the inventory due to the lack of other components. Such inventory is called ``random stock". In the existing literature, there is no effective means to count the random stock, so that an exact cost minimization scheme for ATO system is missing. In this paper, we consider a periodic-review ATO system with base stock inventory replenishment policy. We develop a new concept ``multi-matching", which deviates from previous single item matching in the sense that multiple components must be matched with multiple products. Applying this concept, we are able to define a set of product assembly and component allocation decision variables, which allows us to develop a mathematical program that accurately and simultaneously characterizes the entire operational inventory holding cost and backlogging cost, including random stock holding cost.

We show that there exists an inherent network flow structure in this formulation. We also prove that the formulation is equivalent to a second formulation which only contains the product assembly decision. Furthermore, we compare our cost minimization model with an extension of the reward maximization model of Akcay and Xu (2004). Surprisingly, the reward maximization model can be shown to be an approximation of the cost minimization model under given condition. This relationship allows us to define a new set of holding cost and backlogging cost, which leads to a third equivalent formulation. We also explore the structure of the formulations, and propose an enhanced Benders decomposition algorithm. Our numerical experiments show that the behavior of random stock is very different from classical inventory. Moreover, in general the portion of random stock holding cost is significant enough so that it should be included in any exact cost minimization model. This also implies the necessity of non-FCFS component allocation rules in the future study.

This is a joint SCL/DOS seminar.

SPEAKER: Barry L. Nelson

ABSTRACT:

Simulation models of real-life systems often assume stationary (homogeneous) Poisson arrivals. Therefore, when nonstationary arrival processes are required it is natural to assume Poisson arrivals with a time-varying arrival rate. For many systems, however, this provides an inaccurate representation of the arrival process which is either more or less variable than Poisson, and may exhibit dependence. We extend techniques that transform a stationary Poisson arrival process into a nonstationary Poisson arrival process by transforming a stationary arrival process into a nonstationary, non-Poisson (NSNP) arrival process. We show that the desired arrival rate is achieved, and that certain variability and dependence properties of the base process are passed on to the transformed process. We also provide techniques for specifying the base process when presented with characteristics of, or data from, an arrival process and illustrate them by modeling e-mail arrival data.

Throughout his life, either as participant, support individual, or involved spectator, Richard Tapia has been involved in some aspect of drag racing. As such he has witnessed the birth and growth of many myths concerning dragster speed and acceleration.

In this talk, Professor Tapia will use his mathematical training to identify rather elementary mathematical frameworks for the study of a particular popular belief and then apply mathematics to better understand the belief at hand. In this manner, some myths are explained and validated while others are destroyed. Included in these examples will be attempts to determine how fast dragsters are really going, what is the maximum acceleration achieved, and what is the acceleration curve profile of today's dragsters? Professor Tapia will explain why dragster acceleration is greater than the acceleration due to gravity, an age-old inconsistency and present his "Fundamental Theorem of Drag Racing." The first part of the talk will be a historical account of the development of the sport of drag racing and will include shots of various family members. A component of the presentation will be several lively videos used to illustrate points.

Richard Tapia is a mathematician and professor in the Department of Computational and Applied Mathematics at Rice University in Houston, Texas. He is internationally known for his research in the computational and mathematical sciences and is a national leader in education and outreach. Tapia has authored or co-authored two books and more than 100 mathematical research papers. Among his many honors, Tapia was the first Hispanic elected to the National Academy of Engineering.

Professor Tapia is recognized as a national leader in diversity and has delivered numerous invited addresses at national and international mathematics conferences, served on university diversity committees, and provided leadership at a national level. Two professional conferences have been named in his honor, recognizing his contributions to diversity: Richard Tapia Celebration of Diversity in Computing conference and the Blackwell-Tapia Conference, whose founders described Tapia as a seminal figure who inspired a generation of African-American, Native American and Latino/Latina students to pursue careers in mathematics.

