2016 INFORMS Annual Meeting Program

TC40

INFORMS Nashville – 2016

TC38 206A-MCC Service Science II Contributed Session Chair: Do-Hyeon Ryu, Pohang University of Science and Technology, Pohang, Kyungbuk, Korea, Republic of, dhryu@postech.ac.kr 1 - Financial Valuation Of Wellness Centered Operations Min Kyung Lee, Clemson University, 100 Sirrine Hall, Clemson, SC, 29634, United States, minl@g.clemson.edu, Aleda Roth, Rohit Verma, Bernardo F. Quiroga The importance of individual health and their wellbeing sheds light on the evolution of building design to achieve human sustainability. This study contributes the unstudied area of the financial and market impact of hotel guest rooms with wellness inspired features by comparing the Revenue Per Available Room (RevPAR) and customer satisfaction score. 2 - Extended Warranty Information Availability Formats: Impact On Consumer Purchase Decisions Paul R Messinger, University of Alberta, Faculty of Business, 3-20e Faculty Of Business Bldg, Edmonton, AB, T6G 2R6, Canada, paul.messinger@ualberta.ca, Moein Khanlari Larimi Information about extended service contracts, mainly price, is generally offered to buyers subsequent to their product purchase decision during the checkout. However, extended service contract (ESC) information has also been made available alongside product attribute information. In this research, we ask whether and how the mere availability of ESC information during the product choice phase might affect consumers’ product and ESC purchase decisions. To answer this question, we pit the simultaneous vs. delayed ESC information availability strategies against one another in choice experiments. 3 - Technology In Service – From Dumb To Thinking To Feeling Ming-Hui Huang, Distinguished Professor, National Taiwan University, 1 Sec 4, Roosevelt Road, Taipei, 10617, Taiwan, huangmh@ntu.edu.tw In this paper a three-generational technology evolution (from automated to thinking to feeling technology) is presented and the way they can be used in service is illustrated. Automated technologies are mainly developed for productivity, which achieves greater output with less input by standardization. Thinking technologies are designed to handle cognition-based personalization for customer satisfaction. Feeling technologies are to handle emotion-based personalization that enriches interactions. Given the multiplicity of technology, we should explore how to use the right technology for the right purpose in the right context by the right employees for the right customers. 4 - Development Of An Online To Offline Service Blueprint Do-Hyeon Ryu, Pohang University of Science and Technology, Pohang, Kyungbuk, Korea, Republic of, dhryu@postech.ac.kr, Chie-Hyeon Lim, Kwang-Jae Kim The online to offline(O2O) service is to find and attract users online and direct them to offline stores. Examples include Uber, Zipcar, Groupon, and so on. Although the term, O2O, is frequently used in academia and industries, research on systematic methods for developing O2O services has been scarce. This research aims to develop a new type of service blueprint specially designed for O2O services. This blueprint is expected to help O2O service providers visualize their services from the customer perspective. TC39 207A-MCC Dynamic Learning Applications Sponsored: Applied Probability Sponsored Session Chair: N. Bora Keskin, Duke University, Durham, Durham, NC, 27708, United States, bora.keskin@duke.edu 1 - Dynamic Pricing In Unknown Environments With Memory Abbas Kazerouni, Stanford University, Stanford, CA, 94305, United States, abbask@stanford.edu, Benjamin Van Roy We consider the problem of dynamic pricing in an unknown environment where the demand at any time is governed by the prices at that time as well as the previous time steps. The delayed consequences of the prices introduce new challenges to the problem and require more sophisticated pricing strategies. To deal with this problem, we propose a pricing strategy based on Reinforcement Learning techniques and derive bounds on its performance. We show the efficiency of the proposed strategy by comparing its performance against other strategies as well as the lower bound through some examples.

