![Show Menu](styles/mobile-menu.png)
![Page Background](./../common/page-substrates/page0107.png)
INFORMS Nashville – 2016
105
SD33
203B-MCC
Recent Advances in Simulation
Sponsored: Simulation
Sponsored Session
Chair: Jing Dong, Northwestern University, Evanston, IL, United States,
jing.dong@northwestern.eduCo-Chair: Jose Blanchet, Columbia University, New York, NY,
United States,
jose.blanchet@gmail.com1 - On Calibrating Statistical Distances
Huajie Qian, University of Michigan, 2302 St Francis Drive, Apt
B118, Ann Arbor, MI, 48104, United States,
hqian@umich.eduHenry Lam
We present a general framework to calibrate the statistical distance dictating the
size of the uncertainty sets for distributionally robust optimization used in
stochastic or simulation optimizations under uncertainty. We discuss the
implications on the statistical guarantees of the resulting objective values and
feasibility. We also compare these guarantees to sample average approximation.
2 - Multi-resolution Gaussian Markov Random Fields For Discrete
Optimization Via Simulation
Eunhye Song, Northwestern University, Evanston, IL, United
States,
EunhyeSong2016@u.northwestern.eduBarry L Nelson, Jeremy C Staum
The Gaussian Markov Improvement Algorithm (GMIA), an optimization via
simulation algorithm based on Gaussian Markov random fields (GMRF), has
computational advantages in solving problems on a large discrete solution space.
We extend GMIA to a multiresolution algorithm (MR-GMIA) to solve even larger
problems. The solution space is divided into regions; each region becomes a
“solution” in a region-level GMRF while solutions within each region are
represented by a solution-level GMRF. Using complete expected improvement,
MR-GMIA guides the search toward promising regions and promising solutions
within the selected regions with global inference about the optimality gap for
termination.
3 - Unbiased Monte Carlo Computations For Optimization And
Functions Of Expectations
Yanan Pei, Columbia University,
yp2342@columbia.edu,
Jose Blanchet, Peter W Glynn
We present general principle for the design and analysis of unbiased Monte Carlo
estimators for quantities such as functions of expectations. Our estimators possess
finite work-normalized variance under mild regularity conditions. We apply our
estimator to various settings of interest, such as optimal value estimation in the
context of Sample Average Approximations, unbiased estimators for particle filters
and conditional expectations.
4 - Estimation In The Tail Of The Gaussian Copula
Raghu Pasupathy, Purdue University, West Lafayette, IN, 47907,
United States,
pasupath@purdue.eduWe present ecoNORTA for efficient constrained random vector generation within
the Gaussian and NORTA contexts. We propose three importance-sampling
estimators for such settings, the first of which actively exploits knowledge of the
local structure of the feasible region around a dominating point to achieve
bounded relative error. The second and third estimators, for use in settings where
information about the constraint set is not readily available, do not exhibit
bounded relative error but are shown to achieve a slightly weaker form of
efficiency. Numerical results on various example problems show promise.
SD34
204-MCC
Joint Session HAS/MSOM-HC: Models and Analytics
in Healthcare Operations
Sponsored: Manufacturing & Service Oper Mgmt,
Healthcare Operations
Sponsored Session
Chair: Joel Goh, Harvard Business School, Boston, MA,
United States,
jgoh@hbs.edu1 - Accurate Prediction Of Case Duration
Amirhossein Meisami, University of Michigan, Ann Arbor, MI,
United States,
meisami@umich.edu,Nick Kastango,
Christopher Thomas Borum Stromblad, Mark P Van Oyen
The primary goal of this study is to analyze the abundant data available prior to
surgery and leverage this information to produce accurate case length predictions
via novel statistical learning methodologies. We will be working with a rich
database from Memorial Sloan Kettering Cancer Center to identify the essential
features in defining case duration variability. The research also focuses on
reducing the uncertainties and variations imposed by rare events that may arise in
various procedures during a case.
