Management Sciences Seminar Series 2006-2007
Abstracts

Sept 29 (Fri), 2:30-3:20, W207 PBB

Stochastics in transportation and logistics – some modeling issues

Stein W. Wallace
Professor of Quantitative Logistics
Molde University, Norway

Most decisions are made under some uncertainty about the future. Most models do not take that explicitly into account. Is that a minor issue or a major problem? The purpose of this lecture is to illustrate how deterministic models, even combined with sensitivity analysis, what-if-sessions or other similar tools, can lead to rather bad decisions. Using examples from logistics we show how solutions from deterministic models differ from those of stochastic models in systematic and (sometimes) recognizable ways. We also try to show ways forward in terms of capturing these differences without solving (unsolvable) stochastic optimization problems. So the overall goal of the underlying research is to find good (if not optimal) solutions to stochastic optimization problems without solving such problems.

Oct 13 (Fri), 2:30-3:20, W207 PBB

Real-time Equipment Prognostics for Adaptive Replacement and Spare Parts Inventory Control

Nagi Z. Gebraeel
Assistant Professor
Mechanical and Industrial Engineering

The uncertainty associated with equipment failures poses significant challenges in determining efficient logistical decision making policies. Accurate predictions of equipment failure times are necessary to improve the soundness of these decisions. Most of the existing spare parts logistical models focus on using population-specific reliability characteristics, such as failure time distributions to develop decision making strategies. Since these distributions are unaffected by the underlying physical degradation processes, they do not distinguish between the different degradation characteristics of individual components of the population. Failing to capture the evolution of these characteristics compromises failure predictability and increases the risks associated with maintenance logistical decisions, such as replacement and inventory decisions.

This talk focuses on the development of sensory-driven decision model for component replacement and spare part inventory. First, we develop a sensory-updated degradation modeling framework for computing remaining life distributions (RLDs) using in-situ condition-based sensory data. Next we integrate the sensory-updated residual life estimates with existing renewal theoretic replacement and inventory decision models. This enables the dynamic updating of replacement and inventory decision based on equipment health.

Oct 20 (Fri), 2:30-3:20, W207 PBB

The Hub Covering Flow Problem

Thaddeus Sim
PhD Student

Traditional models for locating hubs in a hub-and-spoke transportation network either solve for hub locations to minimize installation cost of hubs subject to distance restrictions to demand points served (e.g., the hub covering problem), or to minimize total flow costs without consideration of distance constraints (e.g., the uncapacitated hub location problem). We develop and test a model that essentially combines the attributes of these two types of models. The model allows for the use of multiple coverage radii with transportation costs depending upon coverage zones. The model can be used to assist in the design of hub-and-spoke networks and in the selection of types of carriers to serve demand points from hubs.

Oct 27 (Fri), 2:30-3:20, W207 PBB

Opportunity Costs, Dynamic Pricing and Real-time Vehicle Routing in Sequential Auction Marketplaces for Freight Procurement

Hani S. Mahmassani
Charles Irish Sr. Chair in Civil and Environmental Engineering
Director, Maryland Transportation Initiative
Department of Civil and Environmental Engineering
The University of Maryland at College Park

The focus of this presentation is online transportation marketplaces with time-sensitive truckload pickup-and-delivery requests. In this environment, demands arrive randomly over time and are described by pick up, delivery locations and hard time-windows. Upon demand arrival, carriers compete for the loads in a second price auction. Two main questions are addressed in this presentation: (1) how to quantify the opportunity costs, of accepting a current load, in terms of ability to serve future loads, and set prices/bids accordingly; and (2) how to compare different technologies for dynamic vehicle routing; the technologies differ in how they deal with the combinatorial and stochastic elements of the online problem.

Nov 3 (Fri), 2:30-3:20, W207 PBB

Finding the Minimum Network that Maximizes Service

Hui Chen
PhD Student

Many national delivery providers offer delivery time commitments between cities. These commitments direct the design of the delivery network. In this network, each arc defines a direct connection between cities and represents a substantial investment. The minimum cost network that connects all cities is a tree, but a tree may not satisfy all of the delivery commitments. We examine how to find trees that minimize the delivery time violations and present computational results based on the U.S.

