Informs 2015

383 / 552

INFORMS Philadelphia – 2015
381
WA24
24-Room 401, Marriott
Robustness and Approximation in Markov Decision
Processes
Sponsor: Artificial Intelligence
Sponsored Session
Chair: Marek Petrik, IBM, 1101 Kitchawan Rd., Yorktown Heights, NY,
10598, United States of America,
1 - Algorithms for Risk-sensitive Optimization in MDPS
Mohammad Ghavamzadeh, Senior Analytics Researcher, Adobe
Research, 321 Park Ave., E7412, San Jose, CA, 95126, United
States of America,
In many sequential decision-making problems we may want to manage risk by
minimizing some measure of variability in costs in addition to minimizing a
standard criterion. We consider variance-related and percentile-based risk-
sensitive criteria. For each criterion, we devise algorithms for estimating its
gradient and updating the policy parameters in the descent direction. We establish
the convergence of our algorithms and demonstrate their usefulness in a variety
of control problems.
2 - Ambiguous Joint Chance Constraints with Conic
Dispersion Measures
Grani A. Hanasusanto, École Polytechnique Fédérale de
Lausanne, EPFL-CDM-MTEI-RAO, Station 5, Lausanne, 1015,
Switzerland, , Daniel Kuhn,
Wolfram Wiesemann, Vladimir Roitch
We analyse the complexity of a class of ambiguous joint chance constrained
programs where the uncertain parameters are described through their mean
values and through upper bounds on general dispersion measures. We derive
explicit conic reformulations for tractable problem classes and suggest efficiently
computable conservative approximations for intractable ones. We illustrate the
effectiveness of our reformulation in numerical experiments in project
management and image denoising problems.
3 - Learning the Uncertainty in Robust Markov Decision Processes
Xu Huan, Assistant Professor, National University of Singapore, 9
Engineering Drive 1, Singapore, Singapore,
Shie Mannor, Shiau Hong Lim
Robust MDP models the parameter uncertainty as arbitrary element of
uncertainty sets, and seeks the minimax policy. A crucial problem of robust MDP
is how to find appropriate uncertainty. We address this using an online learning
approach: we devise an algorithm that, without knowing the true uncertainty
model, is able to adapt its level of protection to uncertainty, and in the long run
performs as good as the minimax policy knowing the uncertainty model.
4 - Robust Approximate Dynamic Programming
Marek Petrik, IBM, 1101 Kitchawan Rd., Yorktown Heights, NY,
10598, United States of America,
I describe how robust MDPs can be used to improve solution quality of both on-
policy and policy approximate dynamic programming methods. The robustness
addresses both model and sampling error. Finally, I show the utility of robust
optimization when computing implementable policies in MDPs.
WA25
25-Room 402, Marriott
Managing Sustained Participation in
Online Communities
Sponsor: Information Systems
Sponsored Session
Chair: Pratyush Nidhi Sharma, Assistant Professor, University of
Delaware, 010 Purnell Hall, University of Delaware, Newark, DE,
19716, United States of America,
1 - The Impact of Person-organization Fit and Psychological
Ownership on Turnover in Open Source Software Projects
Tingting Rachel Chung, Chatham University,
106 Woodland Road, Pittsburgh, PA, United States of America,
, Pratyush Nidhi Sharma, Sherae Daniel
Open source software projects represent an alternate form of software production
by relying on voluntary contributions. Most projects fail to sustain their
development due to high turnover. Using 574 survey responses from GitHub, we
examined the impact of Person-Organization fit and psychological ownership on
developers’ turnover intentions. Results show value and demands-abilities fit
negatively impact turnover intentions and that psychological ownership
moderates these effects.
2 - Coordinating Co-opetition: Insights from Open-source Cloud
Software Development
Yash Raj Shrestha, ETH Zurich, Weinbergstrasse 56/58, Zurich,
Switzerland, , Shiko Ben-Menahem,
Georg Von Krogh
Using longitudinal data from OpenStackóan open-source cloud computing
software development platform, this study explores why some participating firms
exhibit greater success than others in their ability to coordinate activities in a co-
opetitive ecosystem. Focusing on patterns of strategic task allocation and
completion by firms facing strong competition for highly skilled developers, our
study advances understanding on co-opetition and coordination in new
organizational forms.
3 - The Movement of Open Source Communities
Georg J.P. Link, University of Nebraska Omaha,
6001 Dodge St, Omaha, NE, 68182, United States of America,
Matt Germonprez
In 2010, Oracle acquired OpenOffice during its acquisition of Sun Microsystems.
At that time, community members formed the LibreOffice fork under The
Document Foundation. And in the following year, Oracle transferred ownership
of OpenOffice to the Apache Foundation. We explore thresholds for communal
movement and how the transfer of open source projects affects the community.
We find that understanding communities as movable reveals their nature as
commoditized and consolidated objects of value.
WA26
26-Room 403, Marriott
Production and Scheduling II
Contributed Session
Chair: Katariina Kemppainen, School of Business, Aalto University,
Runeberginkatu 22-24, Helsinki, 00076 Aalt, Finland,
1 - A Capacitated Multi-Item Lot-Sizing Problem with Stochastic
Setup Times
Raf Jans, Professor, HEC Montreal, 3000 Chemin de la
Cote-St-Catherine, Montreal, QC, H3T 2A7, Canada,
Michel Gendreau, Duygu Tas, Ola Jabali
We introduce uncertainty with respect to the setup times in the standard
capacitated lot sizing problem. The company fixes a production plan (i.e. timing
and level of the production quantities). The company can use overtime if the
given capacity is not sufficient due to the specific realizations of the setup times.
We develop an efficient procedure to evaluate the expected overtime assuming a
specific probability distribution. We also present several MIP-based heuristics to
solve this problem.
2 - Issues in Batch Flowshop and Lot Streaming Problems
Ramakrishna Govindu, Instructor, University of South Florida,
8350 N Tamiami Tr, SMC-C263, Sarasota, FL, 34243, United
States of America, Anurag Agarwal
The lot streaming problem attempts to find sublots to reduce the makespan. We
treat this problem as a multiobjective problem that attempts to strike a balance
between makespan and cost of handling the sublots. We propose some heuristics
and properties of the problem.
3 - Routing and Spectrum Assignment in Rings
Sahar Talebi, North Carolina State University, Operations
Research and Computer Science, Raleigh, NC 27695,
United States of America,
We present a theoretical study of the routing and spectrum assignment (RSA)
problem in ring networks. We show that the RSA problem with fixed-alternate
routing in general topology networks is a special case of a multiprocessor
scheduling problem. We then investigate two problems: the spectrum assignment
problem under the shortest path assumption and the general RSA whereby a
routing decision must be made jointly with spectrum allocation. We then develop
a suit of heuristic algorithms.
4 - Cutting Stock with Sequence Dependent Set-up Times:
An Application to a Large Scale Industry Problem
David Wuttke, EBS University, EBS University, ISCM, Burgstr. 5,
Oestrich-Winkel, 65375, Germany, ,
Sebastian Heese, Florian Gojny
We consider a two-dimensional cutting stock problem with sequence dependent
set-up times and tolerances as witnessed in textile and fiber-composite industries.
To solve real-life-instances we provide a decomposition heuristic that first
identifies optimal cutting patterns and then optimizes their sequence by
minimizing the number of knife relocations.
WA26