Speaker Information & Talk Abstracts

Dimitris Bertsimas

Massachusetts Institute of Technology

Title: The Voice of Optimization

Abstract: We present a new way to see optimization problems. Using machine learning techniques we are able to predict the strategy behind the optimal solution in any continuous and mixed-integer convex optimization problem as a function of its key parameters. The benefits of our approach are interpretability and speed. We use interpretable machine learning algorithms such as optimal classification trees (OCTs) to gain insights on the relationship between the problem parameters and the optimal solution. In this way, optimization is no longer a black-box and we can understand it. In addition, once we train the predictor, we can solve optimization problems at very high speed. This aspect is also relevant for non interpretable machine learning methods such as neural networks (NNs) since they can be evaluated very efficiently after the training phase. We show on several realistic examples that the accuracy behind our approach is in the 90%-100% range, while even when the predictions are not correct, the degree of suboptimality or infeasibility is very low. We also benchmark the computation time beating state of the art solvers by multiple orders of magnitude. Therefore, our method provides on the one hand a novel insightful understanding of the optimal strategies to solve a broad class of continuous and mixed-integer optimization problems and on the other hand a powerful computational tool to solve online optimization at very high speed. (joint work with Bartolomeo Stellato, MIT).

[Abstract]

About the speaker: Dimitris Bertsimas is currently the Boeing Professor of Operations Research and the co-director of the Operations Research Center and faculty director of the Master of Business analytics at MIT. He received his SM and PhD in Applied Mathematics and Operations Research from MIT in 1987 and 1988 respectively. He has been with the MIT faculty since 1988. His research interests include optimization, machine learning and applied probability and their applications in health care, finance, operations management and transportation. He has co-authored more than 200 scientific papers and four graduate level textbooks. He is the editor in Chief of INFORMS Journal of Optimization and former department editor in Optimization for Management Science and in Financial Engineering in Operations Research. He has supervised 62 doctoral students and he is currently supervising 25 others. He is a member of the National Academy of Engineering since 2005, an INFORMS fellow, and he has received numerous research awards including the Morse prize (2013), the Pierskalla award for best paper in health care (2013), the best paper award in Transportation (2013) the Farkas prize (2008), the Erlang prize (1996), the SIAM prize in optimization (1996), the Bodossaki prize (1998) and the Presidential Young Investigator award (1991-1996).View More

Shane G. Henderson

Cornell University

Title: Under the Hood of Bike Sharing

Abstract: Cornell’s work on bike sharing with Citi Bike and its parent company Motivate relies on a combination of data analysis, stochastic modeling and optimization to help inform both the design and operation of the largest bike-sharing operations in North America. I’ll discuss our work and its impact, but focus on some of the inner workings of the stochastic modeling. This includes the use of (one of) the Poisson equation(s) in the computation of a central performance measure and a heuristic underlying a simulation-optimization principle that is likely useful in many other contexts. (Joint work with Daniel Freund, Nanjing Jian, Eoin O’Mahony, David Shmoys)

[Abstract]

About the speaker: Shane G. Henderson is professor and director (from July 2017) of the School of Operations Research and Information Engineering at Cornell University. He received his MS and PhD in Statistics and Operations Research from Stanford University in 1995 and 1997 respectively. He has previously held positions in the Department of Industrial and Operations Engineering at the University of Michigan and the Department of Engineering Science at the University of Auckland. His research interests include discrete-event simulation, simulation optimization, and emergency services planning.He is the editor in chief of Stochastic Systems. He has served as chair of the INFORMS Applied Probability Society, and as simulation area editor for Operations Research. He is an INFORMS Fellow. View More

Costis Maglaras

Columbia University

Title: Observational Learning and Abandonment in Congested Systems

Abstract: Demand systems used in operations management and service operations settings often assume that system parameters that may affect user decisions, e.g., to join a system or purchase a service, are known or accurately communicated to the market. In several practical settings this need not be the case, but users may still form estimates of these system parameters through their own observations or experiences in the system. In this talk, I will study the effect of observational learning on user behavior and equilibrium system performance in the context of a queueing model. Specifically, I analyze a congested service system in which delay-sensitive customers have no a priori knowledge of the service rate, but instead join the system and observe their progress through the queue in order to learn the system's service rate, estimate remaining waiting times, and make abandonment decisions. (Joint work with John Yao and Assaf Zeevi.)