Joint Inventory and Pricing Decisions in a Consignment Channel with Competing Suppliers

Dengfeng Zhang
PhD Student

We consider a supply chain with two suppliers of substitutable products on consignment to a single retailer. The retailer (Stakelberg leader) sets the revenue share, while the suppliers (followers) decide on the product prices and inventory levels. We develop equilibrium model solutions to investigate how channel decisions and changes in exogenous parameters affect channel profits. Other channel settings such as exclusive retailer channel and wholesale channel are also explored and compared.

Nov 17 (Fri), 2:30-3:20, W207 PBB

Learning with Model Inaccuracy: Unexpected Consequences

Tito Homem-de-Mello
Associate Professor
Department of Industrial Engineering and Management Sciences
Northwestern University

Many applications in science and engineering are addressed by formulating models that approximate some real system. Oftentimes, these models depend upon some parameters whose values are unknown, in which case learning methods are typically used. Often underlying the use of such techniques is the assumption that there exist parameter values for which the model provides a very good approximation of the real system. However, in many cases - especially in those that involve human behavior - this assumption does not hold. In such cases one can say that the underlying models are inaccurate. Results of attempts at learning with an inaccurate model can be radically different from those at learning with an accurate model, and may result in a systematic decline in the quality of decisions. For example, in airline revenue management practitioners are concerned with a decline in revenues despite the use of apparently "good" revenue management practices. One cause of that particular problem has been identified as the inaccuracy of widely used models, some of which ignore the fact that customers may choose among available alternatives rather than arrive with a specific product in mind.

In this talk we will discuss these issues, focusing on some particular examples - the airline case mentioned above, and pricing models that fail to take competition into account. Results from our research so far suggest that unexpected conclusions may be reached e.g., that in some cases having a more accurate model may be counter-productive.

Joint work with William L. Cooper and Anton Kleywegt.

Dec 1 (Fri), 2:30-3:20, W207 PBB

Learning to Rank by Linear Programming

Kaan Ataman
PhD Student

Ranking is a popular machine learning problem that has been studied extensively for more then a decade. Typical machine learning algorithms are generally built to optimize predictive performance (usually measured in accuracy) by minimizing classification error. However, there are many real world problems, where correct ordering of instances is of equal or greater importance than correct classification. Learning algorithms that are built to minimize classification error are often not effective when ordering within or among classes. This gap in research created a necessity to alter the objective of such algorithms to focus on correct ranking rather then classification.

Area Under the ROC Curve (AUC) is a widely accepted performance measure for evaluating ranking performance in binary classification problems. The AUC is equivalent to the Wicoxon-Mann-Whitney (WMW) statistic. We present a linear programming approach, similar to 1-norm Support Vector Machines (SVM), for instance ranking in the context of binary classification by an approximation to the WMW statistic. Our formulation handles non-linear problems by making use of kernel functions. We have tested our ranking algorithm on several well-known benchmark datasets for performance comparison. The results show that our approach performs favorably compared to state of the art classification algorithms when an RBF kernel is used. The number of constraints in the linear programming formulation increases quadratically with the number of data points considered for the training of the algorithm. We tackle this problem with implementing a number of exact and approximate speed-up approaches inspired by well-known methods such as chunking and clustering.

Apr 13 (Fri), 2:30-3:20, S181 PBB

Economics of Brief Reductions in Setup, Procedure, and/or Cleanup Times in Operating Rooms

Franklin Dexter, MD, PhD (more info)
Director, Division of Management Consulting
Professor, Departments of Anesthesia and Health Management & Policy
University of Iowa

A common question for OR/MS/IE personnel in healthcare is how to reduce process time to increase patient flow. However, often such efforts are relatively unimportant. The costs are typically fixed on the day of surgery. Furthermore, the workday is brief while marginal revenue is small relative to variable costs. The consequence is that queues should be brief, and reducing processing time should not result in increased production. The benefit of discrete event simulation of the processes has, in fact, been mostly as a mechanism (analogous to Monte-Carlo simulation) to test statistical methods to analyze operating room managerial data.