Informs 2015

410 / 552

INFORMS Philadelphia – 2015
408
WB24
24-Room 401, Marriott
Joint Session AI/ICS: Decision Diagrams for
Optimization and Artificial Intelligence
Sponsor: Artificial Intelligence and ICS
Sponsored Session
Chair: Andre Augusto Cire, Assistant Professor, University of Toronto
Scarborough, 1265 Military Trail, Toronto, ON, M1C 1A4, Canada,
1 - Consistency as Projection
John Hooker, Carnegie Mellon University, 5000 Forbes Avenue,
Pittsburgh, PA, 15213, United States of America,
We interpret various forms of consistency in constraint programming as
projection and propose a new form of consistency, J-consistency, that is achieved
by projecting the constraint set onto a subset J of variables. We show how J-
consistency can reduce backtracking when constraints are propagated through
decision diagrams. We introduce J-consistency algorithms for SAT, among,
sequence, regular, and alldiff constraints.
2 - Decision Diagram Decomposition of Optimization Problems
David Bergman, ,
Andre Augusto Cire
Decision diagrams (DDs) can be used to compactly represent solutions to
optimization problems, but the representation grows exponentially large in
general, prohibiting their direct application. However, it is often the case that the
diagram has limited size when representing the solution set for a substructure of a
given problem. In this talk we discuss how DDs can be used when a problem
decomposes in such a way that each individual portion has a limited-size DD.
3 - Decision Diagrams for Efficient Inference and Optimization in
Expressive Continuous Domains
Scott Sanner, Asst. Professor, Oregon State University,
1148 Kelley Engineering Center, Corvallis, OR, 97331,
United States of America,
I will introduce the XADD — an extension of the algebraic decision diagram to
continuous variables — and show how to define and efficiently compute
elementary arithmetic operations, integrals, and maximization for various
restrictions of these functions. I will then cover a range of applications where the
XADD has yielded novel closed-form solutions: (a) structured probabilistic
inference, (b) parametric constrained optimization, and (c) sequential decision-
making problems.
4 - Sentential Decision Diagrams and their Applications
Adnan Darwiche, UCLA, 4532-D Boelter Hall, Los Angeles, CA,
90095-1596, United States of America, ,
Guy Van Den Broeck, Arthur Choi
Sentential decision diagrams (SDDs) are a new class of decision diagrams that
branch on arbitrary sentences instead of individual variables. They generalize
ordered binary decision diagrams (OBDDs) and carry over many desirable OBDD
properties, such as canonicity and support for bottom-up compilation with
Boolean operators, while being more compact than OBDDs. These key properties
have enabled several successful AI applications in recent years, particularly for
reasoning about uncertainty.
WB25
25-Room 402, Marriott
Economics of Information Systems
Sponsor: Information Systems
Sponsored Session
Chair: Liangfei Qiu, Assistant Professor, University of Florida,
Department of ISOM, Gainesville, FL, United States of America,
1 - Cross-market Integration and Sabotage
Hong Guo, University of Notre Dame, 356 Mendoza College of
Business, Notre Dame, IN, 46556, United States of America,
, Yabing Jiang, Asoo Vakharia, Arthur Lim
Recent industry developments motivate the study of cross-market firm
integrations, which often raises controversies and regulatory concerns due to the
potential negative effects through the integrated firms’ sabotage activities. In this
paper, we analyze integrations of firms in two interrelated markets and the
integrated firm’s incentive to sabotage its rival in both markets. Our findings
provide important managerial and policy implications for cross-market firm
integrations.
2 - What Makes Geeks Tick? A Study of Stack Overflow Careers
Tingting Nian, NYU Stern, 44 West 4th Street, New York, NY,
10012, United States of America, ,
Luis Cabral, Lei Xu
We study user contributions to Stack Overflow, the largest online programmers
community. Using a diff-in-diff approach, we show that the event of finding a
new job implies a reduction of 25% in reputation-generating activity, but only a
reduction of 8% in non-reputation-generating activity. We consider a series of
robustness tests to tease out alternative explanations for these variations;
together, the results suggest that career concerns play an important role in
explaining user contributions.
3 - Sentiment Manipulation in Online Platforms and Opinion Forums:
A Regression Discontinuity Approach
Shun-yang Lee, University of Texas at Austin, Austin, TX, United
States of America, , Andrew Whinston,
Liangfei Qiu
Online platforms are prone to abuse and manipulations from strategic parties. We
characterize strategic firms’ manipulation incentives through in a rational
expectation equilibrium framework (REE), and conduct empirical analysis on a
Twitter data set through a regression discontinuity design (RDD) approach. We
find that both the average Twitter sentiment and the proportion of highly positive
tweets exhibit a significant drop on the movie’s release day, suggesting firms’
strategic manipulation.
4 - Open Platform Ecosystem Dynamics and System Efficiency with
Advertising Investment
Gou Qinglong, Associate Professor, University of Science &
Technology of China, No.96, JinZhai Road Baohe District,
Hefei, China, Ruibing Wang, Yonghua Ji
We utilize a dynamic model to explore the advertising strategies for an open
platform ecosystem where the platform and app are to determine their advertising
efforts. We find that a transfer payment can improve system efficiency
significantly.
WB26
26-Room 403, Marriott
Production and Scheduling III
Contributed Session
Chair: Alejandro MacCawley, Assistant Professor, P. Universidad
Catolica de Chile, Vicuna Mackenna 4860 Macul, Santiago, 7820436,
Chile,
1 - Minimizing Demand Oriented Maintenance Cost
Marcus Poggi, PUC-Rio Informatica, Rua Marquís de São Vicente,
225, Gávea, Rio de Janeiro, Brazil, ,
Ivan Lima
In a production environment, machines fail with a given probability, function of
its operating time since last maintenance. Facing a demand scenario for a
planning horizon, one is interested in estimating the efficient frontier for
maintenance cost and risk of not fulfilling a demand SLA. This stochastic problem
with endogenous uncertainty is modeled over a discrete space-time network.
Heuristic and exact approaches are devised.
2 - Minimizing Working Capital Requirements in a Dynamic Lot
Sizing Model with Infinite Capacity
Thomas Yeung, Associate Professor, Ecole des Mines de Nantes,
4 Rue Alfred Kastler, BP20722, Nantes, 44307, France,
, Nathalie Bostel, David Lemoine,
Yuan Bian
Traditional tactical production planning models do not consider a minimum
financial need in terms of working capital requirements (WCR) to maintain
operations. We introduce a first link by proposing a new generic WCR model for
the dynamic lot sizing problem with single-site/level/product and infinite
capacity. We show that our model adheres to the zero-inventory ordering policy
allowing an exact polynomial algorithm to be formulated. Numerical tests
compare the results with traditional models.
3 - Minimizing Tardiness and Maintenance Costs in Flow Shop
Scheduling using Lower-bound-based GA
Andrew Yu, Associate Professor, The University of Tennessee,
Industrial and Systems Engineering, Knoxville, TN, 37996,
United States of America, Javad Seif
A permutation flow shop scheduling problem is reformulated as a mixed-integer
linear program after incorporating flexible and diverse maintenance activities for
minimizing tardiness and maintenance costs. The MILP model is NP-hard for a
large size of problem. Thus a lower-bound based genetic algorithm is developed.
The algorithm is further tuned using design experiment to control and determine
the GA parameters.
WB24