IEOR-DRO Seminar

Title: Multi-Item Online Order Fulfillment in a Two-Layer Network

Abstract: The boom of e-commerce in the globe in recent years has expedited the expansion of fulfillment infrastructures by e-retailers. While e-retailers are building more and more mini-warehouses close to end customers to offer faster delivery service than ever, the associated fulfillment costs have skyrocketed. In this paper, we study a real-time fulfillment problem in a two-layer RDC-FDC distribution network that has been implemented in practice by major e-retailers. In such a network, the upper layer contains larger regional distribution centers (RDCs) and the lower layer contains smaller front distribution centers (FDCs). We allow order split: an order can be split and fulfilled from multiple warehouses at an additional cost. The objective is to minimize the routine fulfillment costs. We study real-time fulfillment algorithms with performance guarantees in both settings with and without demand forecasts.
This is joint work with Yanyang Zhao (Chicago) and Xinshang Wang (Alibaba). An old version of the paper can be found here.

Bio: Linwei Xin is an assistant professor of Operations Management at Booth School of Business, University of Chicago. His primary research is on inventory and supply chain management: designing models and algorithms for organizations to effectively "match supply to demand" in various contexts with uncertainty. His research on stochastic inventory theory by using asymptotic analysis has been recognized with several INFORMS paper competition awards, including the Applied Probability Society Best Publication Award (2019), First Place in the George E. Nicholson Student Paper Competition (2015), Second Place in the Junior Faculty Interest Group Paper Competition (2015), and a finalist in the Manufacturing and Service Operations Management Student Paper Competition (2014). His work with JD.com on dispatching algorithms for robots in intelligent warehouses was recognized as a finalist for the INFORMS 2021 Franz Edelman Award, with an estimate of billions of dollars in savings. His other honors include winning a National Science Foundation grant as a principal investigator. His research has been published in journals such as Operations Research and Management Science. He currently teaches MBA and PhD courses at the University of Chicago.

Title: Demystifying the Efficiency of Reinforcement Learning: Two Recent Stories

Abstract: Reinforcement learning (RL), which is frequently modeled as sequential learning and decision making in the face of uncertainty, is garnering growing interest in recent years due to its remarkable success in practice. In contemporary RL applications, it is increasingly more common to encounter environments with prohibitively large state and action space, thus imposing stringent requirements on the sample and computational efficiency of the RL algorithms in use. Despite the empirical success, however, the theoretical underpinnings for many popular RL algorithms remain highly inadequate even for the tabular setting. 

In this talk, we present two vignettes regarding the effectiveness of RL algorithms. The first vignette demonstrates that a perturbed model-based RL approach is minimax optimal under a generative model, without suffering from a sample size barrier that was present in all past work. The second vignette covers policy optimization in reinforcement learning. On the one hand, we demonstrate that the popular softmax policy gradient method can take exponential time to converge; on the other hand, employing natural policy gradients and enforcing entropy regularization provably achieve fast global convergence. These results cover two distinctive RL paradigms, and might shed light on the efficacy of these algorithms in more complicated scenarios.

Bio: Yuxin Chen is currently an assistant professor in the Department of Electrical Engineering at Princeton University, and is affiliated with Applied and Computational Mathematics, Computer Science, and Center for Statistics and Machine Learning. Prior to joining Princeton, he was a postdoctoral scholar in the Department of Statistics at Stanford University, and he completed his Ph.D. in Electrical Engineering at Stanford University. His research interests include high-dimensional statistics, mathematical optimization, and reinforcement learning. He has received the Princeton graduate mentoring award, Princeton SEAS junior faculty award, the AFOSR Young Investigator Award, the ARO Young Investigator Award, the ICCM best paper award (gold medal), and was selected as a finalist for the Best Paper Prize for Young Researchers in Continuous Optimization.  

