Friday, February 1

1174 Etcheverry Hall, 3:30 p.m. - 4:30 p.m.

Abstract: In today’s urban mobility marketplaces, both operational policies (e.g., matching, routing) and economic mechanisms (e.g., pricing, incentives) affect perceptions of Quality of Service (QoS) and users’ mobility choices. These, in turn, affect both operational objectives (e.g., utilization, vehicle-miles travelled) and economic objectives (e.g., profit, welfare). We study these complex interactions between behavior, economics, and operations in two settings. First, in commercial ridesharing, we study the problem of pricing exclusive vs. shared service to maximize profit, while incorporating QoS guarantees (motivated by recent field experiments that highlight the importance of proactively compensating users when they experience frustrations during service). We introduce a dynamic notion of QoS, called sequential individual rationality, that captures users’ behavior-driven responses to the sequence of utilities experienced during the successive stages of a shared ride, as subsequent users join the ride. The analysis of QoS-sensitive profit not only yields the optimal QoS-aware pricing policy, but also reveals key operational insights, such as an elegant spatial characterization of optimal “shareable regions”. Our framework is general and can be, for example, adapted to the peer-to-peer carpooling context, where the problem involves finding QoS-aware, fair cost sharing schemes. (This part of the talk is based on joint work with Theja Tulabandhula and Koyel Mukherjee.)


Next, we shift our focus from a commercial to a societal perspective, where we study the problem of designing a centralized market for urban mobility that increases social welfare, while ensuring participation from multiple stakeholders (such as the commuters, non-profit public transit agencies, and profit-sensitive on-demand service providers). We show that introducing multi-modal trips, as an operational lever, can soften the dependence of such mechanisms on external subsidies (mandated by fundamental impossibility results in mechanism design), by favorably exploiting structural properties of a city’s transportation network. (This part of the talk is based on ongoing work with Chamsi Hssaine, Siddhartha Banerjee, and Samitha Samaranayake.) Our results in both settings emphasize the importance of an integrated approach for tackling the next generation of challenges facing urban mobility.


Bio: Raga Gopalakrishnan is a postdoc in the School of Civil and Environmental Engineering, jointly with the School of Operations Research and Information Engineering, at Cornell University. His research concerns the design of service systems with strategic entities, especially the interplay between operational policies and economic mechanisms, emphasizing a practice-aware approach to Operations Research. Raga obtained his Ph.D. from the Department of Computing and Mathematical Sciences at California Institute of Technology in 2013, following which he was a postdoc at the University of Colorado Boulder, and a research scientist and manager at Xerox Research.


Monday, January 28

1174 Etcheverry Hall, 3:30 p.m. - 4:30 p.m.

Abstract: Applying sentiment analysis to news articles on large financial companies, we find that an increase in “unusual” negative news predicts an increase in stock market volatility and thus potential market stress. Similarly, unusual positive news forecasts lower volatility. Our analysis is based on more than 360,000 articles on 50 large financial companies, published in 1996–2014. Unusualness interacted with sentiment forecasts volatility – at both the company-specific and aggregate level – several months into the future. Furthermore, unusual news is reflected in volatility more slowly at the aggregate than at the company-specific level. The observed behavior of volatility in our analysis may be explained by attention constraints on investors. This is joint work with Harry Mamaysky.

Bio: Professor Glasserman's research and teaching address risk management, derivative securities, Monte Carlo simulation, statistics and operations. Prior to joining Columbia, Glasserman was with Bell Laboratories; he has also held visiting positions at Princeton University, NYU, and the Federal Reserve Bank of New York. In 2011-2012, he was on leave from Columbia and working at the Office of Financial Research in the U.S. Treasury Department, where he continues to serve as a part-time consultant.

Glasserman's publications include the book Monte Carlo Methods in Financial Engineering (Springer, 2004), which received the 2006 Lanchester Prize and the 2005 I-Sim Outstanding Publication Award. Glasserman is a past recipient of the National Young Investigator Award from the National Science Foundation (1994 - 99), IBM University Partnership Awards (1998 - 2001), the TIMS Outstanding Simulation Publication Award (1992), the Erlang Prize (1996), the IMS Medallion from the Institute of Mathematical Statistics (2006), and a fellowship from the FDIC Center for Financial Research (2004). He received the 2004 Wilmott Award for Cutting-Edge Research in Quantitative Finance and Risk Magazine's 2007 Quant of the Year Award, and he received a U.S. patent for an option pricing method. He was named an INFORMS Fellow in 2008. He is also a two-time recipient of the Dean's Award for Teaching Excellence (1994, 2000). Glasserman serves on the editorial boards of Finance & StochasticsMathematical Finance, the Journal of Derivatives, and Stochastic Systems.


