Informs 2015

98 / 552

INFORMS Philadelphia – 2015
96
4 - The World at Our Fingertips: Consumer Conversion from Search,
Click-through, to Book
Karen Xie, Assistant Professor, University of Denver,
Fritz Knoebel School of Hospitality Mgmt, Denver, CO,
United States of America, Young-Jin Lee
We examine how informational cues displayed influence consumer conversion in
a sequential process of search, click-through, and booking when shopping hotels
online. Using an empirical analysis of a large online travel agent site that provides
information of individual searches, we find consumers are likely to click through
hotels with higher consumer ratings and the industry-endorsed star rating.
However, when committing to booking consumers refer to consumer ratings
rather than the star rating.
SC11
11-Franklin 1, Marriott
Joint Session OPT/OPT Under Uncertainty:
IP Methods for Stochastic Optimization
Sponsor: Optimization/Integer and Discrete Optimization
Sponsored Session
Chair: Siqian Shen, Assistant Professor, University of Michigan,
1205 Beal Avenue, Ann Arbor, MI, 48105, United States of America,
1 - Scenario Decomposition of Risk Averse Stochastic 0-1 Programs
Yan Deng, University of Michigan, 3730 Greenbrier Blvd,
Ann Arbor, Mi, 48105, United States of America,
, Siqian Shen, Shabbir Ahmed
We present a scenario decomposition approach for risk averse stochastic 0-1
programs. The approach uses decomposition that processes subproblems easy to
parallelize, and LP/IP techniques that combine subproblem results to yield strong
valid inequalities. We also design various parallelization paradigms and
demonstrate on stochastic server location problems and stochastic multiple 0-1
knapsack problems that the approach improves computational efficacy
significantly.
2 - Polyhedral Study of Static Probabilistic Lot-sizing Problem
Xiao Liu, Research Associate, The Ohio State University, 210
Baker Systems Building 1971 Neil Ave, Columbus, OH, 43210,
United States of America, , Simge Kucukyavuz
We explore the polyhedral structure of the static probabilistic lot-sizing problem
and propose new valid inequalities. We show the relationship between the
proposed inequalities and those that would be obtained by projecting the natural
or the extended formulation of the deterministic lot-sizing problem for each
scenario on to the space of the production variables. We show that the proposed
inequalities are facet-defining under certain conditions. We report preliminary
computational results.
3 - Computational Experience with a Benders Algorithm for Two-
stage Stochastic Integer Programming
Ted Ralphs, Professor, Lehigh University, 200 W. Packer Ave.,
Bethlehem, PA, 18015, United States of America, ,
Anahita Hassanzadeh
We briefly outline a Benders-type algorithm for solving two-stage stochastic
programs with mixed integer recourse and describe our experience implementing
the algorithm in practice, as well as its theoretical connection to bilevel
optimization. Computational results with benchmark instances from the literature
will be presented.
SC12
12-Franklin 2, Marriott
Mixed Integer Programming and Location Routing
Problem
Sponsor: Optimization/Mixed Integer Nonlinear Optimization and
Global Optimization
Sponsored Session
Chair: Emre Tokgoz, Assistant Professor, Quinnipiac University,
275 Mount Carmel Ave. TH-ENR, Hamden, CT, 06518,
United States of America,
1 - Manifold Location Routing Problem
Emre Tokgoz, Assistant Professor, Quinnipiac University,
275 Mount Carmel Ave. TH-ENR, Hamden, CT, 06518,
United States of America,
A new Location Routing Problem named Manifold Location Routing Problem and
the most recent developments to solve this problem will be explained in this talk.
2 - Cost Effective Energy Optimization by Solving Facility Allocation
on Riemannian Manifolds
Iddrisu Awudu, Assistant Professor, Quinnipiac University,
275 Mount Carmel Ave., Hamden, CT, 06518,
United States of America,
Delivering goods and services in an efficient (cost) and effective (value) manner is
crucial to every supply chain. In this presentation, a recently developed technique
for solving the Location Routing Problem on Manifold surfaces called Manifold
Location Routing Problem (MLRP) is used to design a renewable energy
distribution system for a single ethanol production facility. The results with the
corresponding managerial insights of the model are discussed.
SC13
13-Franklin 3, Marriott
Multi-armed Bandits and Online Optimization
Sponsor: Optimization/Optimization Under Uncertainty
Sponsored Session
Chair: Andrew Lim, National University of Singapore/Department of
Decision Sciences, Mochtar Riady Building, BIZ1 08-69, 15 Kent Ridge
Drive, Singapore, Singapore,
1 - Dynamic Pricing and Product Differentiation with Cost Uncertainty
and Learning
Bora Keskin, Duke University, Fuqua School of Business,
100 Fuqua Drive, Durham, NC, 27708-0120, United States of
America, John Birge
Motivated by applications in the health insurance industry, we consider a seller
who designs and sells a set of products to a population of quality-sensitive
customers. The seller faces an uncertainty about production costs. We prove that,
while a seller facing static cost uncertainty degrades the quality in its product
offering, a dynamically learning seller improves the quality of its products to
accelerate information accumulation.
2 - Efficient Experimentation via the Bootstrap
Ian Osband, Stanford University, 450 Serra Mall, Stanford, CA,
United States of America, ,
Benjamin Van Roy
If you want to use data to make good decisions, you need good data. Often, the
decisions you make influence both performance and the data collected. Balancing
exploration with exploitation can be complex. We present a fully non-parametric
version of Thompson sampling that uses the bootstrap to produce approximate
posterior samples. We show that, under some assumptions, this simple heuristic
satisfies strong performance guarantees. This algorithm performs well in
simulation.
3 - Approximate Learning Trajectories for Bayesian Bandits
Michael Kim, University of Toronto, 5 King’s College Road,
Toronto, Canada, Andrew Lim
It is known that the optimal policy for a multi-armed bandit problem is the Gittins
index policy. For bandit problems with Bayesian learning (Bayesian bandits)
however, computing the Gittins index is intractable due to the curse of
dimensionality. In this talk, we introduce the concept of an approximate learning
trajectory, which approximates the dynamics of future learning. We show how
this can be used to simplify the DP equations, which allows for an efficient
computation of the Gittins index.
4 - Bandits with Global Convex Constraints and Concave Objective
Shipra Agrawal, Researcher, Microsoft Research,
#9 Lavelle Road, Bangalore, India,
I will present recent advances on multi-armed bandit problems that involve
arbitrary concave objective functions and convex constraints on the aggregate of
decisions across time, in addition to the customary limitation on the time horizon.
SC14
14-Franklin 4, Marriott
New Developments in Robust and Adaptive
Optimization
Sponsor: Optimization/Optimization Under Uncertainty
Sponsored Session
Chair: Dimitris Bertsimas, Professor, MIT, 77 Massachusetts Ave.,
Cambridge, MA, 02139, United States of America,
1 - Multistage Robust Mixed Integer Optimization with
Adaptive Partitions
Iain Dunning, PhD Candidate, MIT, 77 Massachusetts Ave,
Building E40, Cambridge, MA, 02139, United States of America,
Dimitris Bertsimas
SC11