Seminar Recording

Title: Value of Promotions with Delayed Incentives: An Empirical Investigation of Gift Card Promotions

Synopsis: A gift card promotion provides customers an incentive to spend more than an expenditure level on regularly priced (as opposed to discounted) products, by rewarding customers with a gift card to be redeemed against a future purchase. This type of promotion is widely used by luxury fashion, department, and consumer electronic stores. In this paper, we empirically test and quantify the causal effects of gift card promotion on the retailers' revenue and the mechanisms through which it impacts customers' purchase behavior. To do so, we collaborate with a major U.S.-based fashion retailer that regularly runs gift card promotions on its online channel by targeting its customers through promotion emails. We utilize our partner retailer’s targeting policy to run a regression discontinuity design analysis. We find that gift card promotion increases retailer's revenue by up to 113%. The magnitude of this effect, however, reduces with the customers’ recency of purchase from the retailer’s website.  Using our estimations, we attribute $554K to $1.14M (i.e., 18.50% to 38.14%) of the retailer's $2.99M total online channel revenue during promotion days to gift card promotion, on average. We find that the gift card promotion email increases sales even among customers who do not participate in the promotion. Specifically, advertising the gift card promotion through an email increases customers’ preference for the retailer but has no significant informative or reminder effects. We attribute 70 to 89% of the promotion’s total impact to this advertisement effect. We also run a propensity score matching analysis and find that customers who redeem their earned gift cards increase their expenditures (beyond the face value of the gift card) by 23.3%, on average. Therefore, we show that redemption of gift cards induces additional expenditure and, hence, can be profitable for the retailer offsetting any promotional costs incurred. 

This paper is joint work with Bharadwaj Kadiyala and Özalp Özer. The paper is available at SSRN: https://ssrn.com/abstract=3499711

Bio: A. Serdar Şimşek is an Assistant Professor within the Operations Management Group of Naveen Jindal School of Management at The University of Texas at Dallas since 2015. He received his Ph.D. degree in Decision, Risk, and Operations from Columbia University, Graduate School of Business. Prior to joining UTD, he spent two years as a researcher/instructor at the ORIE Department of Cornell University.  

His research focuses on understanding and improving firms’ pricing decisions primarily in the retail industry and business-to-business markets. The main goal of his research is twofold: (i) providing actionable insights and methodologies that enable a firm to estimate and quantify the impact of consumers’ strategic and behavioral motives on their purchase decisions, and (ii) designing effective pricing and promotion modalities that optimize profitability as well as consumer satisfaction accordingly. His research appeared in Management Science, Operations Research, Manufacturing and Service Operations Management, and Production and Operations Management. 

Title: Detecting equivalence between iterative algorithms for optimizations

Abstract: When are two algorithms the same? How can we be sure a recently proposed algorithm is novel, and not a minor twist on an existing method? In this talk, we present a framework for reasoning about equivalence between a broad class of iterative algorithms, with a focus on algorithms designed for convex optimization. We propose several notions of what it means for two algorithms to be equivalent, and provide computationally tractable means to detect equivalence. Our main definition, oracle equivalence, states that two algorithms are equivalent if they result in the same sequence of calls to the function oracles (for suitable initialization). Borrowing from control theory, we use state-space realizations to represent algorithms and characterize algorithm equivalence via transfer functions. Our framework can also identify and characterize some algorithm transformations including permutations of the update equations, repetition of the iteration, and conjugation of some of the function oracles in the algorithm. A software package named Linnaeus implements the framework and makes it easy to find other iterative algorithms that are equivalent to an input algorithm. More broadly, this framework and software advances the goal of making mathematics searchable.

Bio:  Madeleine Udell is Assistant Professor of Operations Research and Information Engineering and Richard and Sybil Smith Sesquicentennial Fellow at Cornell University. She studies optimization and machine learning for large scale data analysis and control, with applications in marketing, demographic modeling, medical informatics, engineering system design, and automated machine learning. She has received several awards, including an Alfred P. Sloan Research Fellowship (2021), a National Science Foundation CAREER award (2020), an Office of Naval Research (ONR) Young Investigator Award (2020), a Cornell Engineering Research Excellence Award (2020), an INFORMS Optimization Society Best Student Paper Award (as advisor) (2019), and INFORMS Doing Good with Good OR (2018). Her work is supported by grants from the NSF, ONR, DARPA, the Canadian Institutes of Health, and Capital One.

