2016 INFORMS Annual Meeting Program

WE71

INFORMS Nashville – 2016

2 - Correlation Based Adaptive Sampling Strategy For Online Monitoring Of Correlated High Dimensional Data Streams Mohammad Nabhan, Georgia Institute of Technology, m.nabhan33@gmail.com, Jianjun Shi Effective process control of High dimensional data with embedded spatial structure has been an arising challenge, due to the inability of classical techniques to detect changes in such processes. This article proposes an adaptive sampling technique that achieves better change detection power by identifying and exploiting the hidden spatial structure. The adaptive nature of the proposed method allows for effective monitoring with partial observations. Simulation studies are performed to validate the efficacy of the proposed monitoring scheme. This is followed by real data case studies to evaluate the performance under practical scenarios. 3 - An Effective Online Data Monitoring And Saving Strategy For Large-scale Climate Simulations Xiaochen Xian, University of Wisconsin - Madison, Madison, WI, xxian@wisc.edu, Kaibo Liu Large-scale climate simulation models have been developed and widely used to generate historical data and study future climate. This long-duration simulation process creates huge amount of data; however, how to effectively monitor and record the climate changes still remains to be resolved. To address this issue, we propose an effective online data monitoring and saving strategy over the temporal and spatial domains with the consideration of practical storage and memory capacity constraints. Specifically, our proposed method is able to intelligently select and record the most informative extreme values in the raw data in the context of better monitoring climate changes. 4 - A Wavelet-based Penalized Mixed-effects Model For Multichannel Profile Detection Of In-line Raman Spectroscopy Xiaowei Yue, Georgia Institute of Technology, xwy@gatech.edu, Hao Yan, Jin Gyu Park, Richard Liang, Jianjun Shi Modeling of high-dimension nonlinear profiles is an important and challenging topic in statistical process control. Conventional mixed effect model has limitations in solving the multichannel profile detection in nanomanufacturing. A wavelet-based penalized mixed-effects model (WPMM) is proposed to exploits a regularized high-dimensional regression with linear constraints to decompose profiles into four parts: fixed effect, normal random effect, defective random effect, and noise. An accelerated proximal gradient algorithm is developed to efficiently estimate parameters. Case study shows that the WPMM can realize a better detection power and shorter computation time. WE69 Old Hickory- Omni Dynamic Programming /Control Contributed Session Chair: Jefferson Huang, Postdoctoral Associate, Cornell University, Ithaca, NY, United States, jh2543@cornell.edu 1 - Segmentation Of Anatomical Structures Using Dynamic Identification And Classification Of Edges (dice) Model In Medical Images Maduka M. Balasooriya, Teaching Assistant, Southern Illinois University, Edwardsville, Edwardsville, IL, 62026, United States, mbalaso@siue.edu, Sinan Onal, Xin W Chen Segmentation of anatomical structures using medical images such as MRI, CT scan, and digital fundus image is still ongoing research subject. We developed the dynamic identification and classification of edges (DICE) model that aims to automatically identify edges of a contour of an anatomical structure without any intervention from domain experts. The DICE model includes three sequential but intertwined steps: (a) identifying potential edge points of a contour using moving range control charts; (b) extrapolating additional edge points of a contour through noise reduction; and (c) classifying points into different edges using neighborhood gradient search. 2 - Military Applications Of Approximate Dynamic Programming: Optimizing Helicopter Lift Operations James Grymes, Instructor, United States Military Academy, 646 Swift Rd, West Point, NY, 10996, United States, james.grymes@usma.edu Military commanders rely on helicopter lift assets to serve as force multipliers by moving military personnel and equipment around the battlefield. Inefficiency is a byproduct of uncertainty inherent in military operations which can lead to mission delays and possible failures. This research is designed to explore approximate dynamic programming as a tool for assisting air movement planners. We model Army helicopter lift operations as a Markov Decision Process and learn the value of decisions through machine learning. The algorithm returns an approximating value function for scheduling air lift routes in order to enhance combat power.