[Abstract]

About the speaker: Costis Maglaras is a Professor at the Graduate School of Business at Columbia University in the division of Decision, Risk & Operations. He received his BS in Electrical Engineering from Imperial College, London, in 1990, and his MS and PhD in Electrical Engineering from Stanford University in 1991 and 1998, respectively. He joined Columbia Business School in 1998. His recent work focuses on the application of quantitative modeling in the pricing, risk management and valuation of multi-unit real estate portfolios, and in the design of portfolio trading systems and portfolio trading algorithms. His research has been recognized by the 1999 INFORMS Nicholson Prize for the best student paper in Operations Research and Management Science, the 2008 INFORMS Revenue Management and Pricing Section best research paper award, and his students have won several best paper awards for their doctoral research. He has also received the Dean’s award at Columbia Business School for teaching excellence for the core course Managerial Statistics. He serves as the chair of INFORMS’ Revenue Management and Pricing Section for 2008-2009. He teaches and serves as faculty director for the executive education course on Risk Management offered by Columbia Business School. He holds editorial positions in many of the flagship journals in operations research and management science.View More

David Simchi-Levi

Massachusetts Institute of Technology

Title: Phase Transitions and Cyclic Phenomena in Bandits with Switching Constraints

Abstract: We consider the classical stochastic multi-armed bandit problem with a constraint on the total cost incurred by switching between actions. We prove matching upper and lower bounds on regret and provide near-optimal algorithms for this problem. Surprisingly, we discover phase transitions and cyclic phenomena of the optimal regret. That is, we show that associated with the multi-armed bandit problem, there are phases defined by the number of arms and switching costs, where the regret upper and lower bounds in each phase remains the same and drop significantly between phases. The results enable us to fully characterize the trade-off between regret and incurred switching cost in the stochastic multi-armed bandit problem, contributing new insights to this fundamental problem.

[Abstract]

About the speaker: David Simchi-Levi is a Professor of Engineering Systems at MIT. He is considered one of the premier thought leaders in supply chain management and business analytics. His research focuses on developing and implementing robust and efficient techniques for operations management. His Ph.D. students have accepted faculty positions in leading academic institutes including U. of California Berkeley, Carnegie Mellon U., Columbia U., Cornell U., Duke U., Georgia Tech, Harvard U., U. of Illinois Urbana-Champaign, U. of Michigan, Purdue U. and Virginia Tech. Professor Simchi-Levi co-authored the books Managing the Supply Chain (McGraw-Hill, 2004), the award winning Designing and Managing the Supply Chain (McGraw-Hill, 2007) and The Logic of Logistics (3rd edition, Springer 2013). He also published Operations Rules: Delivering Customer Value through Flexible Operations (MIT Press, 2011). Professor Simchi-Levi is the current Editor-in-Chief of Management Science, one of the two flagship journals of INFORMS. He served as the Editor-in-Chief for Operations Research (2006-2012), the other flagship journal of INFORMS and for Naval Research Logistics (2003-2005). He is an INFORMS Fellow, MSOM Distinguished Fellow and the recipient of the 2014 INFORMS Daniel H. Wagner Prize for Excellence in Operations Research Practice; 2014 INFORMS Revenue Management and Pricing Section Practice Award; 2009 INFORMS Revenue Management and Pricing Section Prize and Ford 2015 Engineering Excellence Award. Professor Simchi-Levi has consulted and collaborated extensively with private and public organizations. He was the founder of LogicTools which provided software solutions and professional services for supply chain optimization. LogicTools became part of IBM in 2009. In 2012 he co-founded OPS Rules, an operations analytics consulting company. The company became part of Accenture in 2016. In 2014, he co-founded Opalytics, a cloud analytics platform company focusing on operations and supply chain intelligence. The company became part of the Accenture Applied Intelligence in 2018. View More