Bio
Dr. Sid Resnick received Masters and PhD. degrees from Purdue University in 1968 and 1970, respectively and has since worked at the Technion (Haifa), Stanford University, Colorado State University and for the last 20 years at Cornell University. He has authored 4 books and coauthored 151 journal articles appearing in major international journals. Sid is also a fellow of the Institute of Mathematical Statistics. His research and scholarship areas cover heavy tails, statistical analysis of extremes, service systems and networks.

Bio
Dr. Barry Nelson is well-known for his contributions to the design and analysis of computer simulation experiments on models of stochastic systems, particularly statistical efficiency, optimization via simulation, and multivariate input modeling and metamodeling. His research is driven by applications from areas such as computer-performance modeling, manufacturing systems, financial engineering and transportation.
Barry received his PhD in 1983 from Purdue University, and joined the faculty of Northwestern University after a decade at The Ohio State University. He currently serves as the Chair of the Department of Industrial Engineering and Management Sciences at Northwestern University. Among his awards and achievements, he has been chosen as an INFORMS fellow and received numerous prestigious best-paper as well as teaching awards.

SPEAKER: Steve Tyber

ABSTRACT:

In 1969, Gomory introduced the master group polyhedron for pure integer programs and derives the mixed integer cut (MIC) as a facet of a special family of these polyhedra. We study the MIC in this framework, characterizing both its facets and extreme points; next, we extend our results under mappings between group polyhedra; and finally, we conclude with related open problems.

No prior knowledge of algebra or the group relaxation is assumed. Terminology will be introduced as needed.

This is a joint work with Prof. Ellis Johnson.

This is a joint ACO/DOS seminar.

SPEAKER: Dr. Yehua Li
Dept. of Statistics, UGA

ABSTRACT:

In this paper, we consider regression models with a functional predictor and a scalar response, where the response depends on the predictor only through a finite number of projections. The linear subspace spanned by these projections is called the effective dimension reduction (EDR) space. To determine the dimensionality of the EDR space, we focus on the principal component scores of the functional predictor, and propose two sequential chi-square testing procedures under the assumption that the predictor has an elliptically contoured distribution. We further extend these procedures and introduce an adaptive Neyman test that simultaneously takes into account a large number of principal component scores. These tests can be used for model building or as goodness-of-fit tests in the context of functional linear models and functional projection pursuit models. The proposed procedures are supported by theory, validated by simulation studies, and illustrated by a real-data example. Although our methods and theory were developed under the functional data framework, they are applicable to general high-dimensional data.

SPEAKER: Dr. Yuzhao Zhang
Fox School of Business
Temple University

ABSTRACT:

We investigate the information content of the call (put) Early Exercise Premium, or EEP, defined as the normalized difference in prices between otherwisecomparable American and European call (put) options. The call EEP specifically captures investors expectations about future lump sum dividend payments aswell as other state variables such as conditional volatility and interest rates. From that perspective, the EEP should also be related to future returns of the underlying security. Little is known about the EEP, largely because it is usually unobservable for most underlying securities. The FTSE 100 index is an exceptionin that regard, because it has both American and European options contracts that are traded in large volumes. We use data of the FTSE 100 index, and its Americanand European options contracts, from which we compute a time series of the EEP. the dynamics of the EEP. We use simulations to show that these empiricalfindings can be explained with a realistically calibrated dividends process with lump-sum payments. Interestingly, we find that the EEP is a good forecaster of returns at daily horizons. This forecastability is not due to time-variation in market risk premia or liquidity. Importantly, we find that the predictability stems primarily from the ability of the EEP to forecast innovations in dividend growth, rather than other components of unexpected returns. Overall, we use several empirical and simulation methods to establish predictability of the underlying with an options market variable, link this predictability to information aboutcash flow fundamentals, and thereby provide clear support for Blacks (1975) conjecture that informed investors prefer to trade on their superior informationabout fundamentals in the options market relative to the underlying.

Bio:
Dr. Yuzhao Zhang is currently an Assistant Professor of Finance in the Fox School of Business at the Temple University. He earned his PhD in finance from the University of California Los Angeles in 2008, an M.S. of mathematics in finance from the New York University in 2002. His research interests include asset pricing, financial econometrics and derivatives. He teaches courses in investments and portfolio theory. He presented his work at University of California Los Angeles, New York University, Temple University, University of Hong Kong, Singapore Management University, Peking University, and European Finance Association Annual Meeting.

