Title: Group-Sparse Matrix Factorization for Transfer Learning of Word Embeddings

Abstract: Unstructured text provides decision-makers with a rich data source in many domains, ranging from product reviews in retail to nursing notes in healthcare. To leverage this information, words are typically translated into word embeddings---vectors that encode the semantic relationships between words---through unsupervised learning algorithms such as matrix factorization. However, learning word embeddings from new domains with limited training data can be challenging, because the meaning/usage may be different in the new domain, e.g., the word "positive" typically has positive sentiment, but often has negative sentiment in medical notes since it may imply that a patient tested positive for a disease. In practice, we expect that only a small number of domain-specific words may have new meanings. We propose an intuitive two-stage estimator that exploits this structure via a group-sparse penalty to efficiently transfer learn domain-specific word embeddings by combining large-scale text corpora (such as Wikipedia) with limited domain-specific text data. We bound the generalization error of our transfer learning estimator, proving that it can achieve high accuracy with substantially less domain-specific data when only a small number of embeddings are altered between domains. Furthermore, we prove that all local minima identified by our nonconvex objective function are statistically indistinguishable from the global minimum under standard regularization conditions, implying that our estimator can be computed efficiently. Our results provide the first bounds on group-sparse matrix factorization, which may be of independent interest. We empirically evaluate our approach compared to state-of-the-art fine-tuning heuristics from natural language processing.

Bio: Kan Xu is currently an Assistant Professor of Information Systems at Arizona State University, W. P. Carey School of Business. Previously, he completed his PhD degree from University of Pennsylvania, Department of Economics. His research focuses on developing novel machine learning methods for data-driven decision making practices, with applications to healthcare, textual analytics, digital platform, and pricing. In particular, his recent works focus on sequential decision making (e.g., bandits, reinforcement learning), and decision making with unstructured (e.g. natural language processing (NLP)) or matrix form (e.g., matrix completion).

Title: Convex Surrogate Loss Functions for Contextual Pricing with Transaction Data

Abstract: We study an off-policy contextual pricing problem where the seller has access to samples of prices that customers were previously offered, whether they purchased at that price, and auxiliary features describing the customer and/or item being sold. This is in contrast to the well-studied setting in which samples of the customer's valuation (willingness to pay) are observed. In our setting, the observed data is influenced by the previous pricing policy, and we do not know how customers would have responded to alternative prices. We introduce suitable loss functions for this setting that can be directly optimized to find an effective pricing policy with expected revenue guarantees, without the need for estimation of an intermediate demand function.

We focus on convex loss functions. This is particularly relevant when linear pricing policies are desired for interpretability reasons, resulting in a tractable convex revenue optimization problem. We propose generalized hinge and quantile pricing loss functions that price at a multiplicative factor of the conditional expected valuation or a particular quantile of the prices that sold, despite the valuation data not being observed. We prove expected revenue bounds for these pricing policies when the valuation distribution is log-concave, and we provide generalization bounds for the finite sample case. Finally, we conduct simulations on both synthetic and real-world data to demonstrate that this approach is competitive with, and in some settings outperforms, state-of-the-art methods in contextual pricing.

Bio: Max is an Assistant Professor at the Darden School of Business in the Quantitative Analysis group. He received his Ph.D. from the Operations Research Center at the Massachusetts Institute of Technology. Prior to joining Darden, Max was a Post-Doctoral Researcher at IBM Thomas J Watson Research Center. His research focuses on data-driven optimization and prescriptive analytics. Specifically, he is interested in using data, machine learning and optimization techniques to help solve operational decisions. Recent application areas include pricing and revenue management, healthcare, and logistics.

Title: Generative AI and Creative Problem Solving

Abstract: The rapid advances in generative artificial intelligence (AI) open up attractive opportunities for creative

problem-solving through human-guided AI partnerships. To explore this potential, we initiated a

crowdsourcing challenge focused on sustainable, circular economy business ideas and assessed the novelty, value, and creativity of solutions created by both the human crowd and the collaborative efforts of one human and AI. The challenge attracted 125 global solvers from various industries and the human-prompted AI solutions were generated using strategic prompt engineering. We recruited 300 evaluators to judge a randomized selection of 13 out of 234 solutions, totaling 3,900 evaluator–solution pairs. Our findings demonstrate that the solutions generated through human-AI collaboration matched the creativity of those from the human solvers. Whereas the human-AI solutions provided more value, the human-only solutions were more innovative—both on average and for highly novel outcomes. Our study explores the potential for incorporating “AI-in-the-loop” into creative problem-solving, offering a scalable and cost-efficient method for enhancing the early phases of innovation. Our research paves the way for future exploration of how AI can be integrated into creative processes to foster more effective innovation.

Bio: Jackie Lane is an Assistant Professor in the Technology and Operations Management Unit at Harvard Business School and a co-Principal Investigator of the Laboratory for Innovation Science at Harvard (LISH) at the Digital Data Design Institute (D^3) at Harvard.

In her academic research, Jackie investigates the strategic utilization of diverse knowledge and perspectives within organizations to identify and champion groundbreaking projects and ideas. Her recent work explores how expertise influences the development and evaluations of early-stage projects, and how artificial intelligence can complement human expertise in decision-making processes. She has partnered with organizations such as NASA and Harvard Medical School to design and run field experiments to solve their innovation problems and challenges.

Jackie earned a bachelor’s degree in operations research and financial engineering from Princeton University, an MBA from Columbia Business School, and a PhD from Northwestern University. Most recently, she was a postdoctoral fellow at Harvard Business School and the Laboratory for Innovation Science at Harvard. Jackie has also worked in sales & trading and equity research at Morgan Stanley and as a finance and operations manager at Microsoft.

Title:  Measuring Strategic Behavior by Gig Economy Workers: Multihoming and Repositioning

Abstract: Gig economy workers make strategic decisions about where and when to work. These decisions are central to gig economy operations and are important policy targets both to firms that operate ridehail and delivery platforms and to regulators that oversee labor markets. We collaborate with a driver analytics company to empirically measure two types of strategic behavior: multihoming, an online change between platforms, and repositioning, a physical change between locations. Using a comprehensive dataset that tracks worker activity across platforms, we estimate a structural model to analyze how workers optimize their earnings and respond to earnings-based incentives to switch platforms or locations. We show that workers are highly heterogeneous in their preferences and find multihoming especially costly, both in absolute terms and relative to the cost of repositioning. Through counterfactual simulations, we show that firms and regulators can substantially improve system efficiency by enabling workers to freely multihome: workers' hourly earnings increase by 2.0% and service levels increase by 53.1%. In contrast, the existing equilibrium is similar to a system without multihoming, in which hourly earnings increase by 1.3% but service capacity decreases by 4.1%. Additionally, we show that policies to limit traffic congestion by increasing travel costs should include incentives to ensure that workers remain able to efficiently reposition. An increase to repositioning costs by $1 per mile increases hourly earnings by 2.3% but substantially decreases service capacity by 29.6%.

Bio: Daniel Chen is a PhD candidate at Wharton OID and an incoming Assistant Professor of Business Analytics at the Boston College Carroll School of Management. His research studies operations strategy on online platforms.

Title:  Watermarking and Uncertainty Quantification for LLMs

Abstract: In this talk, I will present two of our recent works on large language models (LLMs), hopefully as some preliminary efforts and examples of how OR/MS researchers can contribute to the literature of LLMs.

For the first one, we study the problem of watermarking LLMs, where a watermark for LLMs is a hidden pattern injected to the generated text that is imperceptible to humans but can be algorithmically detected as generated by AI. We consider the trade-off between model distortion and detection ability, and formulate the watermarking procedure as a constrained optimization problem based on the green-red algorithm of Kirchenbauer et al. (2023). We show that the optimal solution to the optimization problem enjoys a nice analytical property. This analytical structure renders the formulation fall under the framework of online optimization with constraints where the existing online resource allocation algorithms in OR/MS literature can be naturally applied. These developments also shed light on the properness of model distortion metrics and the repetition phenomenon for the watermarking algorithms.

 For the second one, we consider the problem of quantifying the uncertainty for the LLMs' generated responses. The goal of uncertainty quantification for LLMs is to compute and output an uncertainty score for every response generated by the LLM (to represent its confidence of the response). The existing approaches for the problem are largely heuristic and unsupervised. We propose a simple supervised uncertainty calibration method that (i) takes the advantage of the labeled NLP datasets of question answering, multiple choices, machine translation etc. and (ii) extracts the uncertainty information contained in the hidden-layer activations (of the LLM) when generating the response. We illustrate how the uncertainty quantification problem for the LLMs differs from that for classic machine learning/deep learning models. The numerical experiments show the hidden-layer activations do contain uncertainty information and the learned uncertainty model is transferable across different NLP tasks.

Bio: Xiaocheng Li is an assistant professor at the Analytics, Operations and Marketing group of Imperial College Business School, Imperial College London. Prior to that, he received his Ph.D. from Department of Management Science and Engineering at Stanford University in 2020 and B.S. degree from School of Mathematical Sciences at Peking University in 2014. His research interests cover a wide range of topics in operations research and management science, including optimization, revenue management, dynamic programming, and machine learning applications. Recently, he also works on topics related to machine learning and deep learning.

Title:  Food Subsidies and Substitution: Experimental Evidence from Indian Micro Retail Panel Data

Abstract: What foods do the underprivileged buy? How do in-kind food subsidies affect their food shopping? We installed point-of-sale scanners at 39 food vendors in a Mumbai settlement. Based on 768,074 food transactions made by 23,717 consumers, snacks, sugar and soft drinks represent 12.5 percent of spending, while rice and wheat combined are 13.8 percent. We randomized 1,255 settlement residents into a weekly rice-and-wheat subsidy disbursed via a government-like subsidy store – or none, for six weeks. Treated subjects substituted away from packaged snacks spending by 20 percent more than untreated subjects and increased complementary spices and condiments spending by over 30 percent more, with strongest effects for those in households with children. Overall food spending, calories, and nutrients from purchased foods remained unchanged, indicating that the subsidy likely increased net nutrient consumption.

Bio: Alp is an Assistant Professor at the Operations, Information and Decision department at Wharton. He received his PhD from London Business School. His research uses field experiments and data-driven analytics methods to propose new ways of addressing policy questions on global poverty alleviation. He is interested in studying and advancing technology-enabled interventions that targets social and environmental impact.

Title:  Labor planning for last-mile deliveries

Abstract: Labor planning for last-mile delivery is the process of planning the number of associates (drivers) required each week to deliver all the expected volume for a pre determined time horizon, with the ability to adjust decisions over time under guardrail restrictions. Several challenges arise in the implementation of the labor planning for last-mile delivery.

The first challenge comes from volume uncertainty. In practice, the number of assigned associates per week must be determined several weeks in advance. While there is some room to adjust the number of associates over time, these adjustments are costly and limited. Unfortunately, the exact volume of shipments to deliver is uncertain, with a known distribution. Hence, the planner must consider the costs of over planning and under planning, as well as the volume distribution, when planning the number of associates. The assigned associates are paid regardless of the volume realization, and hence, planning for a large number of associates may result in unutilized paid associates. On the other hand, under planning may result in volume that cannot be distributed. In this work, we assume that in these cases the remaining volume will be distributed using a third-party delivery, with higher delivery costs per package.

The second challenge concerning the labor planning for the last-mile delivery problem comes from the inter-week dependency. Several factors bind the number of associates in consecutive weeks, such as ramp-up time (which implies that new associates will not be able to deliver as many shipments as tenured associates), natural attrition, and restrictions that prevent planners from laying off associates. These inter-week dependencies mean that the planner must consider the dynamic impact of labor planning decisions.

This work explores labor planning for last-mile delivery and addresses the above-mentioned challenges. We propose to model this problem using a sequential newsvendor model. We formulate an optimal Dynamic Programming (DP) formulation and propose a relaxation of the problem that can be solved in tractable time. In addition, we analyze the impact of the relaxation on the optimality of the resulting labor plan to illustrate that the proposed approach can find a near-optimal plan.

Bio: Tamar is an Assistant Professor at the Operations & Logistics Division at the University of British Columbia, Sauder School of Business. Her research is focused on real world problems in the field of Retail Operations Management, including retail logistics, dynamic pricing, and consumer behavior learning and modeling. Tamar’s research is influenced by her many industrial collaborators, including, Amazon, Zara, Oracle Retail Group, Marvell Technology Group, and Teva Pharmaceuticals.

Tamar received an M.Sc. degree with summa cum laude in Information Management Engineering, and a B.Sc. degree in Industrial Engineering both from the Technion - Israel Institute of Technology, and her Ph.D. in Operations Research from MIT.

Prior to joining UBC Sauder, Tamar spent a year as a postdoctoral research scientist at Modeling and Optimization group at Amazon.

Title:  Invoice Tokenization for Deep-Tier Payables Finance

Abstract: Invoices from tier-1 suppliers to the downstream anchor manufacturer can be tokenized onto a blockchain. The tier-1 suppliers are then able to split and transfer the tokens to their own (tier-2) suppliers, enabling deep-tier suppliers to sell tokens and access financing at more affordable rates based on the anchor manufacturer’s credit rating. This paper investigates how invoice tokenization impacts the multitier supply chain’s decisions and profits under different supply network configurations and contractual forms. Our research yields the following main insights. First, invoice tokenization always benefits the downstream manufacturer and the reliable tier-2 supplier, but can hurt the profits of other participants. If the tier-1 supplier allocates the manufacturer’s order between the two tier-2 suppliers with different reliabilities (i.e., in the Y-shaped supply chain), the sourcing strategy switch brought by invoice tokenization makes the unreliable tier-2 supplier worse off with a zero profit. The tier-1 supplier can also be hurt because tier-1’s leverage over the manufacturer is indirectly weakened. If the manufacturer directly decides how much to source from the two tier-2 suppliers through two different tier-1 suppliers (i.e., in the V-shaped supply chain), invoice tokenization reduces the unreliable tier-2 supplier’s production quantity and even decreases the profit when the market potential is relatively large, but the manufacturer never abandons the unreliable branch. Second, when the market potential is relatively small, the manufacturer is more likely to initiate invoice tokenization in the V-shaped supply chain. The result, however, is reversed if the market potential is relatively high. Third, direct control over the wholesale price may hinder the value of invoice tokenization for a supply chain tier. For instance, the tier-1 pricing might allow the manufacturer to extract more value from invoice tokenization in the V-shaped supply chain because of the intensified price competition among tier-1. Finally, a higher credit risk of the manufacturer can increase the value of invoice tokenization, not only for the manufacturer but also for the whole system.

Bio: Fasheng Xu is an Assistant Professor of Operations and Information Management at School of Business, University of Connecticut (UConn). Fasheng’s general research interests lie in the interface of operations, finance, and economics. In particular, he is interested in studying the economic and social implications of emerging technologies and identifying effective designs and policies for innovative markets and platforms. His recent research has been focused on the following areas: Blockchain/Crypto, FinTech, Supply Chain Finance, Data Privacy, Economics of AI and Foundation Models. His research has been published in leading academic journals such as Management Science and M&SOM. He has received several best paper nominations and awards, meritorious service awards for Management Science and M&SOM, and the 2022 Guttag Junior Faculty Award at Syracuse University.

Prior to joining UConn in 2023, Fasheng was an Assistant Professor of Supply Chain Management at Whitman School of Management, Syracuse University. Fasheng received his Ph.D. in Operations Management from Olin Business School, Washington University in St. Louis and B.S. in Industrial Engineering and Operations Research from Shanghai Jiao Tong University.

Title:  The Enigma of Ticket Exchanges (and Other Reselling Markets)

Abstract: It has long been established in the literature (and observed in practice) that sellers can benefit from allowing consumers to purchase in advance of the date of actual consumption (e.g., concert tickets, sporting events, etc.). Because of this advance purchasing, consumers can find themselves either with a ticket that they no longer want, or without a ticket that they wish to have. In the past, scalpers would facilitate transactions among these consumers, for a fee. Sellers historically disliked those practices and actively worked to prevent them. In fact, we obtain a stark finding: an unfettered and efficient reselling market eliminates all of the benefits of advance selling, which justifies sellers’ historic hostility to reselling. But now ticket exchanges are common, growing, and even embraced by the sellers. What changed? We present a theory that demonstrates reselling is actually beneficial for sellers under one crucial condition - the seller must be able to have some control over the reselling process, thereby allowing the seller to earn something from each transaction (either directly, or more likely, through licensing fees to third-party sellers). The old-fashioned paper ticket did not give such control, but technology now enables electronic tickets, which do. In fact, a seller cannot earn more than what it receives from a properly designed and efficient reselling market (i.e., reselling is an optimal mechanism for the seller). And such a market (i) eliminates the opportunities for speculators (i.e., the seller has no need for scalpers nor should fear them), (ii) can also be beneficial to consumers, and (iii) even incentivize the seller to provide more capacity. In sum, our results explain why the seller’s view towards reselling has shifted dramatically.

Bio: Pnina Feldman is an Associate Professor and a Dean's Research Scholar in the Operations and Technology Management Department at the Questrom School of Business, Boston University, and a visiting Associate Professor at the Darden School of Business, University of Virginia. 

Her research focuses on how digital technologies affects operations strategy with emphasis on services, platforms, pricing, and consumer engagement. She serves as an associate editor at Manufacturing & Service Operations Management, Production & Operations Management, and Service Science and her work has been published in academic journals such as Management Science, Manufacturing & Service Operations Management, Operations Research, and Marketing Science. She received many research, teaching, and service awards.

She received her Ph.D. in Operations Management from the Wharton School of the University of Pennsylvania and holds a bachelor’s and master’s degrees from the Technion. Prior to joining Questrom, she was an assistant professor at the Haas School of Business, UC Berkeley.

Title: Recommendations in High-stakes Settings: Diversity and Monoculture

Abstract: Algorithmic recommendation systems -- historically developed for settings such as movies, songs, and media content -- are now well-integrated into online matching platforms for high-stakes settings such as for labor, education, and dating. With this integration comes a renewed importance on challenges such as diversity (are you showing a diverse set to users) and monoculture (what are the consequences of everyone using the same algorithm). I'll describe some of our work in this space, emphasizing how OR techniques are essential to design more efficient, equitable algorithms for such platforms. Joint work with Kenny Peng and many others. 

Bio: Nikhil Garg is an Assistant Professor of Operations Research and Information Engineering at Cornell Tech as part of the Jacobs Technion-Cornell Institute. He uses algorithms, data science, and mechanism design approaches to study democracy, markets, and societal systems at large. Nikhil has received the INFORMS George Dantzig Dissertation Award, an honorable mention for the ACM SIGecom dissertation award, several other best paper awards, and Forbes 30 under 30 for Science. He received his PhD from Stanford University and has spent considerable time in industry -- most recently, he was the Principal Data Scientist at PredictWise, which provides election analytics for political campaigns. 