2 - Online Active Linear Regression Carlos Riquelme, Stanford University, rikel@stanford.edu, Ramesh Johari, Baosen Zhang We study the problem of online active learning to collect data for regression modeling; a decision maker that faces a limited experimentation budget but must efficiently learn an underlying linear model. Our main contribution is a novel threshold-based algorithm for selection of most informative observations; we characterize its performance and fundamental lower bounds. We extend the algorithm and its guarantees to sparse linear regression in high-dimensional settings. Simulations show significant benefits over random sampling in several real-world datasets that exhibit high nonlinearity and high dimensionality - strongly reducing the mean and variance of the squared error. 3 - Learning Preferences With Side-information: Near Optimal Recovery Of Tensors Andrew A Li, MIT, Cambridge, MA, United States, aali@mit.edu, Vivek Farias Many recent problems of great interest in e-commerce can be cast as large-scale problems of tensor recovery in three dimensions. Thus motivated, we study the problem of recovering ‘simple,’ 3D tensors from their noisy observations. We provide an efficient algorithm to recover structurally simple tensors given noisy (or missing) observations of their entries; our version of simplicity subsumes low- rank tensors for various definitions of tensor rank. Our algorithm is practical for large datasets and provides a significant performance improvement over incumbent approaches to Tensor completion. Further, we show theoretical recovery guarantees that under certain assumptions are order optimal. 4 - On Incomplete Learning And Certainty-equivalence Control N. Bora Keskin, Duke University, Durham, NC, United States, bora.keskin@duke.edu, Assaf Zeevi Motivated by dynamic pricing applications, we consider a dynamic control-and- estimation problem. The decision-maker sequentially chooses controls and observes responses that depend on both the chosen controls and an unknown parameter. The decision-maker uses a certainty-equivalence policy, and we characterize the asymptotic accuracy performance of this policy. TC40 207B-MCC Applied Probability and Machine Learning I Sponsored: Applied Probability Sponsored Session Chair: Guy Bresler, Massachusetts Institute of Technology, 32 Vassar St, 32-D672, Cambridge, MA, 02139, United States, guy@mit.edu 1 - Controlling Bias From Data Exploration Using Information Theory Daniel Russo, Northwestern University, Evanston, IL, United States, dan.joseph.russo@gmail.com, James Zou Modern data is messy and high-dimensional, and it is often not clear a priori which questions to ask. Instead, the analyst typically needs to use the data to search for interesting analyses to perform and hypotheses to test. It’s widely recognized that this process, even if well-intentioned, can lead to biases and false discoveries, contributing to the reproducibility crisis in science. We propose a general information-theoretic framework to quantify and provably bound the bias of an arbitrary adaptive analysis process. We prove that our bound is tight in natural models, and then use it to give rigorous insights into when common procedures do or do not lead to substantially biased estimation. 2 - K-nearest Neighbor Methods For Information Estimation Sewoong Oh, University of Illinois, Urbana, IL, United States, swoh@illinois.edu Estimators of information theoretic measures such as entropy and mutual information from samples are a basic workhorse for many downstream applications in modern data science. State of the art approaches have been either geometric (nearest neighbor (NN) based) or kernel based. In this paper we combine both these approaches to design new estimators of entropy and mutual information. Our estimator uses bandwidth choice of fixed k-NN distances; such a choice is both data dependent and linearly vanishing in the sample size and necessitates a bias cancellation term that is universal and independent of the underlying distribution. 3 - Semidefinite Programming Relaxations For Exact Recovery Of Hidden Communities Jiaming Xu, Purdue University, West Lafayette, IN, United States, xu972@purdue.edu, Bruce Hajek, Yihong Wu We study a semidefinite programming (SDP) relaxation of the maximum likelihood for exactly recovering hidden communities under the stochastic block model. It is shown that when the community size is large comparing to the network size, the SDP relaxation achieves the information-theoretic recovery threshold with sharp constants; when the community size is small, the SDP becomes strictly suboptimal comparing to the maximum likelihood estimator.

315