2016 INFORMS Annual Meeting Program

TD77

INFORMS Nashville – 2016

TD77 Legends E- Omni Opt, Integer Programing IV Contributed Session Chair: Joseph B Mazzola, Cleveland State University, 1860 East 18th Street, BU 530, Cleveland, OH, 44115, United States, j.b.mazzola@csuohio.edu 2 - Using Odheuristics To Solve Hard Mixed Integer Programming Problems Alkis Vazacopoulos, Optimization Direct, Inc., 202 Parkway, Harrington Park, NJ, 07640, United States, alkis@optimizationdirect.com, Robert Ashford It is not practical to prove optimality for most large scale MIP models. Indeed, many are so computationally onerous that it is not possible to raise the best bound at all beyond the root solve. ODHeuristics is a general purpose program built on CPLEX for obtaining good feasible solutions to such MIPs. It is designed for scheduling problems but works for any MIP which has a reasonable number of integer feasible solutions. It has been deployed effectively on packing problems, supply chain and telecoms as well as scheduling applications. This talk looks at what ODHeuristics does and how - in general terms - it goes about it with reference to some simple examples. 3 - Objective Scaling Ensemble Approach For Integer Linear Programming Weili Zhang, University of Oklahoma, 202 W. Boyd St., Room 116, Norman, OK, 73019, United States, weili.zhang-1@ou.edu, Charles D. Nicholson The objective scaling ensemble approach is a novel, approximate solution procedure for integer linear programming problems\deleted[id=WZ]{in general} \added[id=WZ]{shown to be effective on a wide variety of ILP problems}. The technique identifies and aggregates multiple partial solutions to modify the problem formulation and significantly reduce the search space. An empirical analysis on widely available difficult problem instances demonstrate the efficacy of our approach by outperforming the existing advanced solution strategies implemented in modern optimization software. 4 - Preventive Maintenance And Replacement Scheduling Farzad Pargar, University of Oulu, Oulu, Finland, farzad.pargar@oulu.fi, Jaakko Kujala In this paper, a pure integer linear programming model is developed to determine the optimal preventive maintenance and replacement schedules for a series of multi-component systems. In this model, we have considered a finite and discretized planning horizon in which three possible actions must be planned for each component in each system, namely maintenance, replacement, or do nothing. The objective is to minimize the total cost of projects by grouping maintenance and replacement operations. Because of the complexity of the model, several heuristic methods are applied to tackle the problem. 5 - Non Monotone Submodular Knapsacks And Applications Avinash Bhardwaj, Postdoctoral Fellow, Georgia Institute of Technology, Room 336, 755 Ferst Drive, NW, Atlanta, GA, 30332, United States, abhardwaj@gatech.edu, Alper Atamturk We study the facial structure of the convex hull of the level sets of a given submodular set function. In particular we derive valid inequalities and their extensions for the general lower level sets of submodular set functions, and propose the facet defining conditions for the same. We relax the monotonicity assumptions on the underlying set function and thus offering a generalization to earlier studies on this subject matter. We derive the appropriate valid inequalities and their extensions from the aggregation of the linear knapsack inequalities corresponding to the extended polymatroid of the set function in context. 1 - Generalizations And Applications Of The Multiperiod Assigment Problem Joseph B Mazzola, Professor and Endowed Chair, Cleveland State University, 1860 East 18th Street, BU 530, Cleveland, OH, 44115, United States, j.b.mazzola@csuohio.edu The Multiperiod Assignment Problem (MultiAP) involves the cost-minimizing assignment of a set of tasks to a set of agents within each period of a finite planning horizon when, in addition, there are transition costs associated with changing agent-task assignments from one period to the next. We review the literature on MultiAP and consider generalizations of the MultiAP including, for example, a model in which task learning occurs when an agent is able to work repeatedly on the same task. We also discuss applications of MultiAP.

5 - An Experimental Study Of Customer’s Risky And Egalitarian Behaviors In a Clearance Sales Problem Junlin Chen, Associate Professor, Central University of Finance and Economics, 39 South College Road, Haidian District, Beijing, 100081, China, chenjunlin@cufe.edu.cn, Yingshuai Zhao We consider a monopolist set prices in a two-period selling season such that low- value customers postpone the purchase for a lower price but subject to rationing risk, whereas high-value customers buy regularly. Traditionally, the optimal regular price is usually set with making high-value customers indifferent between buying early and late, and then basically all high-value customers are assumed to buy regularly. By conducting laboratory experiments, we provide evidence against this basic assumption. We demonstrate that the behavior of subjects can be explained by risky and egalitarian behaviors. We also find evidence about irrational waiting and myopic buying strategic customers. TD76 Legends D- Omni Decision Analysis II Contributed Session Chair: Nikita Korolko, PhD Candidate, Massachusetts Institute of Technology, 1 Amherst st, E40-106 ORC, Cambridge, MA, 2139, United States, korolko@mit.edu 1 - Impact Of Service Payment On Product And Service Supply Chain Considering Time Value Jiayuan Liu, Tsinghua University, Tsinghua University, 14 Zijing Department, Beijing, 100084, China, liujiayuan46@163.com, Wanshan Zhu We investigate a supply chain consisting of a service provider (e.g. AT&T) and a product maker (e.g. Apple), where the payment for service is in installments through a contract time frame. By modeling the installment payment and time value and solving the equilibrium strategies of pricing and inventory decisions, we analyze the impact of the service payment on the structure of the supply chain. 2 - Almost Stochastic Dominance When Utility Is Action-dependent Chunling Luo, National University of Singapore, Singapore, Singapore, c_luo@u.nus.edu, Chin Hon Tan Current stochastic dominance rules assume that utility function is identical across all actions. This assumption makes stochastic dominance rules not applicable under some practical settings. To help reveal decision makers’ preferences under these settings, we generalize almost stochastic dominance by allowing utility functions to differ among actions. 3 - Decision Analysis For Locating Partial Building Renovations Regarding Adaptive Reuse Kristopher Harbin, Doctoral Candidate, The University of Alabama, Tuscaloosa, AL, 35487-0205, United States, kbharbin@ua.edu When considering a building renovation for an adaptive reuse there are numerous building attributes and systems that should be considered. These building attributes should be compared to the proposed reuse and any alterations needed should be noted. The impact of these alterations should be noted and assigned an appropriate weight reflecting the level of impact. This is done for multiple areas of the building which helps ensures a complete listing of the renovation options are seen. 4 - Covariate-adaptive Optimization In Online Clinical Trials Nikita Korolko, PhD Candidate, Massachusetts Institute of Technology, 1 Amherst st, E40-106 ORC, Cambridge, MA, 02139, United States, korolko@mit.edu, Dimitris Bertsimas, Alexander M Weinstein Pharmaceutical companies spend tens of billions of dollars each year to operate clinical trials needed for the approval of new drugs. We present an online allocation algorithm for clinical trials that leverages robust mixed-integer optimization. In simulated experiments involving both single and multiple controlled covariates, our method reduces the number of subjects needed to achieve a desired level of statistical power by at least 35% relative to state-of-the- art allocation algorithms. Correspondingly, we expect that our computationally tractable approach could significantly reduce both the duration and operating costs of a clinical trial.

359