Informs Annual Meeting 2017

MC02

INFORMS Houston – 2017

MC02

MC03A Grand Ballroom A Data-Driven Revenue Management Sponsored: Applied Probability Sponsored Session Chair: Gah-Yi Ban, London Business School, London Business School, London, NW1 4SA, United Kingdom, gban@london.edu 1 - A Nonparametric Self-adjusting Control for Multi-product Pricing with Unknown Demand Function and Finite Resource Capacity George Chen, University of Texas at Dallas, Richardson, TX, United States, georgeqc@umich.edu, Stefanus Jasin, Izak Duenyas We study a multi-period network revenue management problem where the underlying demand function is unknown (in the nonparametric sense) to the seller who uses dynamic pricing to minimize expected revenue loss. It is known that the asymptotic revenue loss of any feasible pricing policy is O(k^{1/2}) (k indicates the size of the problem), but there is a considerable gap between this theoretical lower bound and the performance bound of all existing heuristics. We propose a Nonparametric Self-adjusting Control and show that it guarantees a revenue loss of O(k^{1/2+epsilon} log k) for any arbitrarily small epsilon>0, provided that the underlying demand function is sufficiently smooth. 2 - Assortment Optimization of Vertically Differentiated Products for Quality-conscious Customers Arnoud Victor den Boer, University of Amsterdam, Postbus 94248, Room number F3.33, Amsterdam, 1090 GE, Netherlands, boer@uva.nl, N. Bora Keskin A product is available in a continuum of different quality levels. Potential customers are differentiated according to their minimum required quality level, which is random and unknown to the seller. We discuss the full-information assortment optimization problem of determining which subset of products maximize the seller’s expected revenue. This is an optimization problem over measurable subsets of the positive real line. In addition we discuss a version of the problem with incomplete information, where the distribution of required quality is learned from data. 3 - Personalized Dynamic Pricing with Machine Learning N. Bora Keskin, Duke University, Fuqua School of Business, 100 Fuqua Drive, Durham, NC, 27708-0120, United States, bora.keskin@duke.edu, Gah-Yi Ban Motivated by online retail applications, we consider a seller who offers personalized prices to individual customers. The seller initially does not know the impact of individual customer characteristics on demand, but can learn about this relationship via sales observations. We construct and analyze near-optimal policies that balance the learn-and-earn tradeoff in this setting. 4 - Statistics for Cross-sectional Surveys: Inferring Total Eventual Time in Current State using Only Elapsed Time-to-Date Richard C.Larson, Massachusetts Institute of Technology, E40-233, Cambridge, MA, 02139, United States, rclarson@mit.edu, Louis Cammarata A recent paper by the 2nd author focused on cross-sectional surveys in which a question asked is, “How long have you been in this temporary state?” The result is a derivation of distribution of total duration in the temporary state based only on answers to that question. The method utilizes properties of longevity bias. Here, using nonparametric estimation methods we investigate the statistical accuracy of the method as a function of survey sample size and properties of underlying distributions. We report Monte Carlo simulations using discrete and continuous distributions and provide confidence interval results.

310B Healthcare Decision Analysis I Sponsored: Decision Analysis Sponsored Session Chair: L Robin Keller, University of California-Irvine, Irvine, CA, 92697-3125, United States, LRKeller@uci.edu 1 - A Preliminary Decision Model for Brest Cancer Screening Cristina del Campo, PhD, Universidad Complutense de Madrid, Madrid, Spain, campocc@ccee.ucm.es, Manuel Luque, Cristina Antón Rodríguez, Manuela Parras, Francisco Javier Diez Breast cancer is one of the leading causes of cancer-related deaths in women all over the world. It is therefore imperative to determine an optimal screening pattern for breast cancer that trades off between early cancer detection and avoiding unnecessary procedures for healthy women. We present a preliminary decision model, implemented as a Markov influence diagram, for evaluating the cost-effectiveness of the optimal breast screening policy for an individual patient. The model takes into account major risk factors such as a positive family history or age and uses clinical data to recommend the best screening policy. 2 - Markov Cost-effectiveness Analysis for Cancer Treatment Jiaru Bai, Binghamton University, School of Management, Binghamton, NY, 13902, United States, jiarub@uci.edu, L. Robin Keller We present a way to build a Markov decision tree to model cancer progression and cost-effectiveness analysis for two or more cancer treatments. We propose several problems researchers can encounter in this kind of research and provide possible solutions. 3 - The Effects of Assigning Buddies in Online Health Communities Ali Esmaeeli, University of California, Irvine, 1914 Verano Place, Irvine, CA, 92617, United States, esmaeeli@uci.edu, Cornelia Pechmann We look at the effects of assigning buddies on the relationships between them and its result on abstinence in online smoking cessation groups. We show that directed links to active buddies are stronger than directed links to active non- buddies. We also show that for inactive buddies, the directed relations to them is not looser than the directed relations to inactive non-buddies. Then we show that those participants who have stronger directed relations with their pairs have better performance in quitting smoking. 310C Optimization Methods for Supervised Machine Learning: From Linear Models to Deep Learning Invited: Tutorial Invited Session Chair: Jiming Peng, University of Houston, Houston, TX, 77204, United States, jopeng@Central.uh.edu Co-Chair: Rajan Batta, University at Buffalo (SUNY), 410 Bell Hall, Buffalo, NY, 14260, United States, batta@buffalo.edu 1 - Optimization Methods for Supervised Machine Learning: From Linear Models to Deep Learning Katya Scheinberg, Lehigh University, Bethlehem, PA, United States, katyas@lehigh.edu, Frank E.Curtis The goal of this tutorial is to introduce key models, algorithms, and open questions related to the use of optimization methods for solving problems arising in machine learning. It is written with an INFORMS audience in mind, specifically those readers who are familiar with the basics of optimization algorithms, but less familiar with machine learning. We begin by deriving a formulation of a supervised learning problem and show how it leads to various optimization problems, depending on the context and underlying assumptions. We then discuss some of the distinctive features of these optimization problems, focusing on the examples of logistic regression and the training of deep neural networks. The latter half of the tutorial focuses on optimization algorithms, first for convex logistic regression, for which we discuss the use of first-order methods, the stochastic gradient method, variance reducing stochastic methods, and second- order methods. Finally, we discuss how these approaches can be employed to the training of deep neural networks, emphasizing the difficulties that arise from the complex, nonconvex structure of these models. MC03

202