Title: Search with Recall and Gaussian Learning: Structural Results and Optimal Index Policy

Abstract: The classic sequential search problem rewards the decision maker with the highest sampled value, minus a cost per sample. If the sampling distribution is unknown, then a Bayesian decision maker faces a complex balance between exploration and exploitation. We solve the stopping problem of sampling from a Normal distribution with unknown mean and unknown variance and a conjugate prior, a longstanding open problem. The optimal stopping region may be empty (it may be optimal to continue the search regardless of the offer one receives, especially at the early stages), or it may consist of one or two bounded intervals. While a single reservation price cannot describe the optimal rule, we do find a standardized reservation rule: stop if and only if the standardized value of the current offer is sufficiently high relative to the standardized search cost. We also introduce the index function, which provides a computationally practical way to implement the standardized stopping rule for any given prior, sampling history, and sampling horizon.

Bio: Sasa Zorc joined the Quantitative Analysis area at UVA Darden as assistant professor after having recently earned his Ph.D. in decision sciences from INSEAD in Singapore. He studies incentives in multi-agent systems such as health care and decentralized matching markets. Methodically, his research relies on stochastic dynamic games, search theory, mechanism design, contract theory as well as data-driven simulations of those systems.

Title: The Impact of Information-Granularity and Prioritization on Patients’ Care Modality Choice

Abstract: Over the past few years, healthcare providers have widely adopted telemedicine consultations. On one hand, telemedicine has the potential of increasing the accessibility of medical appointments. On the other hand, due to the limitation of diagnosis and treatment methods, telemedicine may be insufficient for patients' treatment needs and necessitate subsequent in-person follow-up visits. To better understand this tradeoff, we model the healthcare system as a queueing network providing two types of service: telemedicine and in-person consultations. We assume that an in-person visit guarantees successful treatment, whereas a telemedicine visit may fail to replace in-person care with a probability that is contingent on the patient’s features. We formulate patients' strategic choices between these care modalities as a queueing game, and characterize the game-theoretic equilibrium and the socially optimal patients' choices. We investigate the impact of improving patients' perception of their complexity through predictive analytics during online triage. We find that increasing information granularity maximizes the stability region of the system but may not always be optimal in reducing the average waiting time. This limitation, however, can be overcome by simultaneously deploying a priority rule that induces the social optimum under specific conditions. Finally, leveraging real-world data from a large academic hospital in the United States, we perform a comprehensive case study that encompasses both the development of a prediction model for in-person follow-up needs and the implementation of effective information provision and patient scheduling strategies.

Bio: Yue Hu is an Assistant Professor of Operations, Information &Technology at Stanford Graduate School of Business. Her research lies at the intersection of healthcare operations management and applied probability. With particular focus on scheduling, staffing, and patient-flow management in healthcare delivery systems, she studies how to leverage predictive analytics to guide operational strategies and innovations. In addition to solving practically relevant problems, she conducts research in developing new methodologies for the approximation and control of stochastic systems. Hu’s research has been recognized in a number of competitions, including as the finalist of the 2022 INFORMS Doing Good with Good OR Competition, winner of the 2020 INFORMS APS Best Student Paper Award, finalist of the 2019 INFORMS IBM Best Student Paper Award, and honorable mention in the 2017 INFORMS Undergraduate Operations Research Prize. Hu received her PhD from the Decision, Risk and Operations Division at the Graduate School of Business, Columbia University. Prior to pursuing her PhD, she received a BS from the Department of Industrial Engineering and Management Sciences at Northwestern University.

Title: Aggregating Distributed Energy Resources: Efficiency and Market Power

Abstract: The rapid expansion of distributed energy resources (DERs) is one of the most significant changes to electricity systems around the world. Examples of DERs include solar panels, electric vehicles/storage, thermal storage, combined heat and power plants, etc. Due to the small supply capacities of these DERs, it is impractical for them to participate directly in the wholesale electricity market. In this talk, we discuss the question of how to integrate these DER supplies into the electricity market, with the objective of achieving full market efficiency. Specifically, we study three aggregation models, where there is an aggregator who procures electricity from DERs, and sells them in the wholesale market. In the first aggregation model, a profit-maximizing aggregator announces a differential two-part pricing policy to the DER owners. We show that this model preserves full market efficiency, i.e., the social welfare achieved by the aggregation model is the same as that when DERs participate directly in the wholesale market. In the second aggregation model, the profit-seeking aggregator is forced to impose a uniform two-part pricing policy to prosumers from the same location, and we numerically show the efficiency loss of this model. In the third aggregation model, a uniform two-part pricing policy is applied to DER owners, while the aggregator becomes fully regulated but is guaranteed positive profit. It is shown that this third model again achieves full market efficiency. Furthermore, we show that DER aggregation also leads to a reduction on the market power of conventional generators. DER aggregation via profit-seeking and/or regulated aggregators have been investigated by CAISO and NYISO, among others, and the recent FERC Order No. 2222 paved the way for aggregators to bid in the wholesale market. Our efficient aggregation models may shed light on how DERs should be included in the wholesale electricity market.

Bio: Zuguang is an Assistant Professor of Operations and Decision Technologies at the Paul Merage School of Business, University of California, Irvine. His research interests generally lie in sustainable operations, broadly defined, with a major focus on electricity market design, power/energy systems, and environmental policy, as well as some other areas related to sustainability (such as supply chains). Previously, Zuguang received his Ph.D. (2023) and MBA (2022) degrees from the University of Chicago Booth School of Business, advised by Prof. John R. Birge and Prof. Varun Gupta. Prior to Chicago, he received his B.S. (2015) and M.S. (2017) degrees, from the Department of Electrical and Computer Engineering, University of Illinois Urbana-Champaign.

Title: Robust Auction Design with Support Information

Abstract: A seller wants to sell an item to n buyers. Buyer valuations are drawn i.i.d. from a distribution unknown to the seller; the seller only knows that the support is included in [a,b]. To be robust, the seller chooses a DSIC mechanism that optimizes the worst-case performance relative to the first-best benchmark. Our analysis unifies the regret and the ratio objectives. For these objectives, we derive an optimal mechanism and the corresponding performance in quasi-closed form, as a function of the support information and the number of buyers n. Our analysis reveals three regimes of support information and a new class of robust mechanisms. i.) With ``low'' support information, the optimal mechanism is a second-price auction (SPA) with random reserve, a focal class in earlier literature. ii.) With ``high'' support information, SPAs are strictly suboptimal, and an optimal mechanism belongs to a  class of mechanisms we introduce, which we call pooling auctions (POOL);  whenever the highest value is above a threshold, the mechanism still allocates to the highest bidder, but otherwise the mechanism allocates to a uniformly random buyer, i.e., pools low types. iii.) With ``moderate'' support information, a randomization between SPA and POOL is optimal. 

   We also characterize optimal mechanisms within nested central subclasses of mechanisms: standard mechanisms that only allocate to the highest bidder, SPA with random reserve,  and SPA with no reserve. We show strict separations in terms of performance across classes, implying that deviating from standard mechanisms is necessary for robustness. Lastly, we show that the same results hold under other distribution classes that capture ``positive dependence'', namely: i.i.d., mixture of i.i.d., and exchangeable and affiliated distributions, as well as i.i.d. regular distributions. The talk will be based on the following two papers: Robust Auction Design with Support Information, and On the Robustness of Second-Price Auctions in Prior-Independent Mechanism Design.

Bio: Jerry Anunrojwong is a fourth-year PhD candidate at Columbia Business School, advised by Prof. Omar Besbes and Prof. Santiago R. Balseiro. He is broadly interested in strategic behavior and robustness in market design. His work involves both developing methodological tools for robust auction design, as well as applications to electricity markets, battery storage, and the energy transition. He has been awarded a finalist for the George Nicholson Student Paper Competition in 2022. Before his PhD, he was a data scientist at Agoda, and a research affiliate at MIT and Chulalongkorn University. He graduated from Harvard University with a master's degree in statistics and bachelor's degree in applied mathematics.

Title:  Value of Exclusive Doorstep Delivery in the Last-100-meter Distribution

Abstract: The last-100-meter distribution service provided by self-pickup points in residential communities in China serves as the last e-commerce order-fulfillment leg. With repeated interactions with consumers, it shapes their delivery-service experience and online-shopping choices. We investigate the value of upgrading this service leg exclusively by exploiting a large-scale natural experiment. In the experiment, doorstep delivery is provided exclusively for Alibaba packages by self-pickup points as an additional option, while Non-Alibaba consumers can only self-pick their packages. Leveraging difference-in-differences models and using data of more than 1.5 billion packages, we show that the exclusive doorstep-delivery initiative brought significant sales growth (decline) of Alibaba (Non-Alibaba) and reversed the sales growth trends -- outgrown by 3.09% pre-treatment, Alibaba outgrew its competitors by 3.80% post-treatment. We define novel consumer-choice-driven service-quality metrics and uncover that service-quality improvement (deterioration) drives the sales growth (decline) of Alibaba (Non-Alibaba). We reveal that such a negative spillover effect is due to doorstep-delivery-induced service-capacity competition. We also show a counter-intuitive positive spillover effect possibly due to self-benefiting-driven service-capacity sharing. Finally, we provide managerial insights on treatment heterogeneity. Self-pickup points serve different strategical roles depending on Alibaba fast-moving package share and Non-Alibaba overdue package share. Those with one high and the other low can only serve as sales generators or competitor destroyers; those with both high can kill two birds with one stone; those with both low achieve none of the two goals. Policy-wise, Alibaba needs to customize its business-development effort and subsidy based on the two metrics.

Bio: Jiwen Ge is an associate professor working at the Institute of Supply Chain Analytics at Dongbei University of Technology. He is currently visiting the Rotman School of Management at the University of Toronto. He did his PhD in operations management at Eindhoven University of Technology in the Netherlands, and was a post doctoral research fellow at the Tuck School of Business at Dartmouth College. His research focuses on the retail operations in emerging markets. See his Google site https://sites.google.com/view/jiwenge to know more about him and his research.

Title:  Make it Personal: Standardization and Prosocial Behavior

Abstract: While the impact of standardization vs. customization strategies on financial outcomes, such as cost efficiency and willingness to pay, has been widely studied, there has been limited research on the relationship between these strategies and social outcomes. We investigate the relationship between standardization and prosocial behavior, using the lens of psychological ownership. Psychological ownership occurs when individuals feel that an item associated with a task is ‘theirs’, which leads to increased motivation to perform that task. We argue that standardization decreases prosocial behavior because it reduces psychological ownership. We take advantage of a unique dataset that combines public data and proprietary data from an asymptomatic university viral screening program, which switched partway through the program from using name-associated (i.e., customized) to generic (i.e., standardized) test kits. Using a regression discontinuity in time (RDiT) approach, we find that the standardization of kits is associated with a significant reduction in individuals’ participation in testing, controlling for all relevant exogenous factors. We show that this reduction is permanent, rather than transitory, suggesting a long-lasting impact of a reduction in psychological ownership. We further find that participants’ prosocial behavior in larger groups is less influenced by this standardization, which indicates that group dynamics might partially mitigate the importance of psychological ownership with regards to social outcomes. Finally, we offer a preliminary analysis of the influence of group diversity on the relationship between standardization and prosocial behavior. While ethnic diversity appears to strengthen this negative relationship, gender diversity mitigates it. Overall, these findings highlight the complex interplay between group dynamics and production decisions with regards to social outcomes. Beyond developing a framework that extends our understanding of the relationship between standardization and social outcomes, our results have important practical implications for both managers and policymakers. For managers, we highlight the unintended social consequences of standardization, while shedding light on potential mitigating mechanisms. For policymakers, we provide insight into designing healthcare screening and other prosocial services to engage users. 

Bio: Rob Glew is an Assistant Professor (non-tenure track) at Desautels Faculty of Management, McGill University and a researcher at the Institute for Manufacturing, University of Cambridge, where he obtained his PhD after 3 years in 2023. Starting at McGill Desautels age 26, Rob is the youngest member of faculty and plans to go on the academic job market for a tenure-track position in 2024. His research focuses on the relationship between operations management and the social good, applying both empirical and analytical methods. This interest has led him to study operations in a variety of settings, from volunteers responding to the COVID-19 pandemic, to implementing technology to reduce food waste in grocery supply chains. Based in Montreal, Rob is now working on several projects intended to improve our understanding of how operations management can contribute to a healthy and equitable society. More information about these ongoing works can be found at www.robglew.com. 

Title:  Subsidy and Taxation in All-Pay Auctions under Incomplete Information

Abstract: We study all-pay auctions under incomplete information with n contestants who have non-linear cost functions. The designer may award two kinds of subsidy (taxation): one that decreases (increases) each contestant's marginal cost of effort and another that increases (decreases) each contestant's value of winning. The designer's expected payoff is the contestants' expected total effort minus the cost of subsidy or, alternatively, plus the tax payment. We show that when the resource of subsidy (the marginal taxation rate) is relatively small and the cost function is concave (convex), the designer's expected payoff in all-pay auctions with both kinds of subsidy (taxation) is higher than in the same contest without any subsidy (taxation). We then compare both kinds of subsidy and demonstrate that if the resource of subsidy is relatively small and the cost functions are concave (convex), the cost subsidy is better than the prize subsidy for the designer who wishes to maximize his expected payoff.

Bio: Yizhaq Minchuk received his doctorate and master’s degree at Ben-Gurion University, Industrial Engineering and Management Department, specializing in game theory. Yizhaq also holds a B.A. degree in Mathematics from Ben-Gurion University. Yizhaq is a Senior Lecturer at the Department of Industrial Engineering & Management in the Shamoon College of Engineering, Israel. His research interest is Applied Game Theory, focusing on Auction Theory and Contest Theory.

2Title:  Do Mergers and Acquisitions Improve Efficiency: Evidence from Power Plants

Abstract: Using rich data on hourly physical productivity and five thousand ownership changes from US power plants, we study the effects of mergers and acquisitions on efficiency and provide evidence on the mechanisms. We find that acquired plants experience an average of 4% efficiency increase five to eight months after acquisition. Three-quarters of this efficiency gain is explained by increased productive efficiency; the rest comes from dynamic efficiency at the plant level and allocative efficiency at the portfolio level. Our findings suggest that acquisitions reallocate assets to more productive uses; we find that high-productivity firms buy underperforming assets from low-productivity firms and make the acquired assets almost as productive as their existing assets after acquisition. Finally, investigating the mechanism, the evidence suggests that acquired plants achieve higher efficiency through low-cost operational improvements rather than high-cost capital investments.

Bio: Prior to coming to Stanford, Omer completed his Ph.D. in Economics at MIT in 2022 and got his bachelor's degree in Economics from Bilkent University in 2014.

His research focuses on the transition of the energy sector toward a decarbonized and sustainable future. In his research, he utilizes large datasets by using game-theoretical modeling to have practical policy suggestions. 

Title:  30 Million Canvas Records Reveal Widespread Sequential Bias and System-design Induced Surname Initial Disparity in Grading (joint work with Jiaxin Pei and Jun Li)

Abstract: The widespread adoption of learning management systems in education institutions has yielded numerous benefits for teaching staff but also introduced the risk of unequal treatment towards students. We present an analysis of over 30 million Canvas grading records from a large public university, revealing a significant bias in sequential grading tasks. We find that assignments graded later in the sequence tend to (1) receive lower scores, (2) receive comments that are notably more negative and less polite, and (3) exhibit lower grading quality measured by post-grade complaints from students. Furthermore, we show that the system design of Canvas, which pre-orders submissions by student surnames, transforms the sequential bias into a significant disadvantage for students with alphabetically lower-ranked surname initials. These students consistently receive lower scores, more negative and impolite comments, and raise more post-grade complaints because of their disadvantaged position in the grading sequence. This surname initial disparity is observed across a wide range of subjects. For platforms and education institutions, the system-induced surname grading disparity can be mitigated by randomizing student submissions in grading tasks.

Bio: Zhihan (Helen) Wang is a fifth-year PhD candidate at Ross School of Business, University of Michigan, advised by Prof. Damian Beil and Prof. Jun Li. Her research broadly focuses on using empirical and experimental methods to address managerial and operational issues in a variety of education settings, such as EdTech, higher education, K-12, and childcare. The goal of her research is to improve the welfare of the younger generation, families, and educators through more effective and inclusive delivery of care and education services. Her work has been published in MSOM and selected by competitions such as Informs IBM student paper competition (2022), POMS CSOM student paper competition (2023), and ACM EAAMO best student paper (2023). Before she came to UM, she earned her Bachelor's degree in Economics with a minor in Data Science from Fudan University, China.

Title:  On the supply of Autonomous Vehicles in Platforms

Abstract: The likely arrival of autonomous vehicle (AV) technology in the near future has the potential to fundamentally change the transportation landscape. Due to the high cost of AV hardware, the most likely path to widespread use of AVs is via open platforms that can sustain high-utilization, outsource the high capital burden, and complement the network with human drivers joining as individual contractors (ICs). In this paper, we study a supply chain game between a platform, an outside AV supplier, and ICs. We show that such a setting is subject to a risk of AV underutilization because of the need to maintain the ICs' utilization sufficiently high to ensure ICs remain engaged. We show that in a decentralized supply chain, this can have a very significant negative effect on the supply chain efficiency, with an unbounded profit loss. We then study potential contracting solutions and show the benefits of contracts that entail an AV utilization commitment by the platform. 

Bio: Daniel Freund is an Assistant Professor of Operations Management at the MIT Sloan School of Management. Before joining MIT, Daniel received his PhD in applied mathematics from Cornell and was a postdoc at Lyft. His current research applies market design, stochastic modeling, and optimization to problems in transportation, online platforms, and revenue management among others. It has been recognized with the MSOM Best Paper in OR Award (2023), the Applied Probability Society Best Publication Award (2021), the George B. Dantzig Dissertation Award (2018), the Daniel H. Wagner Prize for Excellence in Operations Research and Analytics (2018), and a Best Paper award at the ACM SIGCAS COMPASS conference (2018).

Title: Incentivizing Resource Pooling

Abstract: Resource pooling improves system efficiency drastically in large stochastic systems, but its effective implementation in decentralized systems remains relatively underexplored. This paper studies how to incentivize resource pooling when agents are self-interested, and their states are private information. Our primary motivation is applications in the design of decentralized computing markets, among others. We study a standard multi-server queueing model where each server is associated with an M/M/1 queue and aims to minimize its time-average job holding and processing costs. We design a simple token-based mechanism where servers can earn tokens by offering help and spend tokens to request help from other servers, all in their self-interest. The mechanism induces a complex game among servers. We employ the fluid mean-field equilibrium (FMFE) concept to analyze the system, combining mean-field approximation with fluid relaxation. This framework enables us to derive a closed-form characterization of servers' FMFE strategies. We show that these FMFE strategies approximate well the servers' rational behavior. We leverage this framework to optimize the design of the mechanism and present our main results: As the number of servers increases, the proposed mechanism incentivizes complete resource pooling---i.e., the system dynamics and performance under our mechanism match those under centralized control. We also extend our mechanism to settings with heterogeneous servers, and we show that our mechanism obtains near-optimal performance. (Joint work with Yilun Chen (CUHK-SZ) and Pengyu Qian (Purdue). Paper is available at https://ssrn.com/abstract=4586771.)