Yinyu Ye

Stanford University

Title: The Sample Complexity in Data-Driven Optimization

Abstract: We consider data-driven optimization problems where the objective/reward functions can only be estimated via sample data, for examples, 1) the discounted Markov Decision Process (DMDP) or Reinforced Learning (RL) where we can only access its value functions through a generative sampling model and 2) the stochastic programming (SP) via the sample average approximation (SAA) scheme. In such settings, a fundamental question is how many examples are sufficient and necessary to learn/compute an $\epsilon$-optimal policy/solution with high probability. In this talk, we present two results along this research direction. The first result is on infinite horizon DMDP we provide an algorithm which computes an $\epsilon$-optimal policy with high probability where both the run time spent and number of samples taken are near optimal and it matches the sample complexity lower bound up to logarithmic factors. We also extend our method to the finite-horizon MDP with a generative model and provide a nearly matching sample complexity lower bound. The second result in on general stochastic programming (SP) where the classical SAA dictates that the number of samples should be polynomial in the number of problem dimensions in order to ensure proper optimization accuracy. We now study a modification to the SAA in the scenario where the global minimizer is either sparse or can be approximated by a sparse solution. By making use of a regularization penalty referred to as the folded concave penalty (FCP), we show that, if an FCP-regularized SAA formulation is solved to meet a weaker second-order necessary condition, then the required number of samples can be significantly reduced in approximating the global solution of a convex SP: the sample size is only required to be poly-logarithmic in the number of dimensions. The efficacy of the FCP regularization for nonconvex and non-smooth SPs is also discussed, and the finding allows us to further understand two important paradigms: the high-dimensional statistical learning (HDSL) and the (deep) neural networks (NN).

[Abstract]

About the speaker: Yinyu Ye is currently the Kwoh-Ting Li Professor in the School of Engineering at the Department of Management Science and Engineering and Institute of Computational and Mathematical Engineering and the Director of the MS&E Industrial Affiliates Program, Stanford University. He received the B.S. degree in System Engineering from the Huazhong University of Science and Technology, China, and the M.S. and Ph.D. degrees in Engineering-Economic Systems and Operations Research from Stanford University. Ye's research interests lie in the areas of optimization, complexity theory, algorithm design and analysis, and applications of mathematical programming, operations research and system engineering. He is also interested in developing optimization software for various real-world applications. He is an INFORMS Fellow, and has received several research awards including the winner of the 2014 SIAG/Optimization Prize, the inaugural 2012 ISMP Tseng Lectureship Prize for outstanding contribution to continuous optimization, the 2009 John von Neumann Theory Prize for fundamental sustained contributions to theory in Operations Research and the Management Sciences, the inaugural 2006 Farkas prize on Optimization, and the 2009 IBM Faculty Award. He has supervised numerous doctoral students at Stanford who received the 2015 and 2013 Second Prize of INFORMS Nicholson Student Paper Competition, the 2013 INFORMS Computing Society Prize, the 2008 Nicholson Prize, and the 2006 and 2010 INFORMS Optimization Prizes for Young Researchers.View More

Assaf Zeevi

Columbia University

Title: Learning Theoretic Challenges in Operation

Abstract: Over the past decade or so, machine learning methods have started to enter the OR/MS literature and are playing an increasingly important roll in the study of traditional as well as emerging application areas. The recent creation of “learning theory” (or equivalently named/defined) departments in most of the flagship INFORMS journals has further validated the status of this research area within our community. This talk will provide an abridged (and admittedly biased) tour of learning theory trends in operations, emphasizing the synergy between the two disciplines. In particular, we will indicate how learning theory provides a useful array of tools and concepts that compliment and significantly extend the scope of traditional OR-type analysis techniques, while fundamental problems in operations present novel learning theoretic challenges.

[Abstract]

About the speaker: Assaf Zeevi is Professor and holder of the Kravis chair at the Graduate School of Business, Columbia University. His research focuses on the formulation and analysis of mathematical models of complex systems, with particular research and teaching interests that lie at the intersection of Operations Research, Statistics, Computer Science and Economics. Recent research has centered on problems arising in the data science and machine learning space, with application domains that include healthcare analytics, dynamic pricing and demand learning, the design of recommendation engines, online retail and assortment selection, and the valuation and monetization of digital goods. Assaf received his B.Sc. and M.Sc. (Cum Laude) from the Technion, in Israel, and subsequently his Ph.D. from Stanford University. He is the recipient of several research awards including a CAREER Award from the National Science Foundation, an IBM Faculty Award, Google Research Award, as well as several best paper recognitions. Assaf has recently served as Vice Dean for Research at Columbia Business School, and Editor-in-Chief of Stochastic Systems (the flagship journal of INFORMS Applied Probability Society). He is currently serving on several editorial boards in his professional community, as well as scientific advisory boards for startup companies in the high technology sector.View More