SPEAKER: Dr. Shabbir Ahmed

ABSTRACT:

I will describe a simple scheme for generating a valid inequality for a stochastic integer programs from a given valid inequality for its deterministic counterpart. Applications to stochastic lot-sizing problems will be discussed. This is joint work with Yongpei Guan and George Nemhauser and is based on the following two papers (preprints of the papers are available at http://www2.isye.gatech.edu/~sahmed/publications.html):

Y. Guan, S. Ahmed and G.L. Nemhauser. "Cutting planes for multi-stage stochastic integer programs," Operations Research, vol.57, pp.287-298, 2009

Y. Guan, S. Ahmed and G. L. Nemhauser. "Sequential pairing of mixed integer inequalities," Discrete Optimization, vol.4, pp.21-39, 2007

SPEAKER: Robert Lund

ABSTRACT:

This talk overviews changepoint issues in climate studies. Changepoints are ubiquitous features in climatic time series,occurring whenever stations relocate or gauges are changed. Ignoring changepoints can produce spurious conclusions. Changepoint tests involving cumulative sums, likelihood ratio, and maximums of F statistics are introduced; the asymptotic distributions of these
statistics are quantified under the changepoint-free null hypothesis. We find that cumulative sum procedures work best when the changepoint is near the center of the data record; otherwise, maximums of F statistics perform better. Next, issues of autocorrelation are addressed. Series with positive autocorrelation can have long sojourns above and below mean levels, hence mimicing a mean shift. We show how to modify the above methods to account for autocorrelation features. The methods are illustrated in several applications, including changes in temperatures and Atlantic Basin tropical storm counts.

Bio: Robert Lund received the Ph.D. degree in statistics from The University of North Carolina in 1993. He is currently a Professor in the Department of Mathematical Sciences at Clemson University. He is a Fellow of the American Statistical Association and was the 2005-2007 Editor of the Journal of the American Statistical Association, Reviews Section. He has published over 50 refereed papers and has graduated 8 doctoral students. His interests are in time series, applied probability, and statistical climatology.

Bio
Dr. Mike J. Todd is well-known for his contributions to linear and nonlinear optimization and game theory. His research interests are in algorithms for linear and convex programming, particularly semidefinite programming. He is interested in developing and analyzing interior-point methods; previous research interests include homotopy methods, probabilistic analysis of pivoting methods, and extensions of complementary pivoting ideas to oriented matroids.

Among his several awards, he has received the George B. Dantzig Prize jointly from the Mathematical Programming Society and SIAM in 1988, and the John von Neumann Theory Prize from INFORMS in 2003.

SPEAKER: Mike Todd

ABSTRACT:

We discuss the convergence properties of first-order methods for two problems that arise in computational geometry and statistics: the minimum-volume enclosing ellipsoid problem and the minimum-area enclosing ellipsoidal cylinder problem for a set of m points in R^n. The algorithms are old but the analysis is new, and the methods are remarkably effective at solving large-scale problems to high accuracy.

SPEAKER: Yasuo Amemiya
Manager, Statistical Analysis & Forecasting
Business Analytics & Mathematical Sciences Department
IBM T. J. Watson Research Center

ABSTRACT:

This talk starts with providing an introductory overview of IBM Research, and then presents some of IBM's statistical research activities in business analytics. Business analytics system development is a rapidly growing area in enterprise software and services business. IBM has been involved in this area actively and deeply, as reflected by the introduction of a new business analytics service practice offering, and by the acquisitions of Cognos, ILOG, and SPSS. Statistical research conducted in our department is also aligned closely with this new strategic area for IBM. In this talk, a few specific research projects as related to statistical method development for business analytics are described.