Monday, January 28

Chou Hall, N370, 1 p.m. - 2:30 p.m.

Abstract: We integrate emerging trends intended to improve clinical trial design: design for cost-effectiveness, which ensures health-economic improvement of a new intervention over the current standard intervention; adaptive design, which dynamically adjusts the sample size and allocation of patients to different interventions; and multi-arm trial design, which compares multiple interventions simultaneously. Our goal is to identify a sequential sampling policy that dynamically decides the interventions to which patients should be allocated, as well as when to stop patient recruitment, in order to maximize the expected population-level benefit minus the cost of the trial. The literature on sequential sampling develops indices that either accommodate correlation among the mean rewards of alternatives or are based on optimal stopping times that can dynamically change as samples are taken, but not both. We develop the first tractable allocation and stopping rules whose indices capture both correlation and dynamic stopping times, and our numerical experiments demonstrate the value of considering both problem elements in the context of clinical trials.

Bio: Ozge Yapar is a Ph.D. candidate in the Operations, Information, and Decisions (OID) Department of the Wharton School of the University of Pennsylvania. Her research focuses on healthcare operations management, specifically on the process of developing and marketing new medical treatments. A focus of her work has been the growing pressure that healthcare payers, public health regulators, and the companies that develop new treatments face as they address increases in healthcare spending and seek to bring novel medical treatments to the market as quickly as possible. She uses tools from applied probability, stochastic processes, simulation optimization, and health economics.

In a recently completed work, she addresses the spiraling costs of clinical trials by developing an adaptive trial design that considers the value of multiple correlated alternatives. Her current work analyzes pharmaceutical risk-sharing contracts in which a medical treatment’s price can be updated to reflect what is learned about its performance after it has entered the market and is used by the general population.

Ozge earned her Bachelor of Science degree in Industrial Engineering from Bilkent University in Turkey, and she spent her third year of undergrad studying in Industrial Engineering and Operations Research Department of the University of California Berkeley as a visiting student.



Friday, February 8

1174 Etcheverry Hall, 3:30 p.m. - 4:30 p.m.

Abstract: Maintaining a fleet of buses to transport students to school is a major expense for U.S. school districts. In order to reduce costs by reusing buses between schools, many districts spread start times across the morning. However, assigning each school a time involves estimating the impact on transportation costs and reconciling additional competing objectives. Facing this intricate optimization problem, school districts must resort to ad hoc approaches, which can be expensive, inequitable, and even detrimental to student health. For example, there is medical evidence that early high school starts are impacting the development of an entire generation of students and constitute a major public health crisis. We present the first algorithm to jointly optimize school bus routing and bell time assignment. Our method leverages a natural decomposition of the routing problem, computing and combining subproblem solutions via mixed-integer optimization. It significantly outperforms state-of-the-art routing methods, and its implementation in Boston has led to $5 million in yearly savings, maintaining service quality for students despite a 50-bus fleet reduction. With the routing engine, we construct a tractable proxy to transportation costs, which allows the formulation of the bell time assignment problem as a multi-objective Generalized Quadratic Assignment Problem. Local search methods provide high-quality solutions, allowing school districts to explore tradeoffs between competing priorities and choose times that best fulfill community needs. In December 2017, the development of this method led the Boston School Committee to unanimously approve the first school start time reform in thirty years. In 2018, our collaboration with media and policy specialists generated an intense debate in Boston and many other cities on the use of OR tools for social good.

Bio: Sebastien Martin is a PhD candidate in Operations Research at MIT, advised by Prof. Dimitris Bertsimas and Patrick Jaillet. Beforehand, he obtained a MSc and BSc from Ecole Polytechnique in France. He has worked as a software engineering intern at Google Maps. His research focuses on large scale optimization, with applications in machine learning and transportation, and an emphasis on implementation and policy. His recent work, covered by the Wall Street Journal, the Boston Globe and Wired, led to major policy changes and millions of dollars in yearly saving for Boston and is a 2019 Edelman Award finalist.


