2016 INFORMS Annual Meeting Program

SB33

INFORMS Nashville – 2016

2 - A Model For Scheduling Practical Lessons And Selecting Teaching Assistants At Universities Cristian D Palma, Assistant Professor, Universidad del Desarrollo, Avda. Sanhueza 1750, Pedro de Valdivia, Concepcion, 4040418, Chile, cristianpalma@ingenieros.udd.cl, Pablo Gonzalez, Pamela Riffo Most of the courses at universities includes practical sessions taught by teaching assistants (TA), which are also students. These sessions are usually scheduled as part of the courses, so the day and time when they are taught are known when students register their courses. Since TAs have to attend their own courses, they apply for teaching only in courses that match their own schedules rather than courses they are good at. We propose a framework in which practical sessions are scheduled after course registration, and show a model that schedules practical sessions and simultaneously selects the TAs for each course. We discuss the advantages of using this approach and present results of its application. 3 - Scheduling Virtual Network Embeddings Frank Fischer, University of Kassel, Heinrich-Plett-Str. 40, Kassel, 34132, Germany, frank.fischer@mathematik.uni-kassel.de, Andreas Bley The virtual network embedding problem aims to embed several virtual network (VN) requests, each consisting of several node and connection services that require certain CPU, memory and bandwidth resources, into a shared physical substrate network in such a way that the resources available in the substrate network are not exceeded. We consider the dynamic version of this problem, where VN requests have time windows and durations specifying when and how long they should be embedded. We discuss several mixed integer and constraint programming approaches for this combined embedding and scheduling problem. Our computational results show that a combination of both techniques performs best. 4 - An Improved Genetic Algorithm For Job-shop Scheduling

3 - A Fully Exploratory Reinforcement Learning Algorithm For Solving Semi-Markov Decision Processes Angelo Encapera, Research Assistant, Missouri University of Science & Technology, Rolla, MO, 65409, United States, amet3b@mst.edu, Abhijit Gosavi We study the development of a fully exploratory Reinforcement Learning (RL) algorithm for solving Semi-Markov Decision Processes (SMDPs). Existing RL algorithms, such as R-SMART, for solving SMDPs require gradual decay of exploration. The latter adds a tuning parameter to the algorithm, and indeed its success depends on how the exploration-decay parameter is tuned. Our algorithm uses a “reflective” update that accompanies the main update, based on relative Q- Learning, to estimate the average reward without decaying the exploration. Our algorithm delivers encouraging empirical behavior. 4 - A Bounded Actor Critic Algorithm For Reinforcement Learning Actor-critic algorithms are amongst the oldest reinforcement learning algorithms that can be used to solve Markov decision processes (MDPs) via simulation. Unfortunately, the values of the “actor” in the classical version of this algorithm get unbounded in practice. In practice, the actor’s values are artificially constrained to obtain solutions. Boltzmann action selection is used for this algorithm in which a temperature is used, but the convergence of the algorithm is guaranteed only when the temperature equals 1. We propose a new actor-critic algorithm in which the actor’s values are bounded even when the temperature is set to 1. Our algorithm delivers encouraging numerical behavior. 204-MCC Data-Driven Decisions in Healthcare Sponsored: Manufacturing & Service Oper Mgmt, Healthcare Operations Sponsored Session Chair: Vishal Ahuja, Southern Methodist University, Dallas, TX, United States, vahuja@smu.edu 1 - Optimizing Cancer Prevention Strategies For BRCA1/2 Mutation Carriers Elisa Frances Long, Assistant Professor, UCLA Anderson School of Management, Los Angeles, CA, United States, elisa.long@anderson.ucla.edu, Eike Nohdurft, Stefan Spinler BRCA1/2 mutation carriers face significantly elevated lifetime risks for breast and ovarian cancer. Prophylactic surgery (bilateral mastectomy, oophorectomy, or both) can reduce the risk of cancer but may impact health utility. We developed a Markov decision process model to determine the optimal age to undergo surgery to maximize quality-adjusted life years or survival. Key state variables include age, mutation type, breast cancer stage and sub-type, ovarian cancer stage, and preventive surgery status. We solve the model with linear programming and compute the optimal policy for different parameter settings reflecting varying mutation carrier’s preferences. 2 - Predictive Models For Making Patient Screening Decisions Michael Hahsler, Southern Methodist University, Dallas, TX, United States, mhahsler@lyle.smu.edu, Vishal Ahuja, Michael Bowen A critical dilemma that clinicians face is when and how often to screen patients who may suffer from a disease. The stakes are heightened in the case of chronic diseases that impose a substantial cost burden. We investigate the use of predictive modeling to develop optimal screening conditions and assist with clinical decision-making. We use electronic health data from a major U.S. Hospital and apply our models in the context of screening patients for type-2 diabetes, one of the most prevalent diseases in the U.S. and worldwide. 3 - Belief Perseverance And Experience Bradley R Staats, University of North Carolina at Chapel Hill, bstaats@unc.edu, Diwas S KC, Francesca Gino Many models in operations management involve dynamic decision making that assumes optimal updating in response to information revelation. However, behavioral theory suggests that rather than updating their beliefs, individuals may persevere in their prior beliefs. We examine how individuals’ prior experiences and the experiences of those around them alter their belief perseverance in operational decisions after the revelation of negative news using a field experiment and two lab experiments. SB34 Ryan Lawhead, Research Assistant to Dr. Gosavi, Missouri University of Science and Technology, 223 Engineering Management Building, Rolla, MO, 65409, United States, rjlm97@mst.edu, Abhijit Gosavi, Susan Murray