Seminar Recording

Title: Scoring-based resource allocation: theory and application

Abstract: Scarce resources are often allocated using scoring rules. We will discuss two approaches to deriving optimal scoring rules that also account for fairness considerations. The first approach relies on integer optimization, as well as a scalable heuristic, backed by a performance guarantee. The second approach relies on a novel multi-objective optimization methodology that also leverages machine learning. Furthermore, we will discuss how we applied our methodology to help the re-design of organ transplant allocation policies in the U.S., and elaborate on how we worked with U.S. policymakers to guide the upcoming changes in lung transplant allocation. 

Bio: Nikos Trichakis is Associate Professor of Operations Management at the MIT Sloan School of Management. His research interests include optimization under uncertainty, data-driven optimization and analytics, with application in healthcare, supply chain management, and finance. Trichakis is also interested in the interplay of fairness and efficiency in resource allocation problems and operations, and the inherent tradeoffs that arise in balancing these objectives. His work has been awarded, among others, the INFORMS Optimization Society Young Researchers Prize, the M&SOM Interface of Finance, Operations, and Risk Management Best Paper Award, the INFORMS Junior Faculty Paper Prize, and the INFORMS Koopman Prize.  

Title: Optimal Queue Design

Abstract: We study the optimal design of a queueing system when agents’ arrival and servicing are governed by a general Markov process. The designer chooses entry and exit rules for agents, their service priority—or queueing discipline—as well as their information, while ensuring they have incentives to follow recommendations to join the queue and, importantly, to stay in the queue. Under a mild condition, at the optimal mechanism, agents are induced to enter up to a certain queue length and no agents are to exit the queue; agents are served according to a first-come-first-served (FCFS) rule; and they are given no information throughout the process beyond the recommendations they receive from the designer. FCFS is also necessary for optimality in a rich domain. We identify a novel role for queueing disciplines in regulating agents’ beliefs, and their dynamic incentives, thus uncovering a hitherto unrecognized virtue of FCFS in this regard.

Bio: Yeon-Koo Che is Kelvin J. Lancaster Professor of Economic Theory at Columbia University. His early work contributes to the theory of mechanism and auction design: scoring-rule auctions, auctions with budget constraints, collusion-proof mechanism design, research contest, the incomplete contract paradigm for organization theory, and the matching theory in the context of college and school choice. His current research projects explore the implications of data-driven economic decision making and resource allocation for welfare and distributional consequences. He is Fellow of Econometric Society (elected 2009) and Fellow of Economic Theory (elected 2014) for the Society of Advancement of Economic Theory. He is a member of Council of Game Theory Society (elected 2017) and of Asian Regional Council of Econometric Society (elected 2016). He served as Executive Director of Program for Economic Research (2015-18). He was editor of Journal of Industrial Economics, associate editor of Econometrica, and is currently serving as advisory editor of Games and Economic Behavior. He was the inaugural recipient in 2008 of the Cho Rakkyo Prize, and the KAEA-MK Prize in 2009. He has given numerous Keynote addresses, including the Jacob Marschak Lecture at the Econometric Society meeting in Sydney (2016), Asian Meeting of Econometric Society (2018), and Latin American Meeting of Econometric Society (2018). He has received nine National Science Foundation grants spanning over 20 years. He received a PhD in Economics at the Stanford University. He was Professor at University of Wisconsin before joining the Columbia University as Professor of Economics in 2005.

Seminar Recording

To be announced.