Monday, February 11

1174 Etcheverry Hall, 3:30 p.m. - 4:30 p.m.

Abstract: The transportation and logistics industries are undergoing a round of revolutionary innovation. This innovation is fueled by two key drivers: (1) the growing availability of data, and (2) new operational paradigms in a sharing economy. This talk focuses on showcasing how new models, enabled by the prevalence of data, can lead to significant value in operational decision-making.

We begin by presenting our research that shows how trip data in bike-sharing systems can be mined to infer rider substitution behaviors when there are bike or dock shortages. Based on a non-parametric ranking-based choice model, we propose efficient enumeration procedures and first-order methods to solve the large-scale estimation problem by exploiting problem structure. We prove consistency results of our method. We then demonstrate, with Boston Hubway data, that ridership can be significantly improved through effective inventory allocation operations with better demand modeling.

Next, we describe a recent work in which we propose a new car-pooling mechanism in ride-hailing, called dynamic waiting which varies rider waiting before dispatch. The goal is to limit price volatility in ride-hailing services by reducing the role of surge pricing. We describe a steady-state model depicting the long-run average performance of a ride-hailing service, and characterize the system equilibrium. Calibrating the model using Uber data, we reveal insights on welfare-maximizing pricing and waiting strategies. We show that, with dynamic waiting, price can be lowered, its variability is mitigated and total welfare is increased. 

Bio: Chiwei Yan received his PhD from the Operations Research Center at MIT in 2017. His current research interest is in transportation and logistics, with a focus on data-driven optimization and emerging problems in a sharing economy. He is a recipient of the Best Dissertation Award Honorable Mention and the Outstanding Paper Award in Air Transportation from INFORMS Transportation Science and Logistics Society, the Best Dissertation Award from INFORMS Aviation Application Section, the AGIFORS Anna Valicek Award, and the UPS Doctoral Fellowship, among others. His research involves collaborations with both the private and public sectors, including the Federal Aviation Administration, Sabre Airline Solutions, Boston Hubway Bikes and Uber. Before coming to MIT, he obtained the Bachelor of Science in Industrial Engineering from Tsinghua University.

Currently, Dr. Yan is a data scientist in the marketplace optimization group at Uber, leading the design and implementation of the latest rider surge pricing algorithm which balances supply and demand in real time across 600+ global markets.


Friday, February 15

1174 Etcheverry Hall, 3:30 p.m. - 4:30 p.m.

Abstract: Most of the real-life problems involve uncertainty, which need to be delicately integrated into the decision-making processes. In this talk, we present various stochastic optimization techniques motivated by maintenance, operations and capacity expansion planning problems in energy systems. In the first part of the talk, our aim is to effectively model and solve the integrated condition-based maintenance and operations scheduling problem of a fleet of generators. We develop a data-driven optimization framework that explicitly considers the effect of the sensor-driven generator failure scenarios and operations schedules on the generators’ degradation levels to construct a reliable and cost-efficient plan. In the second part of the talk, we shift our focus to a more generic problem setting in sequential decision-making under uncertainty. Although two-stage and multi-stage stochastic programming are among the key methodologies to address multi-period problems under uncertainty, they might not provide adequate solutions under limited flexibility by resulting in either fully static or dynamic policies. We propose a novel adaptive stochastic programming approach, in which we optimize the time to revise decisions. We provide theoretical bounds on the performance of the proposed approach compared to the static and dynamic approaches, and present practical implications of the choice of the revision time. We also tailor solution algorithms using our analytical analyses and derive their approximation guarantees. To illustrate our results, we study a generation expansion planning problem demonstrating the advantages of the adaptive approach over existing policies. 

Bio: Beste Basciftci is currently a PhD candidate in Operations Research at the H. Milton Stewart School of Industrial and Systems Engineering, Georgia Institute of Technology, with a minor in Statistics. She received her bachelor's degrees in Industrial Engineering and Computer Engineering from Boğaziçi University with High Honors. She also hold a master's degree in Industrial Engineering from Boğaziçi University. She is broadly interested in data-driven decision making problems under uncertainty. Methodologically, her research focuses on developing mixed-integer, stochastic programming and distributionally robust optimization approaches to address operations research/management related problems, specifically for applications in energy, supply chains, production systems, and healthcare operations. Her research also involves developing and integrating statistical modeling and business analytics approaches to the subsequent decision-making processes.