Bio: Mr. Chen Chen is an Assistant Professor of Operations and Business Analytics at New York University Shanghai. He received his Ph.D. from Fuqua School of Business, Duke University. Prior to joining NYU Shanghai, He was a Postdoctoral Researcher in the Operations Management area at Booth School of Business, University of Chicago. He is broadly interested in the design and analysis of simple and efficient algorithms and mechanisms to improve the operations of model marketplaces. His work has been awarded first place in the inaugural 2019 INFORMS Revenue Management and Pricing Jeff McGill Student Paper Prize. He interned at Uber's Marketplace Optimization group in the summer of 2019, working on matching problems.

Title: Customers’ Multihoming Behavior in Ride-Hailing: Empirical Evidence from Uber and Lyft

Abstract: Are customers loyal to a ride-hailing platform or do they see this service as a commodity and multihome (i.e., check several platforms before booking a ride)? Using a large panel dataset on ride-hailing transactions, we investigate to what extent customers multihome. Our dataset offers a unique opportunity to study this question as we observe the repeated choices of riders for both Uber and Lyft. Our dataset comprises more than 1.4 million rides completed by 162 thousand riders in NYC in 2018. We develop a comprehensive structural model that incorporates both operational (price and waiting time) and behavioral factors (e.g., platform stickiness) to explain riders' choices. Our model also accounts for the dynamic interactions between customers and platforms by assuming that riders update their beliefs on price and waiting time in a Bayesian fashion. Finally, the riders' propensity to multihome is modeled by incorporating the consideration set formation of customers into our framework. We find that riders' choices are not fully explained by operational factors, hence indicating that customers view the platforms as differentiated service providers. While 83.4% of riders took rides with a single platform, our model shows that even the remaining 16.6%, who used both Uber and Lyft at least once, considered both platforms only 43.4% of the time. It is crucial for ride-hailing platforms to capture this single (or multi)-homing behavior while designing promotions. Specifically, personalized promotions may be ineffective if the platform is not part of the customer’s consideration set. Our results show that targeting customers earlier in their lifecycle can enhance the platform’s market share by 77.56% more than their current promotional strategy. We also find that targeting customers with low search friction results in a 24.78% increase in market share relative to targeting customers with high search friction.

Bio: Sandeep Chitla is a fifth-year Ph.D. candidate in the Operations Management department at NYU Stern School of Business, advised by Prof. Srikanth Jagabathula. His research broadly focuses on using data and empirical methods to study operational questions in online marketplaces and platforms, with a particular emphasis on structural estimation and machine learning techniques. His work involves both developing novel methodology and applying it to real-world data. Before embarking on his PhD journey, Sandeep graduated from IIT Madras with a bachelor’s degree in Mechanical Engineering. He also worked as a research associate with Professor Milind Sohoni at the Indian School of Business (ISB).

Title: Optimal Low-Rank Matrix Completion: Semidefinite Relaxations and Eigenvector Disjunctions 

Abstract: Low-rank matrix completion consists of computing a matrix of minimal complexity that recovers a given set of observations as accurately as possible and has numerous applications, including product recommendation. Unfortunately, existing methods for solving low-rank matrix completion are heuristics that typically identify high-quality solutions, but without any optimality guarantees. We reexamine matrix completion with an optimality-oriented eye, by reformulating low-rank problems as convex problems over the non-convex set of projection matrices and implementing a disjunctive branch-and-bound scheme that solves them to certifiable optimality. Further, we derive a novel and often tight class of convex relaxations by decomposing a low-rank matrix as a sum of rank-one matrices and incentivizing, via a Shor relaxation, that each two-by-two minor in each rank-one matrix has determinant zero. Across a suite of numerical experiments, our new convex relaxations decrease the optimality gap by two orders of magnitude compared to existing attempts. Moreover, we showcase the performance of our disjunctive branch-and-bound scheme and demonstrate that it solves matrix completion problems over 150x150 matrices to certifiable optimality in hours, constituting an order-of-magnitude improvement on the state-of-the-art. This is joint work with Jean Pauphilet (LBS), Sean Lo (MIT), and Dimitris Bertsimas (MIT) and a preprint is available here: https://arxiv.org/abs/2305.12292. 

Bio: Ryan Cory-Wright is an Assistant Professor in the Analytics and Operations Group at Imperial College Business School (ICBS), affiliated with Imperial-X. Prior to joining Imperial, he was a Herman Goldstine Postdoctoral Fellow at IBM Research in Cambridge, MA. His research follows two main threads. First, developing interpretable algorithms with optimality guarantees to address problems in optimization, machine learning, and statistics. Second, using optimization to facilitate renewable energy integration within electricity markets and other forms of decarbonization. Ryan received his PhD in Operations Research from MIT, where he was advised by Dimitris Bertsimas. Before coming to MIT, he received a BE in Engineering Science from the University of Auckland.

Title: Intertemporal Pricing with Resellers: An Empirical Study of Product Drops

Abstract: Product drops occur when a retailer releases a limited-edition product line on a specific date for a short period of time. Due to limited inventory of the product and the short sales horizon, a resale market emerges where products may resell at higher prices once the firm stocks out. The firm faces a central trade-off: pricing too high may lead to a longer stock-out time and future markdowns; pricing too low may lead to lost revenue with large markups in the resale market. Any firm in this position may ask, "How do resellers impact my profit?" To answer this question we build a dynamic structural model that can be used as a framework for managers to estimate the preferences of consumers that engage in strategic behavior to resell as well as capture the business requirement to sell inventory in a timely fashion. We estimate our model using a unique data set from a retailer of baby clothing with weekly product drops, where customers engage in a resale marketplace through Facebook groups. We find that ignoring the resale market in pricing reduces firm profit by 8.1%.

Bio: Dayton Steele is an Assistant Professor in the Supply Chain and Operations Department at the Carlson School of Management, University of Minnesota. His research uses structural estimation and field experiments as techniques to study new innovative processes at retail chains. Dayton's works span a variety of retail contexts including apparel, e-Commerce, consumer electronics, and automobile spare parts. He received his PhD in Operations at the University of North Carolina at Chapel Hill, and completed his undergraduate studies at the University of Richmond, with a BS in Mathematical Economics. Prior to academia, Dayton worked in management consulting in Richmond, VA, leading the data analytics department.

Title: Restaurant Density and Delivery Efficiency in Food-delivery Platforms 

Abstract: With the rapid growth of food delivery platforms, understanding their operational implications for partnered restaurants is crucial. This paper investigates how an increase in restaurant density on a platform affects individual partnered restaurants in terms of delivery efficiency, sales, revenues, and customers' spending behaviors. Three main externalities are examined: customer expansion, cannibalization, and delivery pooling. Using a queuing model, the study proposes testable hypotheses about the operational impacts of these externalities. Empirical analysis, based on data from a major Chinese food-delivery platform, shows that as restaurant density increases, delivery efficiency for partnered restaurants improves, leading to increased sales and revenues. Furthermore, customers are found to opt for smaller order sizes in denser areas. Delivery pooling emerges as the dominant positive externality, especially benefiting high-traffic restaurants, and is crucial in driving these results. The findings underscore the importance of optimizing restaurant density for platforms, suggesting a focus on maintaining high restaurant density areas to maximize delivery efficiency. This can foster a scenario beneficial for platforms, restaurant partners, and customers. The insights provide guidance for platform expansion strategies and the operational performance of partnered restaurants.

Bio: Wenchang Zhang is an Assistant Professor of Operations & Decision Technologies at Indiana University's Kelley School of Business. He earned his Ph.D. in Operations Management from the University of Maryland's Robert H. Smith School of Business, his M.A. in Statistics from the University of California, Berkeley, and his B.S. in Mathematics and Physics from Beijing's Tsinghua University. Zhang uses econometrics and modeling methods to study operational challenges in online marketplaces and platforms, including market thickness management, information design, and expansion strategies. His work has been published in Management Science and Production and Operations Management.

Title: Constant Approximation for Network Revenue Management with Markovian-correlated Customer Arrivals

Abstract: The Network Revenue Management (NRM) problem is a well-known challenge in dynamic decision-making under uncertainty. In this problem, fixed resources must be allocated to serve customers over a finite horizon, while customers arrive according to a stochastic process. The typical NRM model assumes that customer arrivals are independent over time. However, in this paper, we explore a more general setting where customer arrivals over different periods can be correlated. We propose a model that assumes the existence of a system state, which determines customer arrivals for the current period. This system state evolves over time according to a time-inhomogeneous Markov chain. We show our model can be used to represent correlation in various settings.

To solve the NRM problem under our correlated model, we derive a new linear programming (LP) approximation of the optimal policy. Our approximation provides an upper bound on the total expected value collected by the optimal policy. We use our LP to develop a new bid price policy, which computes bid prices for each system state and time period in a backward induction manner. The decision is then made by comparing the reward of the customer against the associated bid prices. Our policy guarantees to collect at least $1/(1+L)$ fraction of the total reward collected by the optimal policy, where $L$ denotes the maximum number of resources required by a customer.

In summary, our work presents a Markovian model for correlated customer arrivals in the NRM problem and provides a new LP approximation for solving the problem under this model. We derive a new bid price policy and provides a theoretical guarantee of the performance of the policy.

Bio: Jiashuo Jiang is an assistant professor at Industrial Engineering and Decision Analytics at HKUST. He got his PhD degree in Operations from NYU Stern School of Business in 2022 and obtained his bachelor's degree in Mathematics from Peking University in 2017. His research focuses on dynamic decision making and data driven decision making under uncertainty, with applications in supply chain management, revenue management, inventory management, online advertising, and so on. His work has been recognized as finalists for Informs RMP and Nicholson student paper competitions, under the supervision of Prof. Jiawei Zhang and Prof. Will Ma.

Title: Learning to rank under strategic manipulation in small and large markets

Abstract: We consider a dynamic learning and ranking problem of a digital platform. Uninformed of the products' intrinsic qualities, the platform strives to design a ranking rule that learns from historical traffic data while accounting for potential manipulation by sellers through "brushing" activities, such as fake orders or sales. How does the ranking manipulation disrupt ranking efficiency under various market conditions? Are there effective yet simple ranking algorithms to combat ranking manipulation? 

Under an Experiment-Then-Commit (ETC) policy framework, we formulate an $N$-player-$T$-period dynamic "brushing war" game for the sellers. We provide a (static) budget-competition equilibrium characterization and study its limiting behavior when $T$ is large. For a small market with two sellers, we show the nonexistence of pure strategy equilibria and identify a mixed-strategy equilibrium, shedding light on the possibility of efficiency loss. For a large market, we formulate a novel non-atomic game with a continuum of sellers as a limiting case where $N$ is large. We characterize a "self-reinforcing" market equilibrium, where the seller's brushing amount increases in the product's quality. In other words, the sellers' strategic responses "reinforce" complete learning of the platform.

Bio: Yifan Feng is an Assistant Professor at NUS Business School's Department of Analytics and Operations (DAO). With an interest in the intersection of artificial intelligence, operations management, and optimization, he tackles complex e-commerce and marketplace challenges, focusing on information acquisition, experimentation, and demand fulfillment.

Yifan's work has been published or accepted in renowned business journals like Operations Research and Management Science and AI conferences such as NeurIPS, ICML, and EC. He has also secured over one million USD in funding as either Principal or Co-Principal Investigator (PI/Co-PI) from various public and private sources. Before joining NUS, Yifan obtained his PhD in Management Science/Operations Management from the University of Chicago Booth School of Business.

Title: How Competition Shapes Information in Auctions

Abstract: We consider auctions where buyers can acquire costly information about their valuations and those of others, and investigate how competition between buyers shapes their learning incentives. In equilibrium, buyers find it cost-efficient to acquire some information about their competitors so as to only learn their valuations when they have a fair chance of winning. We show that such learning incentives make competition between buyers less effective: losing buyers often fail to learn their valuations precisely and, as a result, compete less aggressively for the good. This depresses revenue, which remains bounded away from what the standard model with exogenous information predicts, even when information costs are negligible. Finally, we examine the implications or auction design. First, setting an optimal reserve price is more valuable than attracting an extra buyer, which contrasts with the seminal result of Bulow and Klemperer (1996). Second, the seller can incentivize buyers to learn their valuations, hence restoring effective competition, by maintaining uncertainty over the set of auction participants.

Bio: Agathe Pernoud is a postdoctoral researcher at the Becker Friedman Institute, University of Chicago. In July 2025, she will join the Booth School of Business as an Assistant Professor of Economics. She works on microeconomic theory, with a particular focus on market design, information economics, and the study of financial networks. She completed her Ph.D. at Stanford University in 2023. 

Title: Time-Dependent Shipment Options and Shipment Fees for E-Commerce Fulfillment Centers

Abstract: Fulfilling online orders faster becomes more and more important. To that end, e-commerce companies increasingly offer same-day shipment services. However, companies may overpromise their shipment services when the offered shipment options and corresponding fees are not aligned with the ability to fulfill these orders, leading to delayed orders and customer dissatisfaction. Indeed, fulfilment centers responsible for collecting and shipping online orders may not be able to hand over the orders to the parcel delivery company upon the agreed deadlines. Thus, the offered shipment options and corresponding fees should be adapted based on the time remaining for the fulfillment center to collect and ship these orders. We build a parsimonious model of a stochastic e-commerce fulfillment center offering time-dependent shipment options and corresponding fees to utility-maximizing customers that arrive according to a Poisson process. For this time-dependent policy, we present an exact analysis based on the underlying periodic Markov chain as well as monotonicity results for the optimal time-dependent shipment fee structure. We then propose a simple time-dependent shipment policy with an attractive two-level fee structure, and study it alongside two benchmarks that are prevalent in practice. Both benchmarks rely on static, time-independent fees for the offered shipment options, but differ in whether they stop offering same-day shipment after a certain cutoff time. Our numerical analysis indicates that including such a cutoff point under static shipment fees increases profits significantly, and that moving to time-dependent shipment fees increases profits by another substantial amount. This is a joint work with Melvin Drent, Ivo Adan and Willem van Jaarsveld.

Bio: 

Dr. Uta Mohring is a postdoctoral researcher at the Rotman School of Management. Her research primarily applies stochastic modeling techniques to guide decision-making in various applications in logistics and transportation, such as online retailing, warehouse operations and shared mobility. She received her doctoral degree from the Karlsruhe Institute of Technology, Germany, where she also earned her bachelor's and master’s degree in Industrial Engineering and Management.

Title: Individualized Dynamic Patient Monitoring Under Alarm Fatigue

Abstract: Hospitals are inundated with alarms, many of which turn out to be false. Such a high frequency of false alarms results in "alarm fatigue," where clinicians become desensitized, potentially overlooking genuine threats. In this research, we've conceptualized a partially observable Markov decision process model that recommends dynamic, patient-specific alarms. This model uniquely incorporates a "cry-wolf" feedback loop, which is characterized by repeated false alarms. One of the significant highlights of our model is its ability to account for patient heterogeneity concerning safety limits for vital signs. Furthermore, it adeptly performs Bayesian updates throughout a patient's hospital stay to continuously learn about a patient’s safety parameters. Through our structural analysis of the optimal policy and a subsequent numerical case study, based on real-world clinical data from an intensive care unit, we unearthed that our model drastically outperforms existing methodologies, substantially diminishing the risk of patient harm.

Bio: Hossein Piri is an Assistant Professor of  Operations and Supply Chain Management at Haskayne School of Business, University of Calgary. He obtained his Ph.D. degree in Business Administration at the University of British Columbia

He is interested in algorithm design for health care and pricing applications.

Hossein's work has been published in Operations Research and selected for competitions such as Medical Decision Making Society student prize competition (2020), and the Canadian Operational Research Society oral presentation competition (2021).

Title: Data-driven Clustering and Feature-based Retail Electricity Pricing with Smart Meters

Abstract: We consider an electric utility company that serves retail electricity customers over a discrete-time horizon. In each period, the company observes the customers' consumption as well as high-dimensional features on customer characteristics and exogenous factors. A distinctive element of our work is that these features exhibit three types of heterogeneity---over time, customers, or both. Based on the consumption and feature observations, the company can dynamically adjust the retail electricity price at the customer level. The consumption depends on the features: there is an underlying structure of clusters in the feature space, and the relationship between consumption and features is different in each cluster. Initially, the company knows neither the underlying cluster structure nor the corresponding consumption models. We design a data-driven policy of joint spectral clustering and feature-based pricing and show that our policy achieves near-optimal performance, i.e., its average regret converges to zero at the fastest achievable rate. This work is the first to theoretically analyze joint clustering and feature-based pricing with different types of feature heterogeneity. Our case study based on real-life smart meter data from Texas illustrates that our policy increases company profits by more than 200% over a three-month period relative to the company policy and is robust to various forms of model misspecification. 

Bio: Yuexing Li is an Assistant Professor of Operations Management and Business Analytics at Johns Hopkins Carey Business School. He obtained his Ph.D. degree in Business Administration in the field of Operations Management at the Fuqua School of Business, Duke University. He is broadly interested in designing and analyzing data-driven algorithms to facilitate decision making under uncertainty. His work advances the theoretical foundation of various data-driven analytics and significantly improves the performance in real applications across multiple industries, including grocery retailing, energy, and digital platforms.

Title: Racial and gender biases in customer satisfaction surveys: Evidence from a restaurant chain 

Abstract: Despite the passage of the Civil Rights Act and other anti-discrimination legislation, racial and gender inequalities are ubiquitous in the workplace. While previous studies have mainly focused on employer discrimination as a factor in these inequalities, little is known about the role of customers in perpetuating such biases, especially in occupations that are female-dominated and racially diverse. We fill this gap by exploring whether, and how, customers discriminate against service workers based on the workers' race and gender. Using a data set of  1,444,044  transactions and  257,656 customer satisfaction surveys from a full-service casual-dining restaurant chain in the U.S., we study racial and gender biases in customer rating of restaurant servers, an occupation where women hold historical majority and racial minorities have a strong presence. We find that customer ratings are biased against racial minority servers, and, interestingly, that customer ratings are biased against female servers despite their majority status in this occupation. We further show that racial biases diminish as the uncertainty about the servers' ability decreases, while gender biases may even increase. These results along with the discrimination theories in the economic and sociology literature suggest that statistical discrimination is the primary driver for racial biases, while status-based discrimination is likely to be the main driver for gender biases. Given the different underlying mechanisms, we propose tailored strategies to mitigate customer racial and gender biases. 

Bio: Masoud Kamalahmadi is an Assistant Professor at the Department of Management Science in University of Miami's Herbert Business School. He also holds a secondary appointment in the Department of Health Management and Policy. Masoud’s research focuses on understanding the dynamics of work within service systems. He investigates how various factors within the work environment, such as workload, shifts, modality, and more, influence the performance of service workers and develop data-driven models to improve productivity. 

