Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

SC37

2 - Asymptotic Analysis of Multiclass Queues with Random Order of Service Mohammadreza Aghajani, University of California San Diego, 9500 Gilman Drive # 0112, La Jolla, CA, 92093, United States Queueing models under Random order of Service (ROS) policy have been used to study molecular interactions of intracellular components in biology. However, these models often assume exponential distributions for processing and patience times, which is not realistic. We study a multiclass queueing model under ROS with reneging and generally distributed processing and patience times. We use measure-valued processes to describe the dynamic evolution of the network, and establish a fluid and diffusion approximations for this representation. These limits are characterized by deterministic and stochastic non-linear Partial Differential Equations, whose analysis requires new techniques. 3 - Ergodic Control of a Class of Jump Diffusions with Finite Levy Measures and Rough Kernels Yi Zheng, The Pennsylvania State University, State College, PA, United States, Ari Arapostathis, Luis Caffarelli, Guodong Pang We study the ergodic control problem for a class of jump diffusions in Rd, which are controlled through the drift with bounded controls. The Levy measure is finite, but has no particular structure. Unstable behavior is ædiscouraged’ by the running cost which satisfies a mild coercive hypothesis. We first study the problem as an optimization problem on the space of infinitesimal ergodic occupation measures, and derive the HamiltonùJacobiùBellman equation, including verification of optimality results, using only analytical arguments. We also examine the regularity of invariant measures. Then, we address the jump diffusion model, and obtain a complete characterization of optimality. 4 - Existence and Uniqueness of Obliquely Reflecting Diffusions in Cusps Cristina Costantini, University of Chieti-Pescara, Pescara, Italy, Thomas G. Kurtz In a 2009 paper, Kang, Kelly, Lee and Williams proposed a diffusion approximation for the workload process in a model for a network operating under a weighted a-fair bandwidth sharing policy. For a<1 the diffusion approximation cannot be proved because the diffusion state space presents cusp- like singularities and suitable uniqueness results are not available. We prove weak existence and uniqueness for a semimartingale reflecting diffusion in a 2- dimensional cusp, with a varying oblique direction of reflection on each side, under the only assumption that there exists a vector that has positive scalar product with the common tangent to the two sides and with the two directions of reflection at the tip. n SC39 North Bldg 226A Machine Learning and Applied Probability Sponsored: Applied Probability Sponsored Session Chair: Daniel Russo, Columbia University, New York, NY, 1 - Optimal Hardness of Questions in Static and Interactive Exams We consider the problem of designing a perfect exam, one that minimizes the mis-classification or mis-grading probability. We consider both static as well as interactive exams. Student’s probability of success in a question is modelled as an increasing function of her ability and decreasing function of question hardness. Ability itself may be a random variable. We use pure exploration Bandit framework to develop sample complexity lower bounds. We also develop algorithms whose computational effort matches the dominant term in the lower bounds. 2 - Distributional Robustness in Machine Learning: Statistical and Computational Guarantees Hongseok Namkoong, Stanford University, Stanford, CA, United States We study distributionally robust approaches to statistical learning, and provide finite-sample and asymptotic results characterizing the theoretical performance of the estimator. We also develop efficient solution methods for such approaches, and empirically verify that distributional robustness is a valuable approach in a number of machine learning applications where reliability is a key concern. Our approach learns the tails of the distribution, hedges against potential covariate shifts, promotes fairness across demographics, and provides robustness against adversarial attacks. Sandeep Juneja, Tata Institute of Fundamental Research, Homi Bhabha Road, Mumbai, Maharashtra, 400005, India, Achal Bassamboo, Assaf Zeevi

n SC37 North Bldg 225A Data, Services, and Queues Sponsored: Applied Probability Sponsored Session Chair: Rouba Ibrahim, University College London, London, WC1E 6BT, United Kingdom 1 - Linking Delay Announcements, Loss Aversion, and Abandonment: A Behavioral Perspective Eric M. Webb, Assistant Professor, Lindner College of Business, University of Cincinnati, Cincinnati, OH, United States, Qiuping Yu, Kurt M. Bretthauer Using field data from a call center, we study the behavioral determinants of customers’ queue abandonment decisions in the presence of delay announcements. Customers exhibit loss aversion with respect to time losses in queue, becoming much more likely to abandon if forced to wait longer than expected. We test the robustness of loss aversion across multiple announcements, customer class types, and functional forms of loss aversion. In addition, we study the effect of the queue experience on the time in service for those customers who do not abandon. 2 - The Impact of Economic Drivers on Gig Economy Workers: Structural Estimation Approach While gig economy firms benefit from increased labor flexibility, ensuring that their services appeal to independent providers poses a great challenge in planning and committing to a service capacity. We study how on-demand workers make labor decisions: when to work and for how long? Our project is in collaboration with a ride-hailing company with the goal to not only improve the way of predicting the number of active drivers, but also understand how to better recruit them, as a way to match supply and demand. We use a structural estimation approach to study how drivers respond to different incentives. 3 - Last Place Aversion in Queues Ryan Buell, Harvard Business School, Morgan Hall 429, Boston, MA, 02163, United States This paper investigates whether people exhibit last place aversion in queues and its implications for their experiences and behaviors in service environments. A combination of field and lab evidence reveals that waiting in last place diminishes wait satisfaction while doubling the probability of switching and quadrupling the probability of abandoning queues. This behavior is partially explained by the inability to make a downward social comparison; namely, when no one is behind a queuing individual, that person is less certain that continuing to wait is worthwhile. The results also demonstrate how queue transparency is an effective design lever for reducing last place aversion in queues. 4 - History Based Priority Policies Brett Hathaway, UNC Chapel Hill, 1800 Baity Hill Drive #310, Chapel Hill, NC, 27514, United States, Seyed Morteza Emadi, Vinayak V. Deshpande We study the behavior of callers in a banking call center. Using a latent class decision model, we estimate that callers differ in their abandonment and redialing behavior based on intrinsic heterogeneity in their behavior and their unique histories of waiting times and decisions with the call center. We introduce a class of policies where callers are prioritized based on our model’s predictions of their behavior, which depend on their history. We find that under history-based policies this call center could reduce average waiting times by 10 percent while increasing sales opportunities by 3 percent. n SC38 North Bldg 225B APS Session Reed Sponsored: Applied Probability Sponsored Session Chair: Joshua Reed, New York University, New York, NY, 10012, United States 1 - A Dirichlet Process Characterization of RBM in a Wedge Josh Reed, NYU, New York, NY, United States, Peter Lakner, Bert Zwart We prove that in the case of 1 < alpha < 2, RBM in a wedge is a Dirichlet process. Specifically, its unique Doob-Meyer type decomposition is given by Z=X+Y, where X is a two-dimensional Brownian motion and Y is a continuous process of zero energy. Furthermore, we show that for p > alpha, the strong p-variation of the sample paths of Y is finite on compact intervals, and, for 0 < p <= alpha, the strong p-variation of Y is infinite on [0,T] whenever Z has been started from the origin. We also show that (Z,Y) satisfies the extended Skorokhod problem for X. Gad Allon, University of Pennsylvania, 3730 Walnut Street, Philadelphia, PA, 19104, United States, Park Sinchaisri, Maxime Cohen

74