Title: Bayesian optimization as black box method for complex engineering applications

Abstract: Systems across automotive, bio-pharma, aerospace, energy, have become increasingly complex, and simulation represents a standard tool to evaluate their performance independently from the purpose of the analysis being optimization, control, certification. As a result, black-box optimization, that can embed simulation as a black box to perform a wide range of analyses, has attracted a lot of attention from the science and engineering communities. This talk centers around Black-box optimization methods, focusing on random search approaches (such randomness is injected in the search independently from the problem being affected by noise). We first present two algorithms that focus on: (1) control and acceleration of the explore/exploit process, (2) scalability into high dimensions (1000’s,10000’s). One approach alternates local and global search using local knowledge while exploring the space of possible solutions. To scale, we decompose the original problem using game theory. The performance of the proposed approaches is analyzed, and key future directions are discussed. In the second part of the talk, we look into algorithms developed in the scope of certification of safety critical systems: Part-X is a family of partitioning informed Bayesian optimizers that can identify regions in which the system can present safety concerns (bugs in the case a software is analyzed), min-BO works to identify faults in systems that have complex requirements that can be decomposed into a set of simpler requirements that need to be simultaneously satisfied by the system. We show the basic ideas behind the design of Part-X and min-BO, and preliminary results. We conclude with pointers into researching to expand Bayesian optimization to embed structure that is usually readily available from complex, but still, engineered systems.

Bio: Giulia Pedrielli (https://www.gpedriel.com/) is currently Assistant Professor for the School of Computing and Augmented Intelligence (SCAI) at Arizona State University. She graduated from the Department of Mechanical Engineering of Politecnico di Milano. Giulia develops her research in design and analysis of random algorithms for global optimization, with focus on improving finite time performance and scalability of these approaches. Her work is motivated by design and control of next generation manufacturing systems in bio-pharma and aerospace applications, as well as problems in the design and evaluation of complex molecular structures in life-science. Applications of her work are in individualized cancer care, bio-manufacturing, design and control of self-assembled RNA structures, verification of Cyberphysical systems. Her research is funded by the NSF, DHS, DARPA, Intel, Lockheed Martin.

Seminar Recording 

Title: Assortment Optimization under the Decision Forest Model

Abstract: The decision forest model is a recently proposed nonparametric choice model that is capable of representing any discrete choice model and in particular, can be used to represent non-rational customer behavior. In this paper, we study the problem of finding the assortment that maximizes expected revenue under the decision forest model. This problem is of practical importance because it allows a firm to tailor its product offerings to profitably exploit deviations from rational customer behavior, but at the same time is challenging due to the extremely general nature of the decision forest model. We approach this problem from a mixed-integer optimization perspective and propose three different formulations. We theoretically compare these formulations in strength, and analyze when these formulations are integral in the special case of a single tree. We propose a methodology for solving these problems at a large-scale based on Benders decomposition, and show that the Benders subproblem can be solved efficiently by primal-dual greedy algorithms when the master solution is fractional for two of our formulations, and in closed form when the master solution is binary for all of our formulations. Using synthetically generated instances, we demonstrate the practical tractability of our formulations and our Benders decomposition approach, and their edge over heuristic approaches. This is joint work with Yi-Chun Chen (UCLA Anderson PhD student). 

Bio: Velibor Misic is an assistant professor of Decisions, Operations and Technology Management at the UCLA Anderson School of Management. Prior to joining UCLA, he received his BASc and MASc degrees in industrial engineering from the University of Toronto and his PhD degree in operations research from MIT. His broad research interest is in analytics, with a focus on customer choice, dynamic decision making under uncertainty and problems at the intersection of optimization and machine learning . His work has been recognized with several awards, including second place in the INFORMS Junior Faculty Interest Group (JFIG) Best Paper Award and as finalist in the INFORMS Data Mining Section Best Paper Competition. At UCLA Anderson, he has received the Master of Science in Business Analytics (MSBA) Faculty Excellence Award twice and the Eric and "E" Juline Faculty Excellence in Research Award. He currently serves as an associate editor for the INFORMS journals Manufacturing & Service Operations Management and Service Science.

To be announced.

To be announced.