Title: The Impact of Historical Workload on Nurses’ Perceived Workload

Abstract: Recent and ongoing nursing shortages have highlighted the valuable and skilled work that nurses provide around the clock in hospital inpatient care. Intense and sustained high nursing workload has been linked to nurse burnout and patient safety concerns, necessitating targeted approaches to better managing nursing workload. In this work, we take an empirical approach to understanding the effect of historical workload on nurses’ perceived workload. We leverage a unique dataset that records detailed patient-to-nurse assignment information, an order-based workload measure, and a clinically perceived workload measure for each patient during each shift. We also address several identification challenges, including endogeneity, missing values, and measurement errors. Our estimation results show that one level of increase in historical order-based workload can lead to a 0.629 increase in the discrepancy between the clinically perceived workload and the order-based workload. Based on the temporal effect of nursing workload, we design an integer program-based patient-to-nurse assignment policy that achieves a more balanced workload over time while maintaining a high level of continuity of care.

Bio: Yi Chen is currently an assistant professor at Hong Kong University of Science and Technology, department of Industrial Engineering and Decision Analytics. Prior to that, he obtained Ph.D. degree in industrial engineering and management science from Northwestern University in 2021. His research interests include data-driven decision-making and empirical research, with their applications in service system operations and healthcare.

Title: Algorithmic Assortment Curation: An Empirical Study of Buybox in Online Marketplaces (joint work with Santiago Gallino, Antonio Moreno) 

Abstract: Online marketplaces have revolutionized the way global sales take place, providing a platform for millions of buyers and sellers to connect. While the presence of numerous third-party sellers attracts customers to the platform, it also leads to a proliferation of listings for each product, making it challenging for customers to choose between the available options. To tackle this, online marketplaces utilize algorithmic tools to curate the presentation of different listings of a product to customers. This paper focuses on the Buybox, an algorithmic tool that chooses and presents prominently one option as the default one to customers. We assess the Buybox's influence on marketplace dynamics by examining its staggered introduction within a major product category in a leading online marketplace. Our results demonstrate that Buybox's implementation mitigates frictions for both customers and sellers and leads to a significant increase in marketplace conversion and orders. On the customer side, we observe a decrease in search frictions, as evidenced by increased conversion rates and a more pronounced Buybox effect on the mobile channel, which inherently has higher search frictions compared to the desktop. On the seller side, the number of sellers offering a product increases after Buybox's introduction which suggest a decrease in frictions for the sellers. Our analysis reveals that customers obtain lower prices and higher average quality levels when competition for Buybox is intense. We also find that the marketplace becomes more concentrated following Buybox's introduction, representing an unintended consequence that platforms and vendors should manage. Our study contributes to the growing literature on algorithms in platforms by examining how algorithmic curation affects marketplace participants and overall marketplace dynamics.  

Bio: Nil Karacaoglu is an assistant professor of operations management at the Fisher College of Business, Ohio State University. Her research focuses on service operations within new technologies, on-demand services, retail, and data-driven decision-making. Exploring operational challenges in today's digital economy, she collaborates with industry partners to understand the effects of digitalization on customer behavior and develop actionable operational improvements. Nil's work has been published in Operations Research and selected for competitions such as the IBM Service Science Best Student Paper (2017), POMS CBOM Junior Scholar Paper (2019), and INFORMS TIMES Best Working Paper (2021). She holds a Ph.D. in Operations Management from Northwestern University's Kellogg School of Management, a Master of Arts in Economics from Northwestern University, and a Bachelor's and Master's in Industrial Engineering from Bilkent University, where she received Academic Excellence in M.S. Studies award.

Title: Design of resale platforms

Abstract: We study resale platforms, a growing type of online marketplaces in developing countries. These platforms allow their users to supplement their income by selling products to their contacts, who do not typically shop online. Resellers on these platforms make two key decisions: which products to sell, and what profit margin to set. By analyzing data from a major resale platform in India, we find that competition among resellers emerged as more of them joined the platform; this not only reduced their earnings, but also led them to exert less effort when searching for products on the platform. Based on these observations, we develop a model of a resale platform. Our model allows us to analyze how competition affects resellers’ equilibrium outcomes and strategies and generates predictions that align with our empirical findings. Moreover, we provide insights into key design aspects of the platform. We compare two structures that different platforms employ to determine resellers’ margins in practice: a decentralized setting, in which resellers freely choose their margins, and a centralized scenario, in which margins are predetermined by the platform. We find that, while resellers may optimize their margins based on the characteristics of their consumer pools under a decentralized structure, competition pressures resellers' margins downwards, which erodes their earnings and weakens their incentives to exert effort in finding the right product for their buyers. As a result, when competition is present, resellers may benefit from a centralized margin structure. Specifically, even though centralization limits resellers’ ability to optimize over margins, the competitive pressure on margins is eliminated and resellers shift the focus of competition to product curation, i.e., towards exerting more effort in finding attractive products for their customers.

Bio: Ilan Morgenstern is a PhD candidate in Operations, Information, and Technology at Stanford GSB, where he is advised by Daniela Saban and Kostas Bimpikis. His research interests relate to the design and operations of online marketplaces and platforms, with an emphasis on the role of data collection and exploitation, privacy, learning, and social commerce. Before starting his PhD, Ilan graduated from ITAM with bachelor’s degrees in applied mathematics and economics, and worked as a management consultant for McKinsey & Company.

Title: Getting Out of Your Own Way: Introducing Autonomous Vehicles on a Ride-Hailing Platform

Abstract: After autonomous vehicles (AVs) are deployed for ride-hailing platforms but before their costs decrease enough to push human drivers off the road entirely, human drivers will compete for rides with AVs. We consider a ride-hailing platform’s strategy to recruit human drivers while also operating a private fleet of AVs. We formulate and solve a game-theoretic model of a ride-hailing platform with a private AV fleet that also recruits self-interested human drivers. The platform sets the human-driver wage and the AV deployment quantity, and human drivers make strategic joining decisions based on a rational anticipation of their expected earnings. We show that growing its AV fleet too quickly while the AV cost is still relatively high can be a costly mistake for the platform. Doing so can trigger a ``race to the top’’ of increasing wages and increasing AV deployment to the detriment of the platform's profits, effectively preventing it from attracting more than a limited number of human drivers and increasing the cost of attracting a given number. Nonetheless, we prove that the platform can break this feedback loop by optimally tuning its AV fleet size, tempering the competition for rides and achieving a profitable balance of AVs and human drivers. Our findings suggest that even while their costs remain high, AVs can be a valuable tool for ride-hailing platforms, as long as the fleet size is carefully set.

Bio: Andrew Frazelle is an Assistant Professor of Operations Management in the Jindal School of Management (JSOM) at the University of Texas at Dallas. Andrew's research focuses on service operations management, the sharing economy, and customer strategic behavior in queueing systems. His work has been published in Management Science and has been recognized as the winner of the MSOM Service Management SIG Best Paper Award (recognizing the best published work on service management over a three-year period), as well as a finalist in the MSOM Student Paper Competition. Andrew earned his Ph.D. in Decision Sciences from the Fuqua School of Business at Duke University. Prior to his doctoral studies, he earned a B.S. in Industrial Engineering with highest honor from the H. Milton Stewart School of Industrial and Systems Engineering at the Georgia Institute of Technology, where he was awarded the Alpha Pi Mu Scholar Award and the Henry Ford II Scholar Award.

Title: Protecting data privacy in personalized revenue management and decision making

Abstract: The prevalence of e-commerce has made detailed customers' personal information readily accessible to retailers, and this information has been widely used in pricing decisions. When involving personalized information, how to protect the privacy of such information becomes a critical issue in practice. In this paper, we consider a dynamic pricing problem over T time periods with an unknown demand function of posted price and personalized information. At each time the retailer observes an arriving customer's personal information and offers a price. The customer then makes the purchase decision, which will be utilized by the retailer to learn the underlying demand function. Using the fundamental framework of differential privacy from computer science, we develop a privacy-preserving dynamic pricing policy, which tries to maximize the retailer revenue while avoiding information leakage of individual customer's information and purchasing decisions. We also extend the framework to a more stringent privacy protection notion (the local privacy) and nonparametric modeling of demand rates in a follow-up paper.

Bio:  Yining Wang is an Associate Professor of Operations Managemen at Naveen Jindal School of Management, University of Texas at Dallas.   Before joining UTD, he was an assistant professor of information systems and operations management at the Warrington College of Business of University of Florida. Yining Wang received his Ph.D. at Carnegie Mellon University. His main research focus is on the development and analysis of sequential decision-making methods under uncertainty, with emphasis to revenue management applications such as assortment optimization and dynamic pricing. Yining's works have been published in Management Science and Operations Research journals.

Title: Financing capacity under moral hazards: Deterring stealing and price deviations

Abstract: We analyze a newsvendor production problem subject to agency issues. An investor provides funds and contracts a producer for capacity planning. However, the producer may commit cash diversion after receiving the funds, which is unobservable to the investor. We derive the optimal contract under ex-ante and ex-post cash diversion scenarios. Optimal contracts have simple forms and are easy to implement. Furthermore, we provide the optimal production quantity and price. In addition, we consider a scenario where the producer’s pricing strategy is also unobservable and find that the investor can still use a simple debt contract to deter both moral hazards. 

Bio: Mingliu Chen is currently an Associate Research Scientist and Adjunct Assistant Professor in the Department of Industrial Engineering and Operations Research at Columbia University. He utilizes stochastic optimization in matching and incentive management problems. In particular, he is interested in applying analytical and modeling techniques in market and mechanism design issues. He has contributed to the literature on dynamic contract designs and developed novel models describing two-sided matching markets. Recently, his research focused on design problems in some emerging markets and the interface between finance and operations. 

Title: Differential privacy via distributionally robust optimization

Abstract: In recent years, differential privacy has emerged as the de facto standard for sharing statistics of datasets while limiting the disclosure of private information about the involved individuals. This is achieved by randomly perturbing the statistics to be published, which in turn leads to a privacy-accuracy trade-off: larger perturbations provide stronger privacy guarantees, but they result in less accurate statistics that offer lower utility to the recipients. Of particular interest are therefore optimal mechanisms that provide the highest accuracy for a pre-selected level of privacy. To date, work in this area has focused on specifying families of perturbations a priori and subsequently proving their asymptotic and/or best-in-class optimality. In this paper, we develop a class of mechanisms that enjoy non-asymptotic and unconditional optimality guarantees. To this end, we formulate the mechanism design problem as an infinite-dimensional distributionally robust optimization problem. We show that the problem affords a strong dual, and we exploit this duality to develop converging hierarchies of finite-dimensional upper and lower bounding problems. Our upper (primal) bounds correspond to implementable perturbations whose suboptimality can be bounded by our lower (dual) bounds. Both bounding problems can be solved within seconds via cutting plane techniques that exploit the inherent problem structure. Our numerical experiments demonstrate that our perturbations can outperform the previously best results from the literature on artificial as well as standard benchmark problems.

Bio:  Aras is a third year PhD student at Imperial College Business School, advised by Wolfram Wiesemann as a member of the Models and Algorithms for Decision-Making under Uncertainty research group. His research interests are at the intersection of robust optimization, machine learning, and computational privacy. He is specifically interested in developing efficient algorithms to solve hard optimization problems which arise in privacy, machine learning, and business applications. 

Title: Distance-based data-driven robust Markov decision processes

Abstract: In this talk, I will discuss Markov decision processes (MDPs) where the decision-maker does not know the true state-transition probabilities. The decision-maker assumes that they belong to certain ambiguity sets, and chooses actions that maximize the worst-case expected total discounted reward. I will work under a rectangular setup wherein the ambiguity set for the whole problem is a Cartesian product of ambiguity sets for individual state-action pairs. Specifically, the ambiguity set for any state-action pair is a ball --- it includes all probability mass functions (pmfs) within a certain distance from an empirical transition pmf. I will show that the optimal values of the resulting robust MDPs (RMDPs) converge to the optimal value of the true MDP, if the radii of the ambiguity balls vanish to zero as the sample-size diverges to infinity. A rate of convergence will be derived. I will also establish that the robust optimal value provides a lower bound on the value of the robust optimal policy in the true MDP, with a high probability. These results rely on a generalized Pinsker's inequality and a concentration inequality. These two inequalities hold for several well-known distances.

Bio: Archis is a Professor of Industrial & Systems Engineering at the University of Washington in Seattle, where he also held a College of Engineering Endowed Professorship for five years. He joined the University of Washington as an Assistant Professor in 2006 after receiving a PhD in Industrial and Operations Engineering from the University of Michigan in 2006, and an MS in Management Science and Engineering from Stanford in 2003. He completed his undergraduate education at the Indian Institute of Technology, Bombay, India, in 2001. Archis is a recipient of the NSF CAREER award and the award for excellence in teaching operations research from IISE. Archis has also received multiple teaching accolades from the University of Washington. His students have won the Dantzig dissertation award, and the Bonder scholarship in healthcare operations research from INFORMS. Archis has served on the editorial boards of several journals. He was the General Chair of the INFORMS 2019 Annual Meeting, and a Program Co-Chair of the 2021 IISE Annual Conference.

Title: Strategic Choices and Routing Within Service Networks: Modeling and Estimation Using Machine Learning

Abstract: Customer decision-making is difficult to study empirically when choice sets are combinatorially complex. For example, in service networks such as physical marketplaces, shopping centers, and amusement parks, customers can choose from combinatorially many network paths that reflect their motivations to visit the stations they like while dynamically avoiding congestion (e.g., by delaying visiting a temporarily crowded store or attraction). Existing methods suffer serious issues when estimating the preferences regarding consumption and waiting that drive customers' network paths. We address this problem by leveraging seminal developments from discrete optimization and machine learning. We prove that customers need only make a small number of local choice comparisons in order to confirm that their choices are optimal; moreover, we need only analyze such comparisons in order to maximally learn customer preferences from data reporting their network paths. Because neural networks excel at identifying such hidden low-dimensional structure, we leverage them to build estimators that discriminate between, hence identify, customer types from their choices of paths. By consistently and tractably estimating customer preferences, our empirical methods enable deeper analyses of service network or marketplace design, capacity management, and customer targeting. 

Bio: Ken Moon is an Assistant Professor and Claude Marion Endowed Faculty Scholar of Operations, Information and Decisions at the Wharton School, University of Pennsylvania.  He received his Ph.D. at the Stanford Graduate School of Business, his J.D. from the Harvard Law School, and his dual bachelor's in Mathematics and Economics from Stanford University.  His research interests center on workforces and marketplaces, and he applies data science methods spanning structural estimation, machine learning, and econometrics.  His work has received the POMS Operational Excellence Best Paper Award (2020) and through supervising doctoral students the MSOM Best Student Paper Award (2021), the POMS Product Innovation and Technology Best Student Paper Award (2020), and the IBM Service Science Best Student Paper Award (2017).

Title: The Challenges of Deploying an Algorithmic Pricing Tool: Evidence from Airbnb

Abstract: We study the deployment of an algorithmic pricing tool, Smart Pricing (SP), on Airbnb's platform. SP is a machine learning algorithm that uses past data to predict demand and employs proxies that are correlated with the host's marginal cost to set prices for listings. The success of such deployments depends on how well the algorithm performs and how sellers use the tool for their business decisions. Our analyses suggest that adopting SP is associated with higher benefits for hosts who rarely change their prices compared to those who flexibly adjust their prices before adoption. However, hosts who rarely change their prices are surprisingly less likely to adopt SP. To understand how the platform can overcome this challenge, we propose and estimate a dynamic structural model in which hosts make adoption decisions based on their expectations of the algorithm's behavior. Our estimation results identify a gap between the actual performance of the SP algorithm and the host's prior belief about it. Specifically, hosts with a pessimistic prior belief about SP think they will need to manually correct algorithmic prices if they adopt SP, and this belief is disproportionately stronger for hosts with higher adjustment costs, making SP adoption less attractive to them. Our counterfactual simulations indicate that the introduction of SP has had a small positive impact on the average host profit and the total platform revenue. But this boost can be significantly raised if Airbnb educates hosts to correct their beliefs about the SP algorithm. This highlights the need for proper communication of the algorithm's benefits and how it works in order to successfully deploy a machine learning tool. The counterfactual analyses also demonstrate that, since the platform does not fully capture the host's private marginal cost in training the algorithm, using the estimated costs from the structural estimation to re-train the algorithm can significantly increase the profitability of SP for both hosts and the platform. It suggests that combining the results of structural models and machine learning tools can help platforms design better algorithms.

Bio: Mohsen Foroughifar is a PhD candidate in Quantitative Marketing at Rotman School of Management, University of Toronto. He got his M.Sc. in economics and his B.Sc. in electrical engineering from the University of Tehran, Iran. Mohsen's primary research is related to the impacts of new technologies on digital platforms. He explores the challenges of deploying an algorithmic pricing tool on Airbnb and the use of cryptocurrencies on advertising platforms. His doctoral dissertation has been nominated for several awards including Sheth Foundation ISMS Doctoral Dissertation Award, ASA Statistics in Marketing Doctoral Dissertation Research Award, MSI Alden G. Clayton Doctoral Dissertation Proposal Award, and Shankar-Spiegel Dissertation Proposal Award. He will join the Tepper School of Business at Carnegie Mellon University as an Assistant Professor in fall 2023.

Title: Real-time delivery time forecasting in online retailing

Abstract: In online retailing, delivery time is subject to volatility and uncertainty from order arrivals, inventory evolution, warehouse operations, and transport operations. As a result of the complexity of order fulfillment and delivery processes, and the aforementioned uncertainties, predicting the delivery time distribution is a challenging task. In this talk, I present a data-driven framework that predicts parcel delivery times for new customer orders in real-time. This framework suggests that through a more accurate estimation of the delivery time distribution, online retailers can strategically set the promised time to maximize customer satisfaction and boost sales. We built and tested the model on the JD.com data set, and through a simulation study we observed that the proposed forecasting framework could improve the sales volume by 6.1% while controlling the delivery delays. This data-­driven framework reveals the importance of modeling fulfillment operations in delivery time forecasting and integrating the decision-­making problem structure with the forecasting model.

Bio: Nooshin Salari is an assistant professor in Engineering Management at the University of Alberta (UofA). Prior to joining UofA, Dr. Salari was a research associate at Saïd Business School, University of Oxford. She was also an NSERC Postdoctoral Fellow jointly appointed at the department of Industrial Engineering & Operations Research, UC-Berkeley and Rotman School of Management, where she spent two years applying her expertise in data analytics and operations research to improve the delivery service quality in online retailing. Dr. Salari's research interests are in data-driven decision-making, predictive and prescriptive analytics, and large-scale optimization using the theory and the application of Markov decision processes.  

