Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

SA36

3 - Optimal Gate Assignment Under the Consideration of Airport Retailing

n SA37 North Bldg 225A Information and Dynamic Decision Making Sponsored: Applied Probability Sponsored Session Chair: Kuang Xu, Stanford Graduate School of Business, Stanford, CA 1 - Privacy and Incentives in Dynamic Games Rachel Cummings, Georgia Tech, 755 Ferst Drive NW, Atlanta, GA, 30332-0205, United States When strategic agents play repeated games against the same opponents, they may be concerned about information leakage across rounds. Examples include repeated posted pricing and repeated auctions, where a seller can learn a buyer’s value to charge higher prices in future rounds. In this talk, we will explore differential privacy as a tool for mitigating these incentive issues and bringing behavior in each round closer to that of the stage game. 2 - Satisficing in Time-sensitive Bandit Learning Daniel Russo, Columbia University, New York, NY, United States, Benjamin Van Roy Much of the recent literature on bandit learning focuses on algorithms that aim to converge on an optimal action. One shortcoming is that this orientation does not account for time sensitivity, which can play a crucial role when learning an optimal action requires much more information than near-optimal ones. We consider instead learning a satisficing action, which is near-optimal while requiring less information, and propose satisficing Thompson sampling, an algorithm that serves this purpose. We establish a general bound on expected discounted regret and discuss relations with the theory of rate distortion, which offers guidance on the selection of satisficing actions. 3 - Capacity of Resource Allocation Under Information Loss Yuan Zhong, Booth School of Business, University of Chicago, Kuang Xu We consider the problem of dynamic resource allocation under information loss, in general stochastic processing systems. Information loss is modeled as an information channel. We characterize the capacity region under various channel models, and analyze the role of memory in policy design. n SA38 North Bldg 225B Applied Probability in Machine Learning and Optimization II Sponsored: Applied Probability Sponsored Session Chair: Chang-Han Rhee, Centrum Wiskunde and Informatica, Amsterdam, 1018 MZ, Netherlands 1 - Optimal Sampling Algorithms for Empirical Risk Minimization We present a large deviations analysis of sample average approximation (SAA), specifically focused on empirical risk minimization (ERM) problems relevant to machine learning and M-estimation. We focus on situations where it is feasible to sample independently and without replacement at predefined grid-points in the design space. The resulting LD rate function is then used to extract a computational-effort allocation algorithm, that efficiently trades-off computational-effort against statistical accuracy. We prove the asymptotic consistency and efficiency of the resulting estimators. 2 - Infinitely Many Armed Bandit Problem with Learned Rewards Yang Cao, University of Southern California, Los Angeles, CA, United States There are infinitely many arms with arm i, i =1 having value vi where vi are the values of i.i.d non-negative random variables following a known distribution F. The values vi are initially unknown to the player, and will be learned once being played. In each game the player can choose either an observed arm or a new arm to play, and will receive a reward equal to the chosen arm’s value. The objective is to find a policy that maximizes the expected sum of the rewards in N games. We show that the optimal policy is a threshold stopping policy when N is deterministic or a random variable having increasing failure rate. We also propose heuristic policies for the case when N has a general distribution. Harsha Honnappa, Purdue University, 315 N. Grant Street, West Lafayette, IN, 47906, United States, Raghu Pasupathy, Prateek Jaiswal

Tulay Flamand, Colorado School of Mines, Division of Economics and Business, Engineering Hall 816 15th Street, Room 313, Golden, CO, 80401, United States, Heng Chen We address a gate assignment problem and its impact on airport retail sales. We develop a mixed-integer programming model that assigns flights to gates in a way that optimizes several objectives, as well as the retail sales. A column generation methodology is used to solve this challenging problem and the results of the computational study are reported. 4 - A Variable Neighborhood Search Approach for the Flight-to-gate Reassignment Problem Moschoula Pternea, University of Maryland, College Park, MD, 20740, United States, Ali Haghani The reallocation of flights to gates in case of schedule disruptions is a key airport recovery operation. Especially in hub airports, disruptions affect connecting passengers by causing failed connections or lost baggage. At the same time, introducing passenger transfers in mathematical programming models makes the problem intractable. In this context, we develop a Variable Neighborhood Search approach to generate near-optimal solutions with the objective to minimize additional schedule disturbances, as well as passenger and baggage misconnections. Our preliminary experiments indicate that our metaheuristic approach shows promising results in terms of both solution quality and time. n SA36 North Bldg 224B Airline Recovery Operations and Route Planning Sponsored: Aviation Applications Sponsored Session Chair: Vikrant Vaze, Dartmouth College, Hanover, NH, 03755, United States 1 - An Integrated Model for Airline Operations Recovery: Emphasizing Passenger Compensation Impacts Luis Cadarso, Rey Juan Carlos University, Camino del Molino s/n, Fuenlabrada, 28943, Spain, Vikrant Vaze The European flight delay compensation regulation (EC) No 261/2004 establishes common rules on compensation to passengers in the event of disruptions. We develop an integrated approach that recovers airline timetable, fleet assignment, aircraft routings, and passenger itineraries capturing the impacts of airlines’ decisions on passenger compensation. We evaluate scenarios involving disruptions, and optimize recovery decisions to maximize profits by modeling passenger no-shows after disruptions. 2 - Fleet Operations - Combining Tail Assignment and Schedule Recovery Andreas Westerlund, Jeppesen, Odinsgatan 9, Gothenburg, 411 03, Sweden, Mattias Gronkvist Tail assignment is the process of planning the assignment of aircraft to flights from the day of operations and a few days, or weeks, into the future. Schedule recovery is the process of deciding how to operate the aircraft at an airline when operational disruptions happen. In this presentation we will discuss how a combined tail assignment system and schedule recovery system gives benefits compared to using separate systems. We will also show how integration with crew tracking and flight planning can give additional benefits. Computational results on real-world data will be presented. 3 - A Two Stage Route Planning Algorithm for Light Aircraft Transportation Systems Shadi Sharif Azadeh, Assistant Professor, Erasmus University Rotterdam, Burgemeester Oudlaan 50, Rotterdam, 3062 PA, Netherlands This paper presents a route planning algorithm for light aircraft operating system under visual flight rules. The problem aims to find a minimum duration, collision free route in three spatial dimensions. The calculated route takes into account the aircraft kinodynamic characteristics and its interaction with external wind. A data processing approach is presented to recast the flying environment as a series of polyhedrons based on which a mixed integer linear model is formulated. A two stage route planning heuristic is developed to solve real life instances. Computational experiments depict the efficiency of this approach.

14