SPEAKER: Prof. Yufeng Liu

ABSTRACT:

Quantile regression is a very useful statistical tool to learn the relationship between the response variable and covariates. For many applications, one often needs to estimate multiple conditional quantile functions of the response variable given covariates. Although one can estimate multiple quantiles separately, it is of great interest to estimate them simultaneously. One advantage of simultaneous estimation is that multiple quantiles can share strength among them to gain better estimation accuracy than individually estimated quantile functions. Another important advantage of joint estimation is the feasibility to incorporate noncrossing constraints of quantile regression functions. In this talk, I will present a new multiple noncrossing quantile regression estimation technique. Both asymptotic properties and finite sample performance will be presented to illustrate usefulness of the proposed method.

Bio: Dr. Liu is an Associate Professor in the Department of Statistics and Operations Research at The University of North Carolina at Chapel Hill. He received his MS and PhD from The Ohio State University. He is the recipient of the NSF Career Award (2008). He is an associate editor of the Journal of the American Statistical Association. His research interests are in Statistical Machine Learning and Data Mining, High Dimensional Data Analysis, Nonparametric Statistics, Bioinformatics and DOE.

SPEAKER: Shuangchi He

ABSTRACT:

We first study G/G/n+GI queues in which customer patience times are independent, identically distributed following a general distribution. We prove that, under some conditions, a deterministic relationship holds asymptotically between the abandonment-count and the queue-length processes under the diffusion scaling, when the number of servers goes to infinity. Using this relationship, we prove that for critically loaded G/Ph/n+GI queues with phase-type service times, a pair of diffusion-scaled customer-count and server-allocation processes, properly centered, converges in distribution to a continuous Markov process. We also develop a numerical algorithm based on the finite-element method to compute the steady-state distribution of the diffusion limits of the queue-length process in the G/Ph/n+GI queues.

Joint work with Jim Dai and Tolga Tezcan.

SPEAKER: Kirk Wang
Director at SystematIQ Research
A Standard and Poors Capital IQ Unit

ABSTRACT:

An introduction into quantitative equity investment process as well as the challenges and opportunities faced by today's quantitative investors. The recent market condition has raised questions about the future alpha-generating potential of quantitative investment strategies. We will discuss this topical issue and the implication of homogeneous approaches and heavily mined data sets in the alpha construction process. We will explore the impact and importance of technology, unconventional data sets and present several new techniques in discovering new alpha sources for institutional quantitative investors. Utilizing Capital IQs Key Development database and ClariFIs industry-leading analytical software, we have examined the impact of event driven information to various traditional quantitative stock selection strategies. Well review the related academic studies and methodologies and share our approach in using the event driven signals to enhance stock selection models.

Bio:

Kirk Wang, Director at SystematIQ Research, Standard and Poors Capital IQ business. His responsibilities include global quantitative equity research, research technologies and product development. Prior to Capital IQ, he was a quantitative analyst with Mesirow Advanced Strategies, a Funds of Hedge Funds group where he performed analysis on various global hedge fund strategies. Prior to Mesirow, he was a quantitative analyst with Quantitative Services Group, an independent quantitative equity research provider. He holds an MBA from the University of Chicagos Booth School of Business.

SPEAKER: Luyi Gui

ABSTRACT:

A network is called a collaborative one if the network users share edge resources. We study a collaborative multicommodity flow (MCF) network model with multiple source-sink pairs and design a collaboration mechanism based on capacity exchange prices so as to subtly regulate the selfish behaviours of the players. It can be shown that efficient prices can be found so that the social welfare is maximized even under a decentralized setting in static MCF networks with capacity and demand levels fixed. Moreover, under certain conditions the individual payoffs induced under the mechanism are fair in terms of that no one profits less by joining the grand coalition. We also consider the sensitivity of such a price mechanism to the variations in capacity and demand levels and introduce some prospective approaches to reinforce the robustness of the mechanism.

This is a joint work with Prof. Ozlem Ergun.