Title: Design of Contingent Free Shipping Policy: The Role of Return Environment

Abstract: A contingent free shipping (CFS) policy offers free shipment of an order only if it satisfies a pre-specified threshold amount. Such a policy may induce customers to pad below-threshold orders to meet the threshold. On the one hand, such padded orders economize the retailer’s logistics cost; on the other hand, it exposes the retailer to enhanced return costs as customers may engage in bubble purchases—orders with spuriously padded items that are later returned. A retailer designing the policy’s terms— threshold and shipping fee—should attempt to balance these competing trade-offs. We study how the selection of these CFS terms is moderated by the retailer’s returns policy and associated customers’ ease-of-return experience. We collaborate with a retailer who switched across multiple CFS policies over time. Our empirical strategy builds on the quasi-natural experiments induced by these switches, and location-based variation in the retailer’s returns policy. We find that in markets with a convenient ease-of-return process, customers pad 15.7% to 23.0% of below-threshold demand, and that 2.9% to 18.5% of these padded orders are bubble purchases. Interestingly, we find that, in markets with modest inconveniences in the returns process, the beneficial order padding is prevalent (13.2% to 20.3%); however, bubble purchases are altogether eliminated. Our counterfactual analysis illustrates that ignoring this moderating role of ease-of-return experience when selecting a CFS policy can result in the selection of suboptimal terms, with a loss of 13.2% in profits. Our study documents a novel determinant of optimal CFS terms: ease-of-return experience. To reflect its impact on the CFS policy’s embedded trade-offs, a manager shall apply the following counterintuitive adjustment; set lenient (resp. stringent) CFS terms when the customer return process is convenient (resp. inconvenient).

Bio: Ashish Kabra is an Assistant Professor at the Robert H. Smith School of Business at the University of Maryland. He conducts empirical and theoretical research using causal inference, structural estimation, game theory, and optimization methods. His research focuses on studying the interplay between platform operations, consumer behavior, and sustainability in the domains such as e-commerce, online B2B platforms, retail stores, and urban transportation. His research work has been published in top journals and has won several prestigious best paper awards. His teaching has been recognized with the prestigious Allen J. Krowe teaching award at the Smith School. Prior to joining Maryland, he completed his graduate studies in Operations Management at INSEAD, France, and undergraduate studies in Computer Science at BITS-Pilani, India.

Title: The Value of Logistic Flexibility in E-commerce

Abstract: Shipping experience improvement has been an essential business strategy in e-commerce. Beyond investing directly in reducing shipping speed, online retailers have recently expanded their focus on other shipping strategies, such as offering consumers the option to pick up orders locally in a station. This paper uses the opening of hundreds of such pick-up stations as a natural experiment to study the impact of these stations on consumers. We find that the introduction of pick-up stations has increased total sales by 3.8%. In contrast with past literature, we show that shipping time reduction is not the driving factor on the impact of pick-up stations. Yet, the logistic flexibility introduced by pick-up stations explains the sales impact. To explicitly examine how logistic flexibility affects consumers' decisions on purchases, we develop and estimate a structural model on consumer choice. In our model, consumers value two types of logistics flexibility--the flexibility to choose to pick up their items in their preferred time, denoted as the value of time flexibility, and the flexibility to delay such picking time decisions after packages arrive, denoted as the value of choice flexibility. We show that the value of time flexibility accounts for 76.2% of the impact on sales, while the value of choice flexibility accounts for the remaining 23.8%. Using our estimated model, we develop a counterfactual strategy in building pick-up stations that could achieve the sales lift with 56.4% fewer stations. Last but not least, using our estimated time flexibility, we also develop a novel shipping strategy without pick-up stations that could improve the sales by 8.4%.

Bio: Bing Bai is a Ph.D. candidate of Supply Chain, Operations, and Technology at Olin Business School, Washington University in St. Louis. And she is joining McGill University as an Assistant Professor of Operations Management starting summer 2023. Her research focuses on data-driven problems in the digitization of online platforms. She implements field experiments, and uses structural models, causal inference and machine learning to study human behaviors and the human-algorithm connections on platforms. For more information.

Title: “Adaptive Clinical Trial Designs with Surrogates: When Should We Bother?”

Abstract: How can we best leverage available data to make decisions on the efficacy of new drugs? The success of a new drug is assessed within a clinical trial using a primary endpoint, which is typically the true outcome of interest—for example, overall survival. However, regulators sometimes approve drugs using a surrogate outcome, an intermediate indicator that is faster or easier to measure than the true outcome of interest—for example, progression-free survival—as the primary endpoint when there is demonstrable medical need. Although using a surrogate outcome (instead of the true outcome) as the primary endpoint can substantially speed up clinical trials and lower costs, it can also result in poor drug-approval decisions because the surrogate is not a perfect predictor of the true outcome. In this paper, we propose combining data from both surrogate and true outcomes to improve decision making within a late-phase clinical trial. In contrast to broadly used clinical trial designs that rely on a single primary endpoint, we propose a Bayesian adaptive clinical trial design that simultaneously leverages both observed outcomes to inform trial decisions. We perform comparative statics on the relative benefit of our approach, illustrating the types of diseases and surrogates for which our proposed design is particularly advantageous. Finally, we illustrate our proposed design on metastatic breast cancer. We use a large-scale clinical trial database to construct a Bayesian prior and simulate our design on a subset of clinical trials. We estimate that our design would yield a 16% decrease in trial costs relative to existing clinical trial designs, while maintaining the same Type I/II error rates.

Bio: Arielle is a PhD Candidate in Wharton’s Operations, Information, and Decisions Department, advised by Prof. Hamsa Bastani, and will be joining Cornell University’s Johnson school this summer as an assistant professor. Her research is primarily focused on healthcare operations, though it does have applications to other fields such as marketing. In recognition of her work in this area Arielle was awarded the 2020 Bonder Scholarship for Applied Operations Research in Health Services. She looks at designing novel adaptive algorithms so that medical professionals can better leverage data to determine efficacy or necessity of treatment. Her primary research stream works to improve the design of clinical trials. 

Title: Trust and Reciprocity in Firms’ Capacity Sharing

Abstract: We study the use of non-monetary incentives based on reciprocity to facilitate capacity sharing between two service providers who have limited and substitutable service capacity. We propose a parsimonious game theory framework, in which two firms dynamically choose whether or not to accept each other's customers without the capability to perfectly monitor each other's capacity utilization state.  We solve the continuous-time imperfect monitoring game by focusing on a class of public strategy, in which firms' real-time capacity sharing decision depends on an intuitive and easy-to-implement accounting device, namely, the current net number of transferred customers. We refer to such an equilibrium as a trading-favors equilibrium. We characterize the condition in which capacity sharing takes place in such an equilibrium. We find that some degree of efficiency loss (as compared to a central planner's solution) is necessary to induce reciprocity. The efficiency loss is small when the two firms have similar traffic intensity even if they are different in service-capacity scale, whereas the efficiency loss can be considerably large when the two firms have significantly different traffic intensities. The trading-favors mechanism, surprisingly, can outperform the perfect-monitoring benchmark when the two firms exhibit high asymmetry in terms of service-capacity scale or traffic intensity, because the smaller firm tends to deviate from collaboration. Firms should consider engaging in non-monetary reciprocal capacity sharing if regulations, transaction costs, or other market and operational frictions make it difficult to use a capacity-sharing contract based on monetary payments. The trading-favors collaboration can improve the firms' payoff close to the centralized upper bound when the firms have similar traffic intensities. However, when their traffic intensities are highly different, firms are better off with a monetary-payment contract to induce more capacity sharing and are worse off investing in increasing their visibility to each other's real-time available capacity, namely, investing in perfect monitoring.

Bio: Dr. Hu is an associate professor in the area of Innovation and Information Management. Her expertise includes online retailing operations, such as dynamic pricing and logistics, revenue management, and sharing economy operations. Her research on these topics has been published in top-tier research journals including 

Management Science, Operations Research, and Manufacturing and Service Operations Management. She received her Ph.D. from the Stern School of Business at the New York University, and obtained her Bachelor of Science in Mathematics from the Peking University. Before joining the University of Hong Kong, she taught at the University of Oregon.

Title: Policy Evaluation and Learning in Small-Data, Weakly-Coupled Settings 

Abstract: For optimization problems in small-data, large-scale settings, leveraging problem structure is crucial to learning effective, data-driven policies. We propose a novel debiasing method for policy evaluation and learning tailored to weakly-coupled optimization problems and policies in small-data, large-scale settings. Our approach exploits the optimization problem's sensitivity analysis to estimate the gradient of the optimal objective value with respect to the amount of noise in the data. It then uses the estimated gradient to debias the policy's in-sample performance. Unlike cross-validation, our method does not sacrifice training data for policy evaluation and, hence, is well-suited to settings where data are scarce. We prove high-probability bounds on the estimation error of our estimator for optimization problems with uncertain linear objectives but known, potentially non-convex, feasible regions. Specifically, we show the error of our estimator that holds uniformly over a policy class and depends on the problem's dimension, the degree of coupling in the problem, and the policy class's complexity. Our bounds show that under mild conditions, the relative error of our estimator vanishes as the dimension of the optimization problem grows, even if the amount of available data remains small and constant. Said differently, we prove our estimator performs well in the small-data, large-scale regime. Finally, we numerically compare our proposed method to state-of-the-art approaches through a case study on dispatching emergency medical response services using real data. Our method provides more accurate estimates of out-of-sample performance and learns better-performing policies.

Bio: Michael is currently a Ph.D. candidate in the Data Sciences and Operations department at the

 University of Southern California Marshall School of Business, advised by Vishal Gupta and Paat Rusmevichientong. His research focuses on developing data-driven decision-making methods that utilize decision-aware or end-to-end learning. These methods are essential for solving optimization problems in data-scarce settings, which arise when: data collection is expensive, signal-to-noise ratios are low, or systems are highly time inhomogeneous. During his Ph.D., he worked on practical data-driven problems while interning at IBM and Boston Consulting Group. Before his Ph.D., he received his M.S. and B.S. in Operations Research at Columbia University.

Title: Sharing Newsboys

Abstract:  We consider a network of socially connected newsvendors facing random demand for a product who need to commit to a stocking level before demand realizes. A newsvendor can share her ex post excess stock to fulfill the unsatisfied demand of a connected newsvendor. The amount of shared supply that a newsvendor anticipates receiving from her network is affected by two factors: sharing magnitude and tie strength. Sharing magnitude (resp., tie strength) measures the portion of excess stock that a newsvendor will share (resp., the likelihood that a newsvendor will share her excess supply) with a neighboring newsvendor. We adopt a Bayesian game framework with incomplete information about the network structure, where a newsvendor has private information about the number of connections she has (as her type) but does not know her neighbors' types, which she believes are consistent with a network's known degree distribution. First, we demonstrate that with more sharing activity (i.e., greater sharing magnitude or stronger social ties) within a fixed network, all newsvendors decrease their stocking levels regardless of their types, which implies that the total consumption level drops. Second, we show that when tied with the number of connections a newsvendor has, the sharing magnitude has a first-order effect on the mean of the shared supply, while the social tie has a second-order effect on the variability of the shared supply. As the degree distribution of the network increases in the sense of usual stochastic dominance, we show that the two factors may have opposite effects on the equilibrium stocking levels. The effect of sharing magnitude is to increase the equilibrium stocking levels. But the effect of tie strength is such that for a high-fractile product, the population's expected consumption level increases, while it is the other way around for a low-fractile product. Lastly, we extend the supply-sharing base model to complete network information under specific networks and to demand sharing, where unsatisfied demand at one newsvendor can be referred to a neighbor in her network.

Bio: Setareh Farajollahzadeh is a Ph.D. candidate at the Rotman School of Management. Her research applies techniques such as network games, queueing, capacity management, learning, statistics, and data-driven algorithms to solve socially responsible and revenue management operations problems. Setareh will join Desautels Faculty of Management as an assistant professor. 

Title: Structuring Online Communities

Abstract: Online Question and Answer communities were started to supplement customer support services. In contrast to conventional customer support, users in online communities can post questions, and other users with more experience or knowledge can answer these questions. Generally, question answerers get rewards and visibility in the community while the askers gain knowledge if their questions get answered. We model the community as a multistage stochastic game where users have different skill levels. We study how users decide to join, leave, and participate in these communities.

We link the user participation decisions to the underlying network structure of the community. Theoretically, we show that under most parameters, only a core-periphery network structure can emerge in such communities. This network structure has been empirically observed in most online communities. Finally, we explore the cost of asking questions as a lever that a community designer can use to balance user participation and the community’s efficiency in providing answers to users’ questions. We find that increasing the cost of asking questions in the community improves the proportion of askers that get answers to their questions. This results in higher user satisfaction. However, a higher asking cost lowers the participation level in the community. This trade-off between participation and community efficiency results in non-monotonicity in the number of users in the community with the participation cost. The communities typically operationalize higher asking costs by either directly penalizing question-asking activity or setting up stricter guidelines for questions to be answered. We find that increasing the cost of asking is not always bad for the community. A higher asking cost can lead to an increase in the number of users in the community. We also discuss how the existence of low-knowledge users in the community (and not necessarily the high-knowledge users) is essential to the survival of such communities.

Bio: Neha Sharma, is a PhD candidate at Kellogg School of Management and an incoming Assistant Professor at Wharton OID. Her research focuses on problems in service operations with novel business models and strategic agents. She has worked on designing peer-to-peer sharing platforms with applications in knowledge-sharing platforms such as Question and Answer communities, and asset-sharing platforms such as Airbnb, Turo. She has been collaborating with multiple firms to work on problems with immense potential for social impact and enjoys using real world data to support and build analytical models.

Title: Project Networks and Reallocation Externalities

Abstract: A project involves several “participants” —including agencies, contractors, and subcontractors —all working concurrently on multiple projects and allocating resources among them. This interdependency creates a network of otherwise unrelated projects. By constructing the largest project network ever mapped, we track the timelines of 2.6 million infrastructure projects involving 140,000 participants. We show that a seemingly localized disruption, affecting only one project site, eventually causes delays and penalties across unrelated projects. This is because self-interest drives participants to opportunistically reallocate resources into disrupted projects, at the expense of other projects, triggering a domino effect of further reallocations in the network. Thus, the costs of on-site disruptions end up being evenly shared by multiple participants within the network, rather than being fully absorbed by the affected project. Performance-based contracts, which reward contractors for timeliness, exacerbate these externalities by encouraging self-interested resource reallocation. 

Bio: Juan Serpa is an Associate Professor and Department Chair at McGill University, where he specializes in addressing empirical problems related to networks and supply chains.  Prior to joining McGill University, Dr. Serpa held an academic position at the Kelley School of Management. He earned his Ph.D. from the University of British Columbia

Dr. Serpa's research is focused on analyzing the behavior of agents in complex networks and investigating the impact of network structures on various phenomena. His work has contributed significantly to the fields of operations management and information systems.

Aside from his academic pursuits, Dr. Serpa is an avid fan of Montreal and enjoys exploring the city's rich culture, history, and cuisine. He is also writing a vegan book with 102 recipes that has one vegan meal of every country, from Afghanistan to Zimbabwe.

Title: Improving Dispute Resolution in Two-Sided Platforms: The Case of Review Blackmail

Abstract: We study the relative merits of different dispute resolution mechanisms in two-sided platforms, in the context of disputes involving malicious reviews and blackmail. We develop a game-theoretic model of the strategic interactions between a seller and a (potentially malicious) consumer. In our model, the seller takes into account the impact of consumer reviews on his future earnings; recognizing this, a malicious consumer may attempt to blackmail the seller by purchasing the product, posting a negative review, and demanding ransom to remove it. Without a dispute resolution mechanism in place, the presence of malicious consumers in the market can lead to a significant decrease in seller profit, especially in settings characterized by high uncertainty about product quality. The introduction of a standard centralized dispute resolution mechanism (whereby the seller can report allegedly malicious reviews to the host platform, which then judges whether to remove the review) can restore efficiency to some extent, but requires the platform's judgments to be both very quick and highly accurate. We demonstrate that a more decentralized mechanism (whereby the firm is allowed to remove reviews without consulting the platform, subject to ex post penalties for wrongdoing) can be much more effective, while simultaneously alleviating -- almost entirely -- the need for the platform's judgments to be quick. Our results suggest that decentralization, when implemented correctly, may represent a more efficient approach to dispute resolution.

Bio: Yiangos Papanastasiou is an Associate Professor of Operations Management. His research focuses on the operations of online platforms and marketplaces. His more recent work addresses research questions pertaining to optimal information provision, social learning and the economics of misinformation. Yiangos completed his PhD in Management Science and Operations at the London Business School, and his undergraduate and master’s degrees in Information Engineering at the University of Cambridge.

Title: Dynamic Relocations in Car-Sharing Networks

Abstract: We propose a novel dynamic car relocation policy for a car-sharing network with centralized control and uncertain, unbalanced demand. The policy is derived from a reformulation of the linear programming fluid model approximation of the dynamic problem. We project the full-dimensional fluid approximation onto the lower-dimensional space of relocation decisions only. This projection results in a characterization of the problem as linear programs, where is the number of nodes in the network. The reformulation uncovers structural properties that are interpretable using absorbing Markov chain concepts and allows us to write the gradient with respect to the relocation decisions in closed form. Our policy exploits these gradients to make dynamic car relocation decisions. We provide extensive numerical results on hundreds of random networks where our dynamic car relocation policy consistently outperforms the standard static policy. Our policy reduces the optimality gap in steady state by more than on average. Also, in a short-term, time-varying setting, the lookahead version of our dynamic policy outperforms the static lookahead policy on average to a greater degree than that observed in the time-homogeneous tests.

Bio: I am an Operations management Ph.D. candidate at Rotman School of Management, University of Toronto. My research focuses on problems in operations management, specifically service operations, online platforms, sharing economy, and diversity and fairness. As a researcher, I worked on challenging dynamic programs such as dynamic programs in networks and dynamic programs with memory in service operations. My research agenda builds on my past work and attempts to develop full-scale models to capture the world’s true nature and design tractable robust algorithms to optimize real-time decision-making under uncertainty. 

Title: Courier Dispatch in On-Demand Delivery.

Abstract: We consider a courier dispatch problem in on-demand delivery. While traditional wisdom indicates potential efficiency gain in routing by batching/pooling orders together for a single delivery trip, on-demand delivery, such as food delivery, brings new characteristics, such as spontaneous orders with less patient customers and smaller batch sizes. We build a spatial queueing model with revenue management to address the question: With the objective of revenue maximization, when is batching/pooling beneficial in on-demand delivery systems? This is joint work with Prof. Ming Hu from the University of Toronto. 

Bio:  Mingliu Chen is currently an Associate Research Scientist and Adjunct Assistant Professor in the Department of Industrial Engineering and Operations Research at Columbia University. His research focuses on design problems in emerging markets and mechanism design by implementing techniques in optimization, dynamic programming, queueing theory, and data analytics.