3 - Value Function Discovery In Markov Decision Processes Sandjai Bhulai, Vrije Universiteit Amsterdam, Faculty of Sciences,

De Boelelaan 1081a, Amsterdam, 1081 HV, Netherlands, s.bhulai@vu.nl, Martijn Onderwater, Robert van der Mei

We introduce a novel method for discovery of value functions for Markov Decision Processes (MDPs). This method is based on ideas from the evolutionary algorithm field. Its key feature is that it discovers descriptions of value functions that are algebraic in nature. This feature is unique, because the descriptions include the model parameters of the MDP. The algebraic expression can be used in several scenarios, e.g., conversion to a policy, control of systems with time- varying parameters. We illustrate its application on an example MDP. 4 - Making College Admission Offers: A Dynamic Programming Approach Subhamoy Ganguly, Indian Institute of Management Udaipur, IIM Udaipur, MLSU Campus, Udaipur, 313001, India, subhamoy.ganguly@iimu.ac.in, Michele Samorani, Ranojoy Basu, Viswanathan Nagarajan College admissions problem refers to the problem where each college seeks to admit the best possible class from a pool of applicants and most applicants apply to multiple colleges, hoping to enroll in one of their most preferred colleges. Most of the extant literature models this problem in the context of a two-sided matching market. However, admissions offices often cannot use these approaches due to practical limitations. We develop a dynamic programming model that could help colleges make optimal decisions of offering admission to secure the best possible class while filling all seats. 5 - Reductions Of Undiscounted Markov Decision Processes And Stochastic Games To Discounted Ones Jefferson Huang, Postdoctoral Associate, Cornell University, Ithaca, NY, United States, jh2543@cornell.edu, Eugene A Feinberg We provide conditions under which certain total and average cost Markov decision processes (MDPs), with possibly uncountable state and action spaces, can be reduced to discounted ones. These reductions are used to obtain complexity estimates for computing an optimal policy for finite MDPs and for computing a nearly optimal policy for infinite MDPs. We also provide analogous reductions for zero-sum stochastic games with possibly uncountable state and action spaces, and show how they can be used to obtain results on the existence of the value and optimal strategies, as well as results on robust MDPs. Game Theory V Contributed Session Chair: Edward Cook, Senior Vice President, Capital One, 4 Bisley Court, Henrico, VA, 23238, United States, Ed.cook2000@gmail.com 1 - Bayesian Opponent Exploitation In Imperfect-information Games Sam Ganzfried, Ganzfried Research, 55 West 26th Street #36E, New York, NY, 10010, United States, sam.ganzfried@gmail.com For all game classes one can potentially do better than following a static Nash equilibrium strategy by learning to exploit perceived weaknesses of opponents. An exact efficient algorithm is known for best-responding to the opponent’s posterior distribution assuming a Dirichlet prior with multinomial sampling in normal-form games; however, for imperfect-information games the best known algorithm is a sampling algorithm approximating an infinite integral without theoretical guarantees. The main result is the first exact algorithm for accomplishing this in certain imperfect-information games. We also present an algorithm for the natural setting where the prior is uniform over a polyhedron. 2 - Strategic Decentralization: Implications For Equity And Equality Omkar D Palsule-Desai, Faculty, Indian Institute of Management Indore, Rau Pithampur Road, Wing C, Ground Floor, Indore, 453331, India, omkardpd@iimidr.ac.in We develop a noncooperative game theoretic model to examine network performance and stability related implications of allocation mechanisms that endogenously balance equity vis-à-vis equality, and hence, the degree of collusion among the network firms in a decentralized setting. We show that inefficiencies and instability of decentralization can be eliminated by incorporating an additional degree of freedom in the network formation game. Our model and the structural results are applicable to networks such as producers’ cooperatives, industrial clusters, joint production and research facilities, etc., wherein the conflicts of equity-equality and degree of collusion are predominant. WE71 Electric- Omni

505