SPEAKER: Paul Kvam

ABSTRACT:

To measure carbon nanotube lengths, atomic force microscopy and special software are used to identify and measure nanotubes on a square grid. Current practice does not include nanotubes that cross the grid, and as a result, the sample is length biased. The selection bias model can be demonstrated through Buffon's Needle Problem, which was extended to general curves that more realistically represent the shape of nanotubes observed on a grid. In this paper, the nonparametric maximum likelihood estimator is constructed for the length distribution of the nanotubes, and the consequences of the length bias are examined. Probability plots reveal that the corrected length distribution estimate provides a better fit to the Weibull distribution than the original selection-biased observations, thus reinforcing a previous claim about the underlying distribution of synthesized nanotube lengths

SPEAKER: Sarang Deo
Kellogg School of Management
Northwestern University

ABSTRACT:

Emergency department (ED) overcrowding and consequent ambulance diversion is one of the most pressing healthcare delivery problems in the US and many other developed countries including the UK. In this paper we develop a two-stage queuing model to study the impact of structural factors such as the size and utilization of the inpatient department and the size of the ED on ambulance diversion status of the ED. In the face of analytically intractability, we develop two approximations based on heavy traffic theory and derive appropriate measures of inpatient occupancy and ED size that are associated with the extent of ambulance diversion. We then test the key insights from the analysis of these approximations using cross-sectional data of hospitals from California. We find that one of our approximations provides a better description of the data than the other. We also try to fit an empirical model with raw measures of size and occupancy and find that it does not explain the data well thereby providing additional support for our theoretical results.

Speaker Bio: Sarang Deo joined Kellogg School of Management in 2007 after completing his PhD from UCLA Anderson School of Management. His primary research interest is application of operations management methods to design efficient and effective healthcare delivery systems. His current work focuses on issues in resource-poor countries (developing models for HIV treatment scale-up, studying the impact of patient load on quality of care) as well as the US (joint planning of HIV screening and treatment at the Veterans Administration, theoretical and empirical investigation into the causes of ambulance diversion).

SPEAKER: Dr. Grace Y. Lin

ABSTRACT:

Supply Chain Management (SCM) has gone through significant changes with some fundamental breakthroughs by both academia and industry in the last two decades. SCM has evolved from back office operations to being the main driver of business performance. As companies transform into globally integrated enterprises, continual cost reductions, functional optimization, and efficiency improvement become the normal fabric of business, but they are not sufficient for today's rapidly changing environments. Companies have to continuously examine market conditions and customer needs, enhance their agility, flexibility, and core competencies, and collaborate with their value net partners to stay competitive and be sustainable in today's market place.

In this talk, Dr. Lin will discuss supply chain evolution and emerging themes. She will share her experience in supply chain research and practice, and describe some of her work with IBM and other companies, including the Edelman award-winning Extended Enterprise SCM, Sense-and-Respond Value Net and latest efforts in building a smarter planet. Finally, she will discuss the key competencies of a sustainable globally integrated enterprise.

Bio

Dr. Lin is an internationally recognized leader and innovator in Supply Chain Management (SCM) and Value Net Optimization. She recently founded the World Resource Optimization, Inc., and the Green Value Net not-for-profit organization to help build a more harmonious, intelligent, and sustainable planet. From 2003 to 2009, she was the Global Sense-and-Respond Leader, then the CTO and Director, Innovation and Emerging Solutions, for IBM's Global Business Services, where she worked with both public and private sector clients. She has served as Researcher, Manager, and Senior Manager for SCM and as the Relationship Manager for IBM's Integrated Supply Chain (ISC) organization at the IBM T.J. Watson Research Center from 1993 to 2003.