Mauricio G. C. Resende, Principal Research Scientist, Amazon.com, Inc., 333 Boren Ave N, Seattle, WA, 98109, United States, resendem@amazon.com José F. Gonçalves

We present a local search, based on a new neighborhood for the job-shop scheduling problem, and its application within a biased random-key genetic algorithm. Schedules are constructed by decoding the chromosome supplied by the genetic algorithm with a procedure that generates active schedules. After an initial schedule is obtained, a local search heuristic, based on an extension of the 1956 graphical method of Akers, is applied to improve the solution. The new heuristic is tested on a set of 205 standard instances taken from the job-shop scheduling literature and compared with results obtained by other approaches. The new algorithm improved the best-known solution values for 57 instances.

SB33 203B-MCC Simulation and Optimization II Contributed Session

Chair: Ryan Lawhead, Research Assistant to Dr. Gosavi, Missouri University of Science and Technology, 223 Engineering Management Building, Rolla, MO, 65409, United States, rjlm97@mst.edu 1 - A Study on The Operations Analysis Using Big Data Developed by Simulations Hongseon Park, University of Central Florida, Orlando, FL, United States, gauss1211@naver.com, Won Il Jung, Yong Bok Lee, Gakgyu Kim, Gene Lee, Rabelo Luis The operations analysis is analyzed under simulation circumstances. A lot of assumptions and limited environments are included and, thus, the results have a bias. To overcome this problem, we proposed a new methodology for the operations analysis using Big Data developed by Virtual and Constructive (VC) simulations. The VC simulations can produce a large volume and variety of data since many variables are used for analyzing the operations close to reality by using 6 Degrees of Freedom models. More than terabytes of data including structured and unstructured types are applied to the current techniques of Big Data. Then we will build the probability map and index to help the commanders make decisions. 2 - Cloud Based Collaborative Information Sharing In Supply Chains Cigdem Kochan, Assistant Professor, Ohio Northern University, Dicke College of Business Administration, 525 S Main St, Ada, OH, 45810, United States, cigdem.kochan@gmail.com, David R. Nowicki This research develops system dynamics models to simulate the effect of cloud based collaborative information sharing in a supply chain. The results suggest that the use of the cloud based information sharing in supply chain reduces inventory levels, reduces actual lead time through demand and inventory visibility, and reduces delivery delays while increasing overall performance of the supply chain.

54