Title: Behavior-Aware Queueing: When Strategic Customers Meet Strategic Servers

Abstract: Service system design is often informed by queueing theory. Traditional queueing theory assumes that customers are indefinitely patient and servers work at constant speeds. That is reasonable in computer science and manufacturing contexts. However, customers and servers in service systems are people, and, in contrast to jobs and machines, systemic incentives created by design decisions influence their behavior. 

First, we study the behavior of strategic servers whose choice of work speed depends on managerial decisions regarding (i) how many servers to staff and how much to pay them, and (ii) whether and when to turn away customers. We develop a game-theoretic many-server Markovian queueing model with a finite or infinite buffer in which the work speeds emerge as the solution to a non-cooperative game. In an asymptotic regime in which demand becomes large and the utility function becomes concave, we establish existence, uniqueness, and monotonicity properties of underloaded, critically loaded, and overloaded equilibria for various regions in the design space. We then extend our model to also incorporate strategic customers' joining decisions, which endogenously induce a finite buffer. By comparing equilibria where strategic individuals maximize their own utility with those that maximize social welfare or net profit, we characterize the managerial costs of ignoring strategic behavior.  

Bio: Yueyang Zhong is a fifth-year PhD candidate in Operations Management at the University of Chicago Booth School of Business, advised by Prof. Amy Ward, Prof. Raga Gopalakrishnan, and Prof. John Birge. Previously, she received a bachelor’s degree in Industrial Engineering and Economics from Tsinghua University in 2018. Her primary research interests are in stochastic modeling and the optimization of “modern” stochastic service systems in which individual human behavior and imperfect systemic information must be considered. Her research has been recognized by INFORMS paper competition awards, including as a finalist in the IBM Service Science Best Student Paper Award (2022), and as a finalist in the INFORMS Conference on Service Science Best Student Paper (2021). 

Title: Operational Risk Management: Optimal Inspection Policy

Abstract: Major banks around the world lost nearly $210 billion from operational risk events between 2011 and 2016 (Huber and Funaro 2018). To mitigate the severe consequences that can be brought by operational risk, the Basel Regulatory Committee has required financial institutions worldwide to conduct inspections on operational risk. Motivated by the importance of operational risk and its current industry regulation, this paper proposes a continuous-time principal-agent model to examine a financial firm's (principal) optimal inspection policy and its employees' (agent) effort toward lowering the occurrences of risk events. First, we characterize the optimal inspection strategy under two commonly used policies in practice: random and periodic policies. This characterization reveals the conditions for two different modes of inspection (effort inducement and error correction) as well as nuanced interactions among inspection frequency, penalty charged on errors, and the wage paid to employees. Next, by comparing random and periodic policies, we find that random policy outperforms periodic policy if and only if the inspection cost is high. Furthermore, we propose a hybrid policy that strictly dominates the random policy and weakly dominates the periodic policy, suggesting that a proper reduction of the random element in the inspection policy can always improve its performance. Finally, we examine the first-best benchmark, supplemental mitigation strategies, and numerical studies to provide further insights and show the robustness of our main findings.

Bio: Yuqian Xu is an assistant professor of Operations Management at Kenan-Flagler Business School, University of North Carolina (UNC), Chapel Hill. Her research studies operations in financial services and digital platforms, with a particular focus on (i) leveraging empirical methods with causal inference to understand worker behaviors and (ii) building stochastic control models to improve worker performances and operational outcomes. In her research, she has been collaborating with different companies, including JD.com, Alibaba, Bank of China, etc. She has given talks at different academic, industry, and government conferences and organizations, such as Federal Reserve Bank and China Banking Regulatory Committee. 

Her research has been published in journals including Management Science, Operations Research, Production and Operations Management, etc. She has a B.S. in Mathematics from the Kuang Yaming Honors School of Intensive Instruction in Science and Arts at Nanjing University, China. She received her Ph.D. degree (Beta Gamma Sigma) in 2017 from NYU Stern School of Business with the Herman E. Krooss Dissertation Award.

Title: Coverage Objective and its Generalizations in Location Analysis

Abstract: Coverage is one of the two classical objective functions (along with the median) in location analysis. The idea is both simple and appealing: instead of worrying about exact travel patterns of customers to facilities, we simply focus on whether a particular location configuration ensures that there is a facility sufficiently close to each customer location.

This has both practical and computational appeal: on the practical side, the potential applications range from public service facilities, including emergency facilities, to retail stores. On the computational side, the coverage focus allows for significant simplification of the underlying optimization models.For example, while the $p-$median problem with planar decision space and discrete demand is much more challenging than its discrete counterpart, the planar $p-$cover problem is easily reducible to its discrete version.

A close relative of the coverage objective is ``obnoxious cover'' or anti-cover, which applies in case of undesirable facilities (such as landfills) where the number of ``covered'' customers should be minimized. This seemingly innocuous change from maximization to minimization already creates certain complications and undesirable model behaviours that need to be corrected with additional constraints.

The classic coverage objective makes several strong implicit assumptions: (1) coverage is an ``all or nothing'' phenomenon, i.e., a customer is either covered or not covered, (2) coverage is determined by the facility that is closest to the customer, (3) coverage (or its absence) is deterministic. In recent years models have been proposed relaxing some or all of these assumptions by examining the very idea of ``coverage'': under what circumstances can we be reasonably sure that a given customer will receive ``adequate'' service from a certain facility configuration? One of the earliest generalizations was the idea of ``gradual cover'', relaxing assumption (1). Concepts such as ``cooperative cover'' and ''multi-cover'', relaxing assumptions (2) and (3) followed. Another extension is that of ``robust cover'', ensuring that adequate coverage is maintained even if some parts of the transportation network fail. The price of relaxing these assumptions are models that are significantly more challenging, particularly when the decision space and/or the demand space are not discrete. In most cases, similar extensions apply to anti-cover versions, often with additional complications. In this talk we will review both the classical models and many of the extensions, review some exact and heuristic solution approaches, and outline some open problems and directions for future research. 

Bio: Dmitry Krass is a Professor of Operations Management and Statistics and Sydney C. Cooper Chair in Business and Technology.  Obtaining his Ph.D. in Operations Research from Johns Hopkins University, he joined the School in 1989.  He consults extensively in the areas of Predictive Analytics, Optimization of Marketing Communications and Operational Effectiveness.  His research and teaching interests include facility location models, transportation, reliability and inventory location modeling and humanitarian logistics. He is also interested in environmental modeling, including regulation of pollution fines, marketing mix management and optimization, and predictive/ prescriptive analytics using “big data” and other tools in business decision making.

Title: Discrete Choice Models and Applications: the Past and Future

Abstract: Modeling choice behavior among multiple options has been an active research area for several decades. In this talk, we first review several classic discrete choice models that are widely used in studying purchase behavior for consumers faced with multiple substitutable products. We provide an overview for a variety of operations management problems under discrete choice models, e.g. pricing, assortment planning, and estimation under different structures of real data. In addition, many other choice models have also been proposed to capture new features that arise in choice process. We will discuss recent development on choice modeling, and present several ongoing and future research topics.

Bio: Dr. Ruxian Wang is a Professor with tenure at Johns Hopkins University, Carey Business School. He received Ph.D. from Columbia University. Before returning to academia, he worked in Hewlett-Packard Company for several years as a research scientist. His research and teaching interests include operations management, revenue management, pricing, discrete choice models, data-driven decision making. His articles appeared in the flagship journals in his field, such as Management Science, Manufacturing & Service Operations Management, Operations Research, Production and Operations Management.

Title: Dynamic Relocations in Car-Sharing Networks

Abstract: We propose a novel dynamic car relocation policy for a car-sharing network with centralized control and uncertain, unbalanced demand. The policy is derived from a reformulation of the linear programming fluid model approximation of the dynamic problem. We project the full-dimensional fluid approximation onto the lower-dimensional space of relocation decisions only. This projection results in a characterization of the problem as linear programs, where is the number of nodes in the network. The reformulation uncovers structural properties that are interpretable using absorbing Markov chain concepts and allows us to write the gradient with respect to the relocation decisions in closed form. Our policy exploits these gradients to make dynamic car relocation decisions. We provide extensive numerical results on hundreds of random networks where our dynamic car relocation policy consistently outperforms the standard static policy. Our policy reduces the optimality gap in steady state by more than on average. Also, in a short-term, time-varying setting, the lookahead version of our dynamic policy outperforms the static lookahead policy on average to a greater degree than that observed in the time-homogeneous tests.

Bio: I am an Operations management Ph.D. candidate at Rotman School of Management, University of Toronto. My research focuses on problems in operations management, specifically service operations, online platforms, sharing economy, and diversity and fairness. As a researcher, I worked on challenging dynamic programs such as dynamic programs in networks and dynamic programs with memory in service operations. My research agenda builds on my past work and attempts to develop full-scale models to capture the world’s true nature and design tractable robust algorithms to optimize real-time decision-making under uncertainty. 

Title: (Some) New Directions of Assortment Optimization

Abstract: Assortment optimization has been a popular topic in the revenue management literature. Assortment planning problems typically assume that the seller offers a set of products to the customer, who chooses at most one product from the provided set. This setup restricts the applications of assortment planning in two crucial ways that lead to two research projects. In the first paper [1], we assume that the seller is interested in the linear combination of several objective terms rather than only maximizing the expected revenue. Each term is a convex function of the choice probability-weighted sum of certain product-specific metrics. This generic form includes many objectives that can be of practical interest, such as revenue, market share, risk, utility (when the choice model belongs to the generalized extreme value family), concave costs, etc. We show that despite the non-linearity of the assortment optimization problem, one can recast it into a single pseudo-revenue maximization problem in which the pseudo-revenues depend on the unknown optimal assortment. This reformulation technique allows us to design efficient algorithms to solve the multi-objective assortment optimization problem under various choice models. In the second paper [2], we consider an assortment optimization problem where the customer may purchase multiple products and possibly more than one unit of each product purchased. We adopt the customer consumption model based on the Multiple-Discrete-Choice (MDC) model. We identify conditions under which the revenue-ordered assortments are optimal and the tight performance guarantee of the revenue-ordered assortments in general settings. Given that the assortment problem is NP-hard, we propose an algorithm framework that connects the problem to the Knapsack problem and facilitates the design of FPTAS for the problem under a range of practical constraints. To deal with the seemingly deterministic choice quantity setup, we consider the mixed customer segments, incorporating random utility uncertainty and incorporating random choice noise.

Bio: Heng Zhang is an assistant professor in the Supply Chain Management Department at Arizona State University, W.P. Carey School of Business. His research interests include revenue management, marketing analytics, supply chain management, data analytics, data-driven optimization, and algorithmic causal inference. He holds a Ph.D. with a concentration in Operations Management from the Data Sciences and Operations Department USC Marshall School of Business. Before joining USC, He obtained his M. Sc. in Systems Engineering from the University of Pennsylvania. In the past, he also worked as an economist at Kwai.com and a research scientist at Amazon.com. 

Title: Revenue Management with Heterogeneous Resources

Abstract: We study revenue management problems with heterogeneous resources, each with unit capacity. An arriving customer makes a booking request for a particular interval of days in the future. We offer an assortment of resources in response to each booking request. The customer makes a choice within the assortment to use the chosen resource for her desired interval of days. The goal is to find a policy that determines an assortment of resources to offer to each customer to maximize the total expected revenue over a finite selling horizon. The problem has two useful features. First, each resource is unique with unit capacity. Second, each customer uses the chosen resource for a number of consecutive days. We consider static policies that offer each assortment of resources with a fixed probability. We show that we can efficiently perform rollout on any static policy, allowing us to build on any static policy and construct an even better policy. Next, we develop two static policies, each of which is derived from linear and polynomial approximations of the value functions. We give performance guarantees for both policies, so the rollout policies based on these static policies inherit the same guarantee. Lastly, we develop an approach for computing an upper bound on the optimal total expected revenue. Our results for efficient rollout, static policies, and upper bounds all exploit the aforementioned two useful features of our problem. We use our model to manage hotel bookings based on a dataset from a real-world boutique hotel, demonstrating that our rollout approach can provide remarkably good policies and our upper bounds can significantly improve those provided by existing techniques.

Bio: Mika Sumida is an Assistant Professor of Data Sciences and Operations in the Marshall School of Business at the University of Southern California. Her research focuses on developing efficient, provably good algorithms for revenue management and resource allocation problems, with applications in the sharing economy, online marketplaces, and delivery systems. She holds a Ph.D. in Operations Research and Information Engineering from Cornell University, and a B.A. from Yale University. Prior to her Ph.D., she spent two years working in operations consulting at Analytics Operations Eng., Inc.

Title: Optimal Capacity Planning for Cloud Service Providers with Periodic, Time-Varying Demand

Abstract: Allocating sufficient capacity to cloud computing services is a challenging task because demand is time-varying and jobs do not queue. To study this issue, we model the service as a multi-station queueing network where the arrival rate is time-varying and the servers represent CPU cores. Jobs are impatient and those that are not immediately serviced may retry several times. We determine an optimal service capacity and retrial interval under a novel voluntary admission control policy where the server informs customers of the time of their next service attempt. We introduce a recursive representation of the offered load which describes its fluid dynamics and develop a calculus-of-variation approach to minimize the total variation in the constructed offered load; we prove that it is similar to maximizing the system throughput. Using a data set of cloud computing requests over a representative 24-hour period, we show that our optimal policy results in a 10% reduction in capacity. We investigate the fidelity of the fluid model and the sensitivity of our recommendations to changes in problem parameters. Our study demonstrates that retrial-time announcements allow a provider to satisfy service level agreements while encouraging retrial jobs to be processed during off-peak periods.

Bio: Adam Diamant is an Associate Professor of Operations Management and Information Systems at the Schulich School of Business (York University). His research uses a variety of mathematical techniques (optimization, decomposition methods, and stochastic modeling) and data-driven methodologies (econometrics, artificial intelligence, and machine learning) to model complex, large-scale systems in health care and supply chain management to obtain insights for better operational decision making. 

Title: Capacity Planning in Stable Matching: An Application to School Choice

Abstract: We introduce the problem of jointly allocating additional seats (given a fixed budget) and finding the best allocation for the students in the expanded market in the context of school choice. We show theoretically and empirically that, when allocating the additional capacity, a trade-off between access (i.e., increasing the number of students assigned) and fairness (i.e., achieving the greatest improvement) arises. We also study the incentive properties of the model and show that the timing of students' applications affects the strategy-proofness of the mechanism. In addition, we provide exact methods to solve the problem and two heuristics to obtain near-optimal solutions in a short amount of time. We empirically evaluate the performance of our approaches in a detailed computational study, and we show that our approach is competitive relative to other methods in the literature. Finally, we use the Chilean school choice system data to demonstrate the impact of capacity planning under stability conditions. Our results show that each additional school seat can benefit multiple students. In addition, depending on the decision-maker, our methodology can prioritize the assignment of previously unassigned students or improve the assignment of several students through improvement chains. Finally, we provide several extensions to accommodate different settings.

Bio: Ignacio Rios is an Assistant Professor at the Jindal School of Management at The University of Texas at Dallas. Prior to joining UTD, Ignacio obtained his Ph.D. in Operations, Information, and Technology from the Stanford Graduate School of Business in 2020. Ignacio’s research interest lies at the intersection of market design and behavioral operations. Specifically, Ignacio studies how to design markets and allocate scarce resources taking into account the behavior of their users to improve efficiency, and fairness, among other goals. His work has been applied and implemented in different markets, including school choice, college admissions, and dating markets, and he constantly collaborates with policy-makers, NGOs, and industry partners.

Title: Service Miner: Automating Data-driven Business Analytics in Congested Systems

Use case: Mining Hybrid Machine Learning and Simulation Models for North York General Hospital (NYGH)

Abstract: Simulation is a popular tool for analyzing and optimizing processes across various domains, including healthcare. However, constructing and tuning these models even in presence of available event data is not a trivial task: One must fit the various building-blocks of the model (e.g., service times) using data, decide on the proper model granularity, account for missing data, and take features related to congestion and customer-specific context (e.g., gender and medical diagnosis) into account. Machine learning is the methodology of choice for predictive analytics: It can seamlessly consider contextual and congestion information when learning and tuning models. Yet, its focus is on modeling the conditional expectation of an outcome (for accurate prediction) rather than generating simulated traces (with the exception of generative machine learning models). In this work, we combine the best of both worlds: we propose a simple hybrid model that provides the ability to simulate (an abstract representation of) the system, while capturing context and congestion via an embedded machine learning model. The main goal of the approach is to learn models that would not only perform well on prediction tasks, but even more importantly, provide a framework for comparative, what if analysis, that is essential in improving the way the process is being currently executed.

Our framework is based on constructing single-stage infinite-server queueing models with state-dependent service durations and data-driven arrivals. The model is highly suitable for datasets where queueing and capacity information is not explicitly observed (e.g., when no resource information is provided). The main idea is to be able to capture the state of the system using various congestion features and feed those features into a machine learning model that generates congestion-dependent service time durations. We use the notion of predictive validation to calibrate our model: if the model is able to predict a customer's length-of-stay well on an out-of-sample dataset, we conclude that the model is validated and we can then use the model for optimizing the process.  We demonstrate the usefulness of the approach using real-world emergency department data from the NYGH hospital in Toronto where we use the model to analyze a potential improvement based on speeding up services for selected types of patients and show that our approach adequately simulate the new process.

Title: Hiding in Plain Sight: Surge Pricing and Strategic Provider

Abstract: Many on-demand service platforms employ surge pricing policies, charging higher prices and raising provider compensation when customer demand exceeds provider supply. There is increasing evidence that service providers understand these pricing policies and strategically collude to induce artificial supply shortages by reducing the number of providers showing as available on the app. We study a stylized mathematical model of a setting in which an on-demand service platform determines its pricing and provider compensation policies, anticipating their impact on customer demand and the participation of strategic providers, who might collectively decide to limit the number of providers showing online as available. We find that collusion can substantially harm the platform and customers, especially when the potential demand is large and the supply of providers in nearby regions is limited. We explore two pricing policies that a platform could employ in the presence of (potential) provider collusion: a bonus pricing policy that offers additional provider payments on top of the regular compensation, and the optimal pricing policy that maximizes the platform’s expected profit while taking strategic provider behavior fully into consideration. Both policies offer a compensation structure that ensures that total provider earnings increase in the number of providers available, thereby encouraging all providers to offer their service. Interestingly, we find that once a platform designs an optimal pricing policy to prevent provider collusion, having an option to collude might harm service providers.

Bio: Jiaru Bai is an assistant professor at the Stony Brook University College of Business. Her recent work addresses issues in crowdsourcing platforms. She received her PhD in management and Master’s degree in statistics from the University of California, Irvine.