Dr. Lin built a world-class R&D team at IBM in the 90's which pioneered extended enterprise supply chain management. She led the IBM team to win the 1999 INFORMS Franz Edelman Award for helping IBM save $750M in supply chain transformation. Dr. Lin initiated IBM's Sense-and-Respond Value Net efforts, and founded IBM's Value Chain Innovation Center in early 2000. She later created IBM's Sense-and-Respond consulting offerings and led cross-functional, cross-geographic teams to bring state-of-the-art business optimization models and technologies to IBM and its customers. The Value-Driven Services and Delivery framework that Dr. Lin led greatly enhanced the way services and solutions are delivered today. Dr. Lin also led the IBM Virtual Command Center effort, and was a key member of the IBM Smarter Planet Technology Team from 2008 to 2009. Some of the awards Dr. Lin's received include: Institute of Industrial Engineers' Doctoral Dissertation Award, IBM's Outstanding Technical Achievement Award, IBM Invention Achievement Awards, and Purdue University's Outstanding Industrial Engineer Award. Dr. Lin was elected to the IBM Academy of Technology in 2002, named an IBM Distinguished Engineer, and an INFORMS Fellow in 2006. Media coverage for Dr. Lin and her teams' SCM achievements included Forbes Magazine, the World Journal News, and Forrester, who cited Dr. Lin as one of six Supply Chain Gurus in its Supply Chain Report in 2002.

Dr. Lin has co-authored more than 60 technical articles and seven patents, with another five patents pending. She is a frequent keynote and invited speaker at international conferences, and has served on four editorial boards, several university advisory boards, and on National Science Foundation panels of US, Canada and Ireland. Dr. Lin has chaired two INFORMS Practice conferences, an IEEE conference, and an INFORMS Women in ORMS Award committee, and was twice elected as the VP Practice for INFORMS. Dr. Lin received her Ph.D. in Industrial Engineering and an M.S. in Applied Mathematics from Purdue University, and an M.S. and a B.S. in Mathematics from Tsing-Hua University in Taiwan. Dr. Lin has dedicated herself to promoting diversity and mentoring for many years. She is grateful for the support she has received from her husband, two children, family, friends, and colleagues throughout her professional career and personal life.

SPEAKER: Mor Harchol-Balter

ABSTRACT:

It is well-known that when job size variability is high, one needs to prevent short jobs from getting stuck behind long jobs. In a server farm setting, one way to achieve this goal is to allocate short jobs their own server (or set of servers). This is the theory behind the popular Size Interval Task Assignment policy (SITA) for server farms,which assigns each server a unique size range, so that short jobs are given isolation from long ones. The SITA policy is prevalent throughout compute server farms and manufacturing systems, whenever job size variability is high. The higher the job size variability, the more important it is to provide short jobs some isolation from long ones, via a SITA policy, or some variation thereof.

This talk questions the above common wisdom. To understand what's going on, we study the performance of task assignment policies, in the limit, as the variability of job sizes (service demands) approaches infinity. Results in this limiting regime reveal that the SITA policy can be far inferior to much simpler greedy policies, like Least-Work-Left (LWL), for many common job size distributions, including a range of Pareto distributions. Regimes are also defined where SITA's performance is good, and here simple closed-form bounds are proved on its performance. Towards the end of the talk we will also consider the performance of SITA variants/hybrids.

Parts of this work appeared in ACM SIGMETRICS 2009.

JOINT WORK WITH: Alan Scheller-Wolf and Andrew Young

BIO: Mor Harchol-Balter is Associate Department Head of the Computer Science Department at Carnegie Mellon University. She received her doctorate from the Computer Science department at the University of California at Berkeley under the direction of Manuel Blum. She is a recipient of the McCandless Chair, the NSF CAREER award, the NSF Postdoctoral Fellowship in the Mathematical Sciences, multiple best paper awards, and several teaching awards, including the Herbert A. Simon Award for Teaching Excellence. She is heavily involved in the ACM SIGMETRICS research community, and recently served as Technical Program Chair for SIGMETRICS. Mor's work focuses on designing new resource allocation policies (load balancing policies, power management policies, and scheduling policies) for server farms and distributed systems in general. Her work spans both queueing analysis and systems implementation, and emphasizes integrating measured workload distributions into the problem solution.

"Model-Based Data Mining for Functional Data Under Spatial Interdependence" 

Abstract:

In this seminar, I will present data mining methods for discovering and summarizing patterns in functional data observed under spatial interdependence. The field of functional data analysis has already provided a series of competitive data mining approaches, but they are generally limited to the assumption of independence between the random functions. This assumption is rather restrictive in many research applications.