3 - Admission Of Long Stay Patients In A Busy Pediatric ICU
Fernanda Bravo, Assistant Professor, UCLA Anderson School of
Management, Los Angeles, CA, United States,
fernanda.bravo@anderson.ucla.edu, Michael McManus
This work studies admission policies for complex patients in the ICU of a large
pediatric academic hospital. There are four different patient types: medical,
emergency, surgical, and transfers. Within these, long-stay-patients use a large
amount of resources and limit the access to the unit. The ICU must always remain
available for emergencies before accommodating elective admissions. As a result,
many children are queued for complex surgeries and medical workups. We study
policies to decide when to admit an long-stay-patient depending on the current
ICU status, and future patients’ arrivals.
4 - Scheduling Work In Radiology
Maria R. Ibanez, Harvard Business School,
mibanez@hbs.eduUsing detailed data on millions of radiological studies interpreted by physicians,
we study the drivers of speed and quality of the interpretation, and identify
implications for scheduling and allocation of work across workers.
SD35
205A-MCC
Strategic Queueing
Sponsored: Manufacturing & Service Oper Mgmt,
Service Operations
Sponsored Session
Chair: Laurens Debo, Dartmouth College,
Tuck School of Business, Hanover, NH, 03755, United States,
Laurens.G.Debo@tuck.dartmouth.eduCo-Chair: Luyi Yang, University of Chicago, Booth School of Business,
Chicago, IL, 60637, United States,
luyi.yang@chicagobooth.edu1 - Queueing With Strategic Balking And System Design
Yichen Tu, University of North Carolina, Chapel Hill, NC,
United States,
yichen1@live.unc.edu, Nur Sunar, Serhan Ziya
We analyze a queueing system where customers decide whether to join or balk
the system depending on the expected benefit they receive by joining a system. In
this setting, we characterize the optimal choice of system design from the
perspective of a social welfare optimizer. We also conduct numerical studies to
shed light on the benefit of different system design choices.
2 - Risk/return Trade-off In Queues With A Nonlinear Waiting
Cost Function
Hossein Abouee-Mehrizi, University of Waterloo, 200 University
Avenue West, Waterloo, ON, N2L 3G1, Canada,
haboueemehrizi@uwaterloo.ca,Ata G Zare, Renata Konrad
We consider an M/M/1 queueing system and assume that each customer receives
a value by getting served and suffers from a waiting cost. To analyze customers’
behavior, we consider the risk/return trade-off and a nonlinear waiting cost
function. Customers are impatient and have a mixed attitude with respect to the
risk. Before reaching a certain point in time, customers are risk-seeking, but after
that they become risk-averse. We assume that customers follow a joint balking
and abandonment strategy. We fully characterize the equilibrium joint balking
and abandonment strategy and show that three types of equilibria may exist:
global, myopic, and farsighted.
3 - Optimal Information Disclosure In M/M/1 Queues
Shiliang Cui, Georgetown University, 548 Rafik B. Hariri Building,
37th & O Streets, NW, Washington, DC, 20057, United States,
shiliang.cui@georgetown.edu, Jinting Wang
Queue length is a very important parameter for customers to make a joining
decision or not. We study optimal information disclosure policies in M/M/1
queues.
4 - Want Priority Access? Refer Your Friend To Move Up In Line
Luyi Yang, University of Chicago,
lyang6@chicagobooth.eduLaurens G Debo
This paper studies the referral priority program, an emerging business practice
adopted by a growing number of technology companies that manage a waitlist of
customers. The program enables existing customers on the waitlist to gain priority
access if they successfully bring in new customers. We find that the effectiveness
of this novel mechanism as a marketing tool for customer acquisition and an
operational approach for waitlist management depends crucially on the base
arrival rate of the system. Referrals may not be generated when the base arrival
rate is either too high or too low. Even when customer refer, the program could
backfire (i.e., reduces the system throughput and customer welfare).
SD35