Title: Incentive Design and Pricing under Limited Inventory

Abstract: A firm faces random demand for a service it delivers on a given future date. To boost the demand, the firm hires a sales agent who exerts unobservable effort continuously over time. The firm is concerned not only about boosting current demand, but also about smoothing the demand over time to avoid the loss of goodwill when the realized demand exceeds its limited inventory. We model the firm’s incentive design problem using a continuous-time principal-agent framework in which demand drifts over time in response to unobserved agent effort and the price the firm charges. To induce the agent’s sales effort, the firm chooses an incentive scheme that depends on the remaining inventory and the time to the service (e.g., time to departure in the case of airlines). We characterize the firm’s optimal incentive scheme under both static and dynamic pricing policies. Using parameters calibrated from the airline industry, we numerically show that under dynamic pricing, a static incentive scheme provides nearly all of the benefit of the corresponding dynamic incentive scheme. By contrast, using a fully static strategy causes a substantial efficiency loss. We also compare two partially dynamic strategies under which the firm practices either dynamic pricing or dynamic contracting, but not both. Among other findings, we show that all else being equal, under a high inventory level, the dynamic-contracting-only strategy tends to outperform the dynamic-pricing-only strategy; under a tight inventory level, however, the dynamic-pricing-only strategy tends to perform better.

Bio: Professor Keppo teaches risk management and analytics courses, and directs analytics executive education programs at NUS Business School. He is also Research Director of the Institute of Operations Research and Analytics at NUS. Previously, he taught at the University of Michigan.

He has several publications in the top-tier journals such as Journal of Economic Theory, Review of Economic Studies, Management Science, Operations Research, and Journal of Business on topics such as investment analysis, banking regulation, learning, and strategic incentives. His research has been featured also in numerous business and popular publications, including the Wall Street Journal and Fortune.

Professor Keppo’s research has been supported by several Asian, European, and US agencies such as the National Science Foundation. He serves on the editorial boards of Management Science, Mathematics of Operations Research, and Journal of Risk. He has consulted several startups, Fortune 100 companies, and financial institutions. 

Title: Labor Cost Free-Riding in the Gig Economy

Abstract: We propose a theory of gig economies in which workers participate in a shared labor pool utilized by multiple firms. Since firms share the same pool of workers, they face a trade-off in setting pay rates; high pay rates are necessary to maintain a large worker pool and thus reduce the likelihood of lost demand, but they also lower a firm’s profit margin. We prove that larger firms pay more than smaller firms in the resulting pay equilibrium. These diseconomies of scale are strong too; firms smaller than a critical size pay the minimal rate possible (the workers’ reservation wage), while all firms larger than the critical size earn the same total profit regardless of size. This scale disadvantage in labor costs contradicts the conventional wisdom that gig companies enjoy strong network effects and suggests that small firms have significant incentives to join an existing gig economy, implying gig markets are highly contestable. Yet we also show that the formation of a gig economy requires the existence of a large firm, in the sense that an equilibrium without any firms participating only exists when no single firm has enough demand to form a gig economy on its own. The findings are consistent with stylized facts about the evolution of gig markets such as ride sharing.

Bio: Zhen Lian is currently a postdoctoral fellow at Lyft Rideshare Labs. Her research focuses on marketplaces, focusing on both their economics and physical operations. Drawing on optimization, stochastic process, microeconomics, and industrial organization theories, her research provides insights into both firm-level decisions and industry-level dynamics. Zhen obtained her Ph.D. from Cornell University in August 2022 and is joining Yale School of Management (SOM) in July 2023.

Title: A Unified Analysis for Assortment Planning with Marginal Distributions

Abstract: We study assortment problems under the marginal distribution model (MDM), a semiparametric choice model that only requires marginal error information without assuming independence. It is known that the multinomial logit (MNL) model belongs to MDM. In this paper, we further show that some multi-purchase choice models, such as the multiple-discrete-choice (MDC) model, and threshold utility model (TUM), also fall into the framework of MDM, although MDM does not explicitly model multi-purchase behavior. For the assortment problem under MDM, we characterize a general condition for the marginal distributions under which a strictly profit-nested assortment is optimal. Moreover, though the problem is shown to be NP-hard, we prove that the best strictly profit-nested assortment is a 1/2-approximate solution for all MDMs. We further construct a simple case of MDM such that the 1/2-approximate bound is tight. These results either generalize or improve existing results on assortment optimization under MNL, MDC, and TUM.

Bio: Xiaobo Li is an assistant professor in the Department of Industrial Systems Engineering and Management at the National University of Singapore. He received his Ph.D. in Industrial Engineering from the University of Minnesota in 2018. His research mainly focuses on robust optimization, discrete choice modelling and dynamic programming, with applications in revenue management, data-driven decision making and supply chain management.

Title: Taming the Long Tail: The Gambler’s Fallacy in Intermittent Demand Management

Abstract: “Long tail” products with intermittent demand often tie up valuable warehouse space and capital investment of many companies. Furthermore, the paucity of demand data poses additional challenges for model estimation and performance evaluation. Traditional inventory solutions were not designed for products with intermittent demand. In this paper, we propose a new framework to optimize the choice of “replenishment timings” and “replenishment quantities”, to manage the inventory performance metrics of long tail products, when evaluated over a finite horizon. Our analysis is motivated by the recent interesting observation that, in a finite number of coin tosses, the gambler’s fallacy phenomenon holds due to the finite horizon effect. We use this phenomenon to analyze the inventory problem for intermittent demand to show that classical inventory models using KPIs such as fill rate, cost per cycle or cost per unit, will need to “bias” the underlying demand distribution to capture the effect of the finite horizon. We provide the exact closed-form expression of the biased distribution to account for the length of the horizon in performance evaluation. The results show that the choice of replenishment timings, together with the replenishment quantities, are key to the performance on several key inventory metrics. More surprisingly, the policy of replenishment (immediately) up to a constant base stock level, often used in practice, can be improved (in both cost and space dimensions) using a variant of staggered base-stock policy, where the stock levels are gradually increased over a horizon. These findings hold even when demands are independent and identical across periods! For long tail products, the belief that it is less likely for another demand to arrive shortly after a preceding one (Gambler’s Fallacy), turns out to be true when the empirical probabilities are tabulated over a finite horizon, even if demands across time are independent. Managers can therefore optimize over the replenishment timings, on top of replenishment quantities, to streamline the performance metrics of several classes of inventory problems. This approach not only dominates the traditional method in terms of inventory cost but also has the potential to reduce warehouse space usage. The latter is especially useful for companies dealing with a huge number of long tail products.

Bio: Long He is an associate professor of decision sciences at the School of Business, George Washington University. Prior to joining GW, Long was an associate professor in the Department of Analytics & Operations (DAO) at NUS Business School, National University of Singapore. He received his Ph.D. in Operations Research from the University of California, Berkeley, and his B.Eng. in Logistics Management and Engineering from HKUST. His current research involves using data-driven approaches to address problems in smart city operations (e.g., vehicle sharing, last-mile delivery) and supply chain management. This line of research has been recognized with the M&SOM Journal Best Paper Award and Transportation Science & Logistics (TSL) Best Paper Award from INFORMS.

Title: Information Frictions and Heterogeneity in Valuations of Personal Data

Abstract: We investigate how consumer valuations of personal data are affected by real-world information interventions. Proposals to compensate users for the information they disclose to online services have been advanced in both research and policy circles. These proposals are hampered by information frictions that limit consumers' ability to assess the value of their own data. We use an incentive-compatible mechanism to capture consumers’ willingness to share their social media data for monetary compensation, and estimate distributions of valuations of social media data before and after an information treatment. We find evidence of significant dispersion and heterogeneity in valuations before the information intervention, with women and Black and low-income individuals reporting systematically lower valuations than other groups. After an information intervention, we detect significant revisions in valuations, concentrated among individuals with low initial valuations. Dispersion and heterogeneity in valuations across these demographic groups decrease but persist after the information intervention. The findings suggest that strategies aimed at reducing information asymmetries in markets for personal data may increase consumer welfare. At the same time, the findings highlight how consumer valuations of personal data are only in part influenced by market information.

Link: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3974826

Bio: Avinash (Avi) Collis is an Assistant Professor in the Information, Risk and Operations Management department at the McCombs School of Business at the University of Texas at Austin. He is also a digital fellow at the MIT Initiative on the Digital Economy and the Stanford Digital Economy Lab. He holds a PhD in Management Science from MIT Sloan School of Management. His research interests include the economics of digitization focusing on measuring the welfare gains from digital goods. His research has been published in top tier academic and practitioner journals including the Proceedings of the National Academy of Sciences, Nature Human Behavior, Nature Communications and the Harvard Business Review and has been covered in major media outlets and policy reports including the New York Times, Wall Street Journal, Washington Post, the Economist and reports by the US White House, Federal Reserve, Senate and UK treasury. 

Title: Pseudo-Competitive Games and Algorithmic Pricing

Abstract: Algorithmic pricing is increasingly a staple of e-commerce platform operations; however, while such data-driven pricing techniques are known to work well in non-strategic environments, their performance in competitive settings remains poorly understood. To this end, we investigate market outcomes that may arise when multiple competing firms deploy local price experimentation algorithms while treating their market environment as a black-box. For price-competition games induced by a broad class of well-validated customer behavior models, we demonstrate that price trajectories resulting from natural local learning dynamics may converge to outcomes in which firms can experience unbounded losses in revenue compared to the best price equilibrium. We moreover design a novel learning algorithm to address this concern.

Bio: Chamsi Hssaine is a postdoctoral scientist in Amazon's Supply Chain Optimization Technologies group, working under the supervision of Garrett van Ryzin. She will be joining the Department of Data Sciences and Operations at USC's Marshall School of Business in Summer 2023 as an Assistant Professor. Chamsi's research lies at the intersection of online decision-making and incentive design, finding applications in dynamic pricing, inventory management, and smart transit operations. Her paper "Real-Time Approximate Routing for Smart Transit Systems" (joint with Sid Banerjee, Noémie Périvier, and Samitha Samaranayake) was a finalist for the 2021 INFORMS Minority Issues Forum Paper Competition. In addition to this, she has been awarded a Simons-Berkeley Research Fellowship for Fall 2022, and was named a 2020 Rising Stars in EECS. Prior to joining Amazon, Chamsi obtained her Ph.D. in Operations Research at Cornell, where she was advised by Siddhartha Banerjee, and her B.S. from Princeton University.

Title: Human performance at solving 0-1 knapsack problems: Results from a large-scale field experiment

Abstract: The 0-1 knapsack problem is a fundamental problem in discrete choice that permeates many diverse fields and applications. Although there has been much research into characterizing the instances of such problems that are difficult for machines to solve, comparatively less is known about what instances human solvers find more difficult than others. Previous work has primarily investigated this in controlled laboratory settings with small numbers of participants and problem instances. In this talk, I will discuss the results of a large field experiment to study this question, conducted through a limited time event on a mobile gaming platform with a global playerbase. A total of 100,000 different knapsack instances were pre-designed for this study and randomly assigned to each player attempt at the knapsack problem. The final dataset comprised 17,950,838 distinct player accounts, who made a total of 150,284,195 submitted solution attempts. Our findings reveal that conventional notions of “more difficult” knapsack problems do not necessarily translate to poorer human performance. Conversely, we find that other measures of difficulty are more directly related to lower human performance on these choice tasks.

Bio: My research interests are in the domains of Healthcare Analytics and Supply Chain Management. In the first domain, I'm interested in understanding how mathematical models can be applied to real-world problems in healthcare in order to inform, improve, and enhance medical decision-making and health policy. In the second, I'm interested in understanding how new business models, enabled by digital technology, can be harnessed to unlock hidden efficiencies in supply chains. I also have methodological interests in optimization theory and co-created Robust Optimization Made Easy (ROME), a software package for modeling robust optimization problems.

Title: Designing Approximately Optimal Search on Matching Platforms

Abstract: We study the design of a two-sided matching market in which agents' search is guided by a platform. The platform determines the rates at which agents of different types meet, while agents strategically accept or reject the potential partners whom they meet. We focus on the platform's problem of optimal search design in a continuum matching market model where agents have symmetric pairwise preferences. The platform's objective is to find meeting rates that maximize the equilibrium social welfare of the resulting game. Incentive issues arising from congestion and cannibalization make this design problem intricate. Nonetheless, we give an efficiently computable solution that achieves 1/4 the optimal social welfare. Our solution shows the platform can substantially limit choice while maintaining approximately optimal welfare through a carefully chosen search design.

Bio: Alexander Wei is a third-year Ph.D. student at UC Berkeley advised by Nika Haghtalab, Michael I. Jordan, and Jacob Steinhardt. His research is centered around the intersections of machine learning, economics, and algorithm design. His work has been recognized by a SODA Best Student Paper award and an INFORMS Auctions and Market Design Rothkopf Junior Research Paper Prize (third place). Prior to Berkeley, Alex received an A.B. in Computer Science and Mathematics and an S.M. in Computer Science from Harvard University.

Title: Unmasking human trafficking risk in commercial sex supply chains with machine learning

Abstract: The covert nature of sex trafficking provides a significant barrier to generating large-scale, data-driven insights to inform law enforcement, policy and social work. Existing research has focused on analyzing commercial sex sales on the internet to capture scalable geographical proxies for trafficking. However, ads selling commercial sex do not reveal information about worker consent. Therefore, it is challenging to identify risk for trafficking, which involves fraud, coercion, or abuse.  We leverage massive deep web data (collected globally from leading commercial sex websites) in tandem with a novel machine learning framework (combining natural language processing, active learning, and network analysis) to study how and where sex worker recruitment occurs. This allows us to unmask deceptive recruitment patterns (e.g., an entity that recruits for modeling, but sells sex). Our analysis provides a geographical network view of commercial sex supply chains, highlighting deceptive recruitment-to-sales pathways that signal high trafficking risk. Our results can help law enforcement agencies along trafficking routes better coordinate efforts to tackle trafficking entities at both ends of the supply chain, as well as target local social policies and interventions towards exploitative recruitment behavior frequently exhibited in that region.

Bio: Pia Ramchandani is a doctoral student at the Wharton School at the University of Pennsylvania in the Operations, Information and Decisions department. Her research focuses on using big data for social good, with topics including counter-human trafficking, human rights in supply chains, and responsible AI. She has worked with organizations such as the United Nations, the Clinton Foundation, the TellFinder Alliance of counter-human trafficking partners, and the World Economic Forum. Prior to her PhD, she was a director in PricewaterhouseCoopers' AI Innovation Accelerator lab. 

Title: Self-Adapting Network Relaxations for Weakly Coupled Markov Decision Processes

Abstract: Weakly coupled Markov decision processes (WDPs) are standard formulations in dynamic decision-making and reinforcement learning, serving as the basis model of multi-armed bandit problems, inventory routing, and marketing applications.  These models are often high dimensional but decompose into smaller component MDPs when coupling constraints are relaxed, leading to extensive research in Lagrangian relaxations that dualize the linking constraints to compute heuristic policies and (optimistic) bounds. While computationally appealing, this Lagrangian approach averages away combinatorial information embedded in the linking constraints. We present a class of network relaxations, dubbed feasibility network relaxations (FNRs), that embed an exact network encoding of the linking constraints into a linear programming flow model and provide (weakly) stronger bounds than the Lagrangian approach. We also develop a procedure to obtain the minimally sized such relaxation, which we refer to as self-adapting FNR, as its size automatically adjusts to the structure of the linking constraints. We show that self-adapting FNR provides bounds and policies that match the well-known approximate linear programming (ALP) approach, but is substantially smaller in size, even polynomially sized when existing ALP formulations are exponentially large, such as in the case of classical multi-armed bandits. We also leverage our framework to demonstrate when the Lagrangian and the ALP bounds match, which generalize existing results from the MDP literature. Finally, we discuss trades-off in bounds and policy performance on knapsack bandits and on a class of WDPs arising in a telecommunications maintenance application.

Bio: Andre Augusto Cire is an Associate Professor at the Department of Management at the University of Toronto Scarborough, cross-appointed with the Operations Management area at the Rotman School of Management. Andre's research investigates new methodologies to address large-scale, difficult optimization problems arising in organizations, with a focus on discrete optimization, scheduling, and healthcare applications. Andre's research agenda considers formal approaches that incorporate concepts from mathematical programming, dynamic programming, and artificial intelligence to derive actionable insights and computationally efficient techniques. 

Title: Bypassing the Monster: A Faster and Simpler Optimal Algorithm for Contextual Bandits under Realizability

Abstract: We consider the general (stochastic) contextual bandit problem under the realizability assumption, i.e., the expected reward, as a function of contexts and actions, belongs to a general function class F. We design a fast and simple algorithm that achieves the statistically optimal regret with only O(log T) calls to an offline regression oracle across all T rounds. The number of oracle calls can be further reduced to O(loglog T) if T is known in advance. Our results provide the first universal and optimal reduction from contextual bandits to offline regression, solving an important open problem in the contextual bandit literature. A direct consequence of our results is that any advances in offline regression immediately translate to contextual bandits, statistically and computationally. This leads to faster algorithms and improved regret guarantees for broader classes of contextual bandit problems.

Bio: Yunzong Xu is a fourth-year Ph.D. student in the Institute for Data, Systems, and Society at MIT. In Summer 2021, he was a research intern at Microsoft Research NYC. Prior to joining MIT, he received his dual bachelor's degrees in information systems and mathematics from Tsinghua University in 2018. Yunzong is broadly interested in statistical machine learning and operations research. His current research interests include data-driven decision making, online and reinforcement learning, econometrics and causal inference, with applications to revenue management and healthcare. His research has been recognized by several awards, including finalists in the INFORMS George Nicholson and Applied Probability Society best student paper competitions.

Title: Can Autonomous Vehicles Solve the Commuter Parking Problem?

Abstract: We investigate how autonomous vehicles (AVs) may change the morning commute travel pattern and improve downtown parking. We develop a continuous-time traffic model that takes into account key economic deterrents to driving, such as parking fee and traffic congestion, and characterize the departure time and parking location (downtown or outside downtown parking area) patterns of commuters in equilibrium. To illustrate our results, our model is calibrated to data from Pittsburgh. For the calibrated model, our analysis shows that all AV commuters choose to park outside downtown, increasing both vehicle hours and vehicle miles traveled as compared to the case with all human-driven vehicles. This change increases the total system cost and suggests a potential downtown land-use change (e.g., repurposing downtown parking spots to commercial and residential areas) in Pittsburgh after mass adoption of AVs. To reduce the total system cost, a social planner may be interested in regulating commuters' decisions by adjusting parking fees and/or imposing congestion tolls as a short-term measure, or adjusting infrastructure, e.g., converting downtown parking spaces to curbside drop-off spots for AVs. Our results indicate that these measures can reduce the total system cost substantially (e.g., up to 70% in our calibrated model).

Bio: Neda Mirzaeian is an assistant professor of operations management at the Jindal School of Management at University of Texas at Dallas. Her research interest revolves around smart city operations. More specifically, she studies the potential effects of innovative technologies, such as autonomous vehicles, on highway traffic congestion, parking and the sharing economy. Neda’s research has won multiple awards, including the IBM Best Student Paper Award in Service Science. Prior to joining UTD, Neda completed her PhD in operations management at the Tepper School of business, Carnegie Mellon University.