In the first part of this seminar, I will introduce a model-based method for clustering random functions which are spatially interdependent. The time functions are decomposed into spatial global and time-dependent cluster effects using a semi-parametric model. We assume that the clustering membership is a realization from a Markov random field. In the case study presented in this paper, we focus on obtaining temporal cluster trends for racial-ethnic diversity for five southeast states in the US.

In the second part of this seminar, I will introduce a computational efficient and theoretically-founded cross-correlation analysis. Under the proposed semi-parametric model, we show that the cross-correlation estimators are asymptotically unbiased under the conditions that the sample size is large and the intrinsic dimensionality of the functional processes is much smaller than the sample size. We illustrate this correlation analysis within a demographic study, in which we analyze the association between per capita income and racial-ethnic diversity.

This is joint work with Huijing Jiang, PhD student in ISyE, Georgia Institute of Technology

Bio:

Nicoleta Serban is an assistant professor in ISyE. She received her B.S. in Mathematics and an M.S. in Theoretical Statistics and Stochastic Processes from the University of Bucharest. She went on to earn her Ph.D. in Statistics at Carnegie Mellon University. Before joining Georgia Tech, Dr. Serban's research focused on nonparametric statistical methods motivated by recent applications from proteomics and genomics. Dr Serban's current research focusses on multiple functional estimation and clustering with applications to industrial performance, service site location, socio-economics and NMR biomolecular studies. Visit Dr. Serban's website.

Please join us for a reception preceding the seminar outside Klaus 1324, beginning at 1:30pm.

To receive future announcements, please sign up to the cse-seminar email list: https://mailman.cc.gatech.edu/mailman/listinfo/cse-seminar

"Generalized Fast Multipole Method"

Abstract:
I will present a novel FMM technique applicable to both oscillatory and non-oscillatory
kernels. The method is very flexible and accurate. It is more general than existing FMMs and can often compete with nearly optimal schemes for specific kernels like 1/r. I will present the progress of the research with some preliminary numerical results.

Bio:

 Visit Dr. Darve's website.

You are cordially invited to attend a reception preceding the seminar to chat informally with faculty and students. Reception will take place outside Klaus 1324 between 1:30 - 2:00.

To receive future announcements, please sign up to the cse-seminar email list: https://mailman.cc.gatech.edu/mailman/listinfo/cse-seminar

Professor, Mathematics and Computer Science Department, Emory University 

"Efficient Iterative Methods for Large Scale Inverse Problems"

Abstract:

Ill-posed problems arise in many image processing applications, including microscopy, medicine and astronomy.  Iterative methods are typically recommended for these large scale problems, but they can be difficult to use in practice.  In this talk we describe a hybrid approach that combines the Golub-Kahan bidiagonalization algorithm with Tikhonov regularization and a weighted generalized cross validation scheme.  We also show how this method can be used as an effective scheme for choosing regularization parameters for certain nonlinear inverse problem.  Applications from image processing illustrate the effectiveness of the resulting numerical schemes.

This is joint work with Julianne Chung, University of Maryland.

Bio:

James Nagy is a Professor of Mathematics and Computer Science at Emory University.  He received his Ph.D. in Applied Mathematics from North Carolina State University in 1991.  Before joining Emory University in 1999 he had postdoctoral research fellowships with the IMA at the University of Minnesota, with the NSF at the University of Maryland, and was on the faculty at Southern Methodist University.  He is on the editorial boards of SIAM Journal on Scientific Computing (SISC), SIAM Journal on Matrix Analysis and Applications (SIMAX), and the SIAM Book Series "Fundamentals of Algorithms".  His research interests include numerical linear algebra, structured matrix computations, and numerical solution of inverse problems in image processing. Among his many publications includes the recent SIAM book, "Image Deblurring: Matrices, Spectra and Filtering", which is co-authored with Per Christian Hansen and Dianne O'Leary. Visit Dr. Nagy's website.

Please join us for a reception preceding the seminar outside Klaus 1324, beginning at 1:30pm.

To receive future announcements, please sign up to the cse-seminar email list: https://mailman.cc.gatech.edu/mailman/listinfo/cse-seminar