2016 INFORMS Annual Meeting Program

451 / 561

INFORMS Nashville – 2016
451
WC77
Legends E- Omni
Opt, Large Scale I
Contributed Session
Chair: Gerdus Benade, Carnegie Mellon University,
6235 Fifth Avenue, Apt B17, Pittsburgh, PA, 15232, United States,
1 - An Agglomerative Clustering Based Large-Scale Minimum Cost
Flow Algorithm
Yuan Zhou, PhD Candidate, University of Memphis,
174 Windover Cove, Apt 3, Memphis, TN, 38111, United States,
Stephanie Ivey
When the size of an input network exceeds computer hardware capabilities,
minimum cost flow (MCF) becomes problematic in terms of computational and
memory requirements. This research focuses on the large-scale MCF problem by
adopting a divide-and-conquer policy during the optimization process. We
propose an agglomerative clustering based tiling strategy which can ensure
consistency between local and global optimum solutions, decomposes the input
network into sub-networks, and then solves MCF in each sub-network
independently to reduce peak memory consumption and improve efficiency.
2 - Large Scale Scheduling Problems On Internet Of Things
Carlos Eduardo De Andrade, Senior Inventive Scientist, AT&T Labs
Research, 200 S Laurel Avenue, Room A5-1E33, Middletown, NJ,
07748, United States,
The number of connected devices has been increasing exponentially over past
years. Such devices do not only use the network for their main purposes, such as
collect and share data but also in the maintenance mode for updates. We present
a scheduling problem to deal with massive download updates on millions of
devices connected to radio networks. One of the main challenges of this problem
is the fact of the devices are mobile and can connect to several access points in the
network over the time, and the server usually can not control the download
process other than suggesting a start time to the download. The problem also
includes other constraints such as network and server capacities, and devices
limitations.
3 - Optimum Product Introduction And Strategic Capacity Planning
Including Demand And Price Cannibalization
Gorkem Yilmaz, Assistant Professor, Ozyegin University,
Cekmekoy, Istanbul, 34794, Turkey,
, Amaia Lusa Garcia,
Ernest Benedito
In literature, product introduction and strategic capacity planning typically
address optimal timing, pricing decisions and capacity management
independently regardless of demand and price cannibalization effects. We develop
a Mixed Integer Linear Program (MILP) model for solving strategic capacity
planning and product introduction problems, simultaneously, including demand
and price cannibalization. With dynamic demand and pricing senario, we analyze
several numerical examples to display the interaction between demand and price.
4 - Benders Decomposition And Column-and-row Generation
For Solving Large-scale Linear Programs With
Column-dependent-rows
Kerem Bulbul, Associate Professor, Sabanci University, Faculty of
Engineering & Natural Sciences, Orhanli, Tuzla, Istanbul, 34956,
Turkey, , Ilker S Birbil, Ibrahim Muter
We study a general class of LP problems - known as problems with column-
dependent-rows (CDR-P). These LPs feature two sets of constraints with mutually
exclusive groups of variables in addition to a set of structural linking constraints,
in which variables from both groups appear together. The number of linking
constraints grows very quickly with the number of variables, which motivates
generating both columns and their associated linking constraints simultaneously
on-the-fly. The structure of CDR-P is amenable to Benders decomposition and
leads to an effective approach. The unavailability of a full description of the
Benders subproblem over the iterations is our main theoretical challenge.
5 - The Branching Dual Of A Discrete Optimization Problem
Gerdus Benade, Carnegie Mellon University, 6235 Fifth Avenue,
Apt B17, Pittsburgh, PA, 15232, United States,
, John Hooker
We formulate the branching dual of a discrete optimization problem as
maximizing an inferred lower bound over the space of partial search trees. We
identify a class of optimization problems with limited information transfer for
which a greedy node selection rule is approximately optimal. A computational
study compares the efficiency of the branching dual with other methods for
proving bounds on the minimum bandwidth problem. It is found that the
branching dual requires significantly fewer nodes than the alternatives to prove a
given bound.
WC78
Legends F- Omni
Health Care, Modeling
Contributed Session
Chair: Harshal Lowalekar, Indian Institute of Management-Indore,
Prabandh Shikhar, Rau-Pithampur Road, Indore, 453331, India,
1 - A Microsimulation Cost-effectiveness Analysis Of Nalmefene To
Reduce Heavy Alcohol Consumption In An Alcohol-dependent
Canadian Population
Estefania Ruiz Vargas, Ivey Business School, 1255 Western Rd,
London, ON, N6G 0N1, Canada, Richard Zur,
Greg Zaric
Nalmefene is an opioid antagonist believed to reduce the analgesic and positive
reward effect of alcohol. A microsimulation of alcohol consumption was
developed to determine if nalmefene combined with psychosocial support is cost-
effective for reducing alcohol consumption in an alcohol-dependent Canadian
population. We considered a number of different nalmefene treatment
approaches, and evaluated the public health benefit of implementing nalmefene
as a pharmacological intervention for alcohol dependence in Canada.
2 - Managing A Multi-physician Clinic With Heterogeneous Patients
Wen-Ya Wang, San Jose State University, San Jose, CA,
United States, , Hao-Wei Chen
This paper focuses on the physician panel design problem of determining the
patient composition (i.e. panel size and types of patient) for physician panels. In
particular, we investigate how clinics would allocate a given pool of two types of
patient flows that have different appointment patterns (i.e. low frequent/routine
service vs. high frequent/complex care needs). We identify the characteristics of
patient and capacity allocation strategies from the perspectives of the clinics and
the patients
3 - A Markovian Model For Blood Components Production System
Harshal Lowalekar, Indian Institute of Management-Indore,
Prabandh Shikhar, Rau-Pithampur Road, Indore, 453331, India,
We develop a Markovian model for the production of major components such as
red blood cells, platelets and plasma from whole blood at a blood bank. The
supply quantity of whole blood is assumed to be a random variable. Using the
Markovian approach we determine the optimal target collection level for whole
blood and the percentage of whole blood to be separated into components.
WC79
Legends G- Omni
Opt, Combinatorial I
Contributed Session
Chair: Christopher John Wishon, Ph.D. Candidate, Arizona State
University, Tempe, AZ, United States,
1 - Reformulation-linearization Technique Application On Integer
Programming Models For Organ Exchange Program
Seokhyun Chung, Korea University, Seoul, Korea, Republic of,
Junsang Yuh, Taesu Cheong
Kidney exchange allows a potential living donor whose kidney is incompatible
with the intended recipient to donate a kidney to another patient so that the
donor’s recipient receives a compatible kidney from another donor. This can be
modeled as maximum weight cycle packing problems in a directed graph. In this
study, we verify relationship between existing models via reformulation-
linearization technique (RLT), and develop a new integer programming model for
kidney exchange program by implementing the RLT. In addition, we devise new
integer programming model for liver exchange program, which has distinct
character. Moreover, we enhance the integer programming model by applying
RLT.
2 - Solution Approaches To Network Design Problems With Decision
Dependent Uncertainty
Nathaniel Richmond, University of Iowa, 446 4th Avenue, Iowa
City, IA, 52241, United States, ,
Pavlo A Krokhmal, Dmytro Matsypura
Robust network design problems have many important practical applications, and
for this reason are well-represented in the operations research literature.
However, until the last decade very little research has examined stochastic robust
network design problems in which the user’s decisions affect the underlying
probability distribution of the random outcome. In this talk, we present a model
for the stochastic network design problem with decision-dependent uncertainties.
We discuss computational challenges and explore different solution approaches.
In particular, we present a metaheuristic algorithm that draws from strengths of
GRASP and genetic algorithms.
WC79