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

Abstract: Delayed promotion incentives in the form of retailer-specific gift cards are becoming increasingly popular in the retail industry. These gift cards are offered to customers as a reward for spending more than an expenditure level on regularly priced (as opposed to discounted) products, which can be redeemed by customers towards a future purchase at the retailer. In theory, gift card promotions are an attractive proposition for retailers to potentially stimulate sales during the promotion (at regular price) and lock future demand. Despite the potential benefits, little is known about whether and how gift card promotions impact customer purchase behavior. We address these questions by taking a causal inference approach. We collaborated with a major U.S.-based department store that runs gift card promotions on its online channel by targeting its customers through emails. We utilize discontinuities in the retailer’s targeting policies (based on customers’ purchase recency) across several gift card promotions to estimate localized causal effects using a collection of fuzzy regression discontinuity designs. We find that gift card promotion email increases average customer expenditure during the promotion by $5.64, $1.37, and $1.30 for customers with 4, 13, and 16 months purchase recencies, respectively, corresponding to 21.84%, 29.18%, and 133.55% increase in sales. Furthermore, we find that customers are induced to spend more (beyond the gift card face value) while redeeming their gift card, thus validating the dual value of delayed incentives. We find that majority (78%–100%) of the increase in sales due to gift card promotion email can be attributed to the effect of the email channel rather than the effect of participation in the promotion. Our results also suggest that, contrary to popular belief, a customer’s incremental expenditure while redeeming a gift card recovers the retailer’s promotion cost, on average, and, therefore, makes redeeming a gift card more beneficial to retailers than not redeeming it (i.e., slippage).

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

Bio: Serdar Şimşek is an Associate Professor at the Jindal School of Management of The University of Texas at Dallas and an Amazon Visiting Academic. His research focuses on empirical pricing and revenue management and supply chain management problems primarily in the retail industry and business-to-business markets. The main goal of his research is twofold: (i) developing methodologies that enable a firm to estimate and quantify the impact of customers’ strategic and behavioral motives on their purchase decisions, and (ii) using these motives to design effective pricing/procurement/ inventory management mechanisms that optimize profitability as well as consumer satisfaction. His research appeared in Management Science, Operations Research, Manufacturing and Service Operations Management, and Production and Operations Management journals.

Title: Optimal Matchmaking Strategy in Two-sided Marketplaces

Abstract: Online platforms that match customers with suitable service providers utilize a wide variety of matchmaking strategies: some create a searchable directory of one side of the market (i.e., Airbnb, Google Local Finder); some allow both sides of the market to search and initiate contact (i.e., Care.com, Upwork); others implement centralized matching (i.e., Amazon Home Services, TaskRabbit). This paper compares these strategies in terms of their efficiency of matchmaking, as proxied by the amount of communication needed to facilitate a good market outcome. The paper finds that the relative performance of the above matchmaking strategies is driven by whether the preferences of agents on each side of the market are easy to describe. Here, ``easy to describe'' means that the preferences can be inferred with sufficient accuracy based on responses to standardized questionnaires. For markets with suitable characteristics, each of the above matchmaking strategies can provide near-optimal performance guarantees according to an analysis based on information theory. The analysis provides prescriptive insights for online platforms.

Link: https://pubsonline.informs.org/doi/abs/10.1287/mnsc.2022.4444

Bio: Peng Shi is an Assistant Professor of Data Sciences and Operations in the USC Marshall School of Business. His current research focuses on optimization in matching markets, with applications in school choice, public housing, organ allocation, and online marketplaces. His research has won multiple awards, including the MSOM Responsible Research in OM Award, the MSOM Service Management SIG Best Paper Award, the ACM SIGecom Doctoral Dissertation Award, the INFORMS Public Sector Operations Best Paper Competition, and the INFORMS Doing Good with Good OR Student Paper Competition. Prior to joining USC, he completed a PhD in operations research at MIT, and was a post-doctoral researcher at Microsoft Research.

Title: Provably Good Region Partitioning for On-Time Delivery

Abstract: This paper studies the optimal region partitioning policy to minimize the expected delivery time of customer orders in a stochastic and dynamic setting. We allow both the order locations and on-site service times to be random and generally distributed. This policy assigns every driver to a subregion, hence making sure drivers will only be dispatched to their own territories. We characterize the structure of the optimal partitioning policy and show its expected on-time performance converges to that of the flexible dispatching policy in heavy traffic. The optimal characterization features two insightful conditions that are critical to the on-time performance of delivery systems. We then develop partitioning algorithms with performance guarantees, leveraging ham sandwich cuts and 3-partitions from discrete geometry. This algorithmic development can be of independent interest for other transportation and logistics problems.

Bio:  Sheng Liu is an Assistant Professor of Operations Management and Statistics at the Rotman School of Management. His research interests lie in smart city operations (especially transport, last-mile logistics, and sustainable/climate-resilient infrastructure planning) and data-driven decision-making (the integration of predictive and prescriptive analytics). He received a PhD in Operations Research from UC Berkeley in 2019 and a BSc in Industrial Engineering from Tsinghua University in 2014.

Title:  Pricing Strategies for Online Dating Platforms

Abstract: "Online dating is now the most common way for new couples to meet, with three-in-ten Americans have used dating apps, and with revenues from dating apps swelling to more than five billion annually. The majority of these dating apps earn revenue via subscription-based pricing, where subscriptions for a period of access to the app are sold at a fixed, reoccurring price. Subscription-based pricing is a ubiquitous way to monetize mobile apps, however, in the context of online dating different subscription periods lengths can lead to very different behaviors for the users and the platform. The purpose of this work is to understand the profit and welfare trade-offs associated with a class of subscription-based pricing strategy for online dating platforms, parameterized by period length."

Bio: Michael Hamilton is an Assistant Professor of Business Analytics and Operations at the University of Pittsburgh, Katz Graduate School of Business.  He is broadly interested in problems related to pricing, prescriptive analytics, and market design. He received his Ph.D. in Operations Research from Columbia IEOR in 2019.

Title: Mediating social influence and network effects with targeted interventions, descriptive norms messaging and algorithmic rewiring

Abstract: I will cover a number of recent results on intervention design in social network contexts. In “seeding with costly network information” (forthcoming in Opre. Res.), we present a unifying framework for data collection and targeted interventions with theoretical guarantees to trade off the cost of data collection with increasing intervention size. I will show the utility of this framework in analyzing privacy and fairness implications of data collection and intervention designs. In “long ties accelerate noisy threshold-based contagions”, we argue for targeted structural interventions based on the properties of the adoption behavior (simple or complex contagions, with or without inertia) and show conditions under which adoptions spread much faster with increasing long, rather than short, ties. I will then present results from a large, pre-registered, randomized experiment (N=484,239) embedded in an international survey that show messaging with accurate descriptive norms can substantially increase intentions to accept a vaccine for COVID-19. Finally, I present some early results from an online experiment with a multiplayer collaborative prediction game (N=704), where participants were randomly assigned  to one of the four 16-person network treatments that algorithmically medicated their communications (including a static network condition with no mediation). Although we found no statistically significant treatment effect on post-communication collective predictions, the within-group effects were significant. I contextualize this potential of rewiring algorithms to enhance collective decision with some theoretical results for “when social influence promotes the wisdom of crowds”. Based on joint works with Dean Eckles, Elchanan Mossel, Hossein Esfandiari, Abdullah Almaatouq, Kiran Garimella, Subhabrata Sen, Sinan Aral, Alex Moehring, Avinash Collis, Jason Burton and Ulrike Hahn. 

Bio: Amin Rahimian joined Pitt IE in the fall of 2020. Prior to that, he was a postdoc with joint appointments at MIT Institute for Data, Systems, and Society (IDSS) and MIT Sloan School of Management. He received his PhD in Electrical and Systems Engineering from the University of Pennsylvania, and Master’s in Statistics from Wharton School. Broadly speaking his works are at the intersection of networks, data, and decision sciences. He borrows tools from applied probability, statistics, algorithms, as well as decision and game theory. Some of his current focus is on the challenges of inference and intervention design in complex, large-scale sociotechnical systems, with applications ranging from online social networks, public health, e-commerce and collective decision/action platforms to modern civilian cyberinfrastructure and future battlefields. He is especially interested in the critical role that information plays in the operation of sociotechnical institutions and its societal implications, including on diversity, fairness, and privacy. He has recently served on the program committees of the 2021 ACM Economics and Computation conference and the 2022 IISE annual conference (as the operations research track co-chair), as well as the advisory council of the vaccine confidence fund (a new industry alliance). He has published in the Proceedings of the National Academy of Sciences, Nature Human Behaviour, the Operations Research journal, the Automatica journal, and several IEEE Transactions. At Pitt, he leads the sociotechnical systems research lab and teaches Stochastic Processes (IE 2084), Design of Experiments (IE 1072) and a new engineering elective on “Data for Social Good (IE 1171)” that he has developed through the Pitt Year of Data and Society Initiative. 

Title: Planning Bike Lanes with Data: Ridership, Congestion, and Path Selection

Abstract: Urban infrastructure is essential to building sustainable cities. In recent years, municipal governments have invested heavily in the expansion of bike lane networks to meet growing demand, promote ridership, and reduce emissions. However, re-allocating vehicle capacity in a road network to cycling is often contentious due to the risk of amplifying traffic congestion. In this paper, we develop a method for planning bike lane networks that accounts for ridership and congestion effects. We first present an estimator for recovering unknown parameters of a traffic equilibrium model from features of a road network and observed vehicle flows, which we show asymptotically recovers ground-truth parameters as the network grows large. We then present a prescriptive model that recommends paths in a road network for bike lane construction while endogenizing cycling demand, driver route choice, and driving travel times. In an empirical study on the City of Chicago, we bring together data on the road and bike lane networks, vehicle flows, travel mode choices, bike share trips, driving and cycling routes, and taxi trips to estimate the impact of expanding Chicago's bike lane network. We estimate that adding 25 miles of bike lanes as prescribed by our model can lift ridership from 3.9% to 6.9%, with at most an 8% increase in driving times. We also find that three intuitive heuristics for bike lane planning can lead to lower ridership and worse congestion outcomes, which highlights the value of a holistic and data-driven approach to urban infrastructure planning.

Bio: Jingwei Zhang is a 4th-year PhD student in operations management at the UCLA Anderson School of Management. Her research combines tools from optimization, game theory, and data analytics with applications to urban mobility and sustainable operations. Her work has been awarded 1st place in the POMS College of Sustainable Operations Student Paper Competition. In the summer of 2022, she will work as a data scientist intern at Lime.  

Title: On the impact of treatment restrictions for the indigent and chronically-ill: The case of compassionate dialysis.

Abstract: We analyze a congested healthcare delivery setting resulting from emergency treatment of a chronic disease on a regular basis. A prominent example of the problem of interest is congestion in the emergency room (ER) at a publicly funded safety net hospital resulting from recurrent arrivals of uninsured end-stage renal disease patients needing dialysis (a.k.a. compassionate dialysis). Unfortunately, this is the only treatment option for un/under-funded patients (e.g., undocumented immigrants) with ESRD, and it is available only when the patient’s clinical condition is deemed as life-threatening after a mandatory protocol, including an initial screening assessment in the ER as dictated and communicated by hospital administration and county policy. After the screening assessment, the so-called treatment restrictions are in place, and a certain percentage of patients are sent back home; the ER, thus, serves as a screening stage. The intention here is to control system load and, hence, overcrowding via restricting service (i.e., dialysis) for recurrent arrivals as a result of the chronic nature of the underlying disease. In order to develop a deeper understanding of potential unintended consequences, we model the problem setting as a stylized queueing network with recurrent arrivals and restricted service subject to the mandatory screening assessment in the ER. We obtain analytical expressions of fundamental quantitative metrics related to network characteristics along with more sophisticated performance measures. The performance measures of interest include both traditional and new problem-specific metrics, such as those that are indicative of deterioration in patient welfare because of rejections and treatment delays. We identify cases for which treatment restrictions alone may alleviate or lead to severe congestion and treatment delays, thereby impacting both the system operation and patient welfare. The fundamental insight we offer is centered around the finding that the impact of mandatory protocol on network characteristics as well as traditional and problem-specific performance measures is nontrivial and counterintuitive. However, impact is analytically and/or numerically quantifiable via our approach. Overall, our quantitative results demonstrate that the thinking behind the mandatory protocol is potentially naive. This is because the approach does not necessarily serve its intended purpose of controlling system-load and overcrowding.

Bio: Olga Bountali is an Assistant Professor of Operations Management at the University of Toronto. She is cross-appointed in the Institute of Communication, Culture, Information and Technology and the Department of Management at University of Toronto Mississagua, and the Rotman School of Management. She holds a Ph.D. from the Operations Research Group of the Department of Mathematics, at the University of Athens. Prior to joining the University of Toronto, Olga was a postdoctoral fellow at Southern Methodist University (OREM Department) and Koç University (Industrial Engineering Department). Her research interests are in operations management with a focus on stochastic modeling and complex decision-making under competition. In particular, her expertise lies in applying economic analysis to quantify how the behavior of self-interested customers impacts service operations, revenues, and the delivery of health care. She has active collaborations with people from diverse disciplines, including physicians and practitioners at Parkland Memorial Hospital and UT Southwestern in the DFW Metroplex. Olga is also a research fellow at the Sandra Rotman Centre for Health Sector Strategy.

Title: Experimentation Platform and Learning Treatment Effects in Panels

Abstract: Experiments in brick-and-mortar retail are contaminated for myriad reasons. Pragmatic inference in such settings is more akin to learning from observational data, as opposed to the typical setup one might consider for a carefully designed randomized experiment. So motivated, we consider the problem of causal inference in panels with *general* intervention patterns. We provide a novel, near-complete solution to this problem that allows for rate-optimal recovery of treatment effects. Our work expands the applicability of the synthetic control paradigm. In doing so, we extend the analysis of non-convex optimization techniques for matrix completion to non-random missing-ness patterns and noise; a non-trivial feature of independent interest. Our algorithms form the core of a new testing platform we co-developed with a USD 100B drink company. Over the last six months, the platform has hosted approximately 300 experiments that have run across 200,000 stores in Central and South America representing 12 billion dollars of transactions.

Link: https://proceedings.neurips.cc/paper/2021/hash/7504adad8bb96320eb3afdd4df6e1f60-Abstract.html

Bio: Tianyi Peng is a final Ph.D. student at MIT. He is advised by Vivek Farias, and also mentored by Andrew Li at CMU. He is broadly interested in developing algorithms for learning and causal inference in large-scale dynamic decision-making systems. In particular, he is interested in developing the next generation Experimentation Platform, which aims to provide scalable and low-cost solutions for identifying beneficial strategies/policies that are otherwise impossible to be discovered. In translating these ideas, he is working with Anheuser-Busch InBev, the world's largest beer producer, to co-develop their physical retailer experimentation platform. He also engaged with Tiktok, Takeda Pharmaceuticals, Liberty Mutual, and Broad Institute, to develop learning algorithms from a causal lens. He is a graduate of the 2017 Yao Class at Tsinghua University.

Title:  Dynamic Pricing and Auction Design for Freight Transportation Marketplaces

Abstract: The U.S. trucking market is enormous in size but is fragmented and inefficient. In recent years, digital freight marketplaces that provide automated matching services have emerged and are reshaping the freight industry. In this talk, we study a freight platform operator that serves as an intermediary between shippers and carriers in a truckload transportation network. The market operator faces uncertainty in both supply and demand and can adjust shipper and carrier prices based on market dynamics. We analyze three types of mechanisms commonly used in practice: posted price mechanisms, auction mechanisms, and hybrid mechanisms that offer both pricing and bidding channels. 

Bio: He Wang is an Assistant Professor and Colonel John B. Day Early Career Professor in the School of Industrial and Systems Engineering at Georgia Tech. He is also an Applied Scientist in Uber Freight’s Marketplace Dynamics group. His research interest includes pricing, supply chain, transportation, and machine learning.

Title: Analysis of Farm Equipment Sharing in Emerging Economies

Abstract: In recent years, there is a growing number of farm equipment sharing platforms in emerging economies that help connect smallholder farmers with tractor owners who are willing to fulfill farmers' requests for mechanization services. Although these platforms are often labeled as "Uber for Tractors," they face a distinct set of challenges due to the extremely small sizes of individual farms and the high cost of transporting the farm equipment. As a result, it is generally not profitable for tractor owners to serve individual farmers' requests. To address this challenge, a new role that has emerged in these platforms is the so-called "booking agents," who exert a costly effort by collecting demand from individual farmers and submitting the aggregated demand on the platform. In this paper, we study how the presence of booking agents affects the platform's optimal pricing and wage decisions, and how the key supply and demand characteristics of such a platform affect the optimal platform decisions and the equilibrium outcomes. Our analysis yields several insights with managerial implications. First, we show that the platform should consider paying out a higher percentage of the price to booking agents when the number of tractors on the platform increases, when tractor owners’ operating cost decreases, and surprisingly, when the platform puts more emphasis on its profit (over farmer surplus). Second, in contrast to conventional sharing settings (such as ride sharing), we show that in the presence of booking agents, an increase in the number of service providers (i.e., tractors) may lead to a lower optimal platform commission rate (i.e., the percentage of the price the platform keeps). Finally, while efforts to enhance the supply side of the platform (such as an increase in the number of tractors on the platform or a decrease in tractor owners’ operating cost) generally lead to a decrease in price and an increase in total farmer surplus, we find that a decrease in the cost for demand aggregation can lead to an increase in price and a decrease in total farmer surplus.

Bio:  Can Zhang is an Assistant Professor in the OM area at the Duke Fuqua School of Business. He is broadly interested in socially responsible and sustainable operations with a particular focus on underserved populations. Specifically, his research spans three areas: 1) nonprofit and public sector operations, 2) healthcare supply chains in resource-limited settings, and 3) agricultural supply chains in developing and emerging economies. His research has won several awards, including the winner for the MSOM Best Paper Award, the MSOM Award for Responsible Research in OM, the MSOM Student Paper Competition, and the Honorable Mention for the George Dantzig Dissertation Award.

Title: Routing optimization with stochastic and correlated data

Abstract: Interaction between decisions is defined as a situation where the solution cost/benefit corresponding to a decision is affected (i.e., non-additive effects) by other decisions. In many planning problems arising in transportation, supply chain management, and logistics, the interactions between individual events or decisions contain crucial information that cannot be neglected. However, people tend to consider them exclusively when modeling decisions, mainly because it is easier to manage. For example, when decisions are made in the presence of large-scale stochastic data, it is common to pay more attention to the easy-to-see statistics (e.g., mean) instead of the underlying complex correlations. One reason is that it is often much easier to solve a stochastic optimization problem by assuming independence across data. This puts a large gap between the research and reality in situations where significant correlations exist in data.

   In this talk, I will focus on routing optimization where travel times among different road segments are