Informs Annual Meeting 2017

MB01

INFORMS Houston – 2017

MA82

important operational implications. For example, network externalities prompt the firm to provide better services to customers. We also propose a dynamic look- ahead heuristic policy with optimality gaps decaying exponentially in the forward-looking time length. 3 - Analysis of Ordering Strategies under Market Competition and Capacity Constraint Weili Xue, Southeast University, Hankou Road 22, Nanjing, China, weili@seu.edu.cn In this paper, we consider a supply chain with a manufacturer and two retailers, in which the two retailers compete in the same consumer market and the manufacturer has capacity constraint. Each manufacturer can order before or after demand uncertainty realization. We characterize the equilibrium ordering strategies and show that an asymmetric equilibrium may exist even for the symmetric setting when the supplier’s capacity is constrained and the wholesale prices are exogenously determined. We further extend our result to the case when the capacity is unlimited and when the wholesale prices are endogenously determined.

382B Optimization and Learning in Decision Processes Sponsored: Optimization, Optimization Under Uncertainty Sponsored Session Chair: Mahshid Salemi Parizi, Seattle, WA, 98105, United States, msalemip@uw.edu Co-Chair: Archis Ghate, University of Washington, University of Washington, Shoreline, WA, 98155, United States, archis@u.washington.edu 1 - Weakly Coupled Markov Decision Processes with Parametric Uncertainty Mahshid Salemi Parizi, 3801 Brooklyn Ave NE, Apt M304, Seattle, WA, 98105, United States, msalemip@uw.edu We consider a class of stochastic dynamic programs where multiple, otherwise independent, subproblems are linked via constraints. This problem class is known as weakly coupled Markov decision processes (MDP). We consider the case where transition probabilities for each subproblem are unknown to the decision maker. The decision maker thus attempts to balance learning (exploration) with reward maximization (exploitation). Tractable approximate dynamic programing methods rooted in the Bayesian learning framework will be presented. 2 - Backward ADP with Hidden Semi-Markov Information Models Joseph L.Durante, Princeton University, Princeton, NJ, United States, jdurante@princeton.edu, Juliana Nascimento, Bolong Cheng, Warren B.Powell We consider an energy storage problem involving a wind farm with a forecasted power output, a power demand, an energy storage device, and a connection to the larger power grid (with associated electricity prices). The stochastics are modeled using a novel hidden semi-Markov model which accurately replicates both forecast error distributions and the periods of time for which a process stays above or below its forecast. We derive near-optimal time-dependent policies using a technique we call backward approximate dynamic programming, which overcomes the computational hurdles of classical (exact) backward DP, and achieves higher quality solutions than the more familiar forward ADP methods. 3 - Optimal Learning with General Nonlinear Belief Models Xinyu He, Princeton University, Princeton, NJ, 08544, United States, xinyuhe@princeton.edu, Warren B.Powell We consider the problem of optimizing an unknown function over a multidimensional continuous space, where function evaluations are noisy and expensive. We assume that a globally accurate model of the function is not available, but there exist some parametric models that can well approximate the function locally. We propose an algorithm, using local approximation technique and optimal learning policies, to quickly learn the shape of the function and find the optimal design. Experiments on both synthetic problems and a real materials science application show the strong performance of our algorithm. 382C Joint Session OPT/Practice: Stochastic Inventory Management Sponsored: Optimization, Optimization Under Uncertainty Sponsored Session Chair: Xiting Gong, The Chinese University of Hong Kong, Hong Kong, xtgong@se.cuhk.edu.hk 1 - Dynamic Inventory Management under Stockout Substitutions Tianxiao Chen, The Chinese University of Hong Kong, ERB615, CUHK, Shatin, Hong Kong, chentx@se.cuhk.edu.hk, Xiting Gong In this paper, we consider a periodic-review inventory control problem with two products which can be substituted if one of them is out of stock. We formulate the problem as a two-dimensional dynamic program and characterize the structure and asymptotic bounds of the optimal policy under a wide range of parameter settings. 2 - Dynamic Pricing and Inventory Management under Network Externalities Renyu Zhang, Assistant Professor, New York University Shanghai, 1555 Century Avenue, Shanghai, 200122, China, renyu.zhang@nyu.edu, Nan Yang We study the impact of network externalities on a firm’s joint pricing and inventory policy under demand uncertainty. A customer’s willingness-to-pay and, thus, the potential demand are increasing in the past sales of the product. We show that the presence of network externalities gives rise to the tradeoff between generating current profits and inducing future demands, thus having several MA83

Monday, 10:00 - 10:50AM

Monday Plenary Hilton- Ballroom of Americas, Level 2 Optimization: Past, Present, Future Plenary Invited Session 1 - Sports Scheduling Meets Business Analytics Bob Bixby, Gurobi

For the vast majority of business applications, optimization means linear and mixed-integer programming. Beginning with Dantzig’s simplex method in 1947, optimization experienced a slow, uneven period of development into the mid 1980s. Then, beginning in the late 1980s, developments ensued that completely transformed optimization and its applications, driven by truly remarkable performance improvements in the underlying solvers. What’s coming next may be even more exciting. Driven by an explosion in available business data, a new broad corporate focus on extracting value from that data, increased computing power, and the continually expanding power of optimization solvers, optimization promises to become an indispensable tool in managing the modern enterprise.

Monday, 11:00 - 12:30PM

MB01

310A Risk Attitudes Sponsored: Decision Analysis Sponsored Session Chair: Andrea Hupman Cadenbach, University of Missouri-St. Louis, St Louis, MO, 63121, United States, cadenbach@umsl.edu 1 - Risk Attitudes in Disaster Planning Vicki Bier, University of Wisconsin, Madison, WI, United States, vicki.bier@wisc.edu This talk discusses some implications of risk attitudes (specifically, risk aversion) for disaster planning. Surprisingly, if we take risk aversion seriously, it may mean that we are being overly protective against radiation-induced fatalities following nuclear incidents. However, “psychic numbing” may make it difficult to use risk aversion to encourage preparedness. 2 - Risk Attitude Modeling in Homeland Security Decisions Ali E. Abbas, University of Southern California, OHE 310R, 3650 McClintock Ave, Los Angeles, CA, 90089, United States, aliabbas@usc.edu Decision Analysis is a normative approach for making decisions in the face of uncertainty, and can be applied to a variety of decisions. The field of utility theory is often misused and replaced by other non-normative approaches under the umbrella of decision analysis. This talk will discuss an application of utility theory in homeland security decisions and discuss some widely used (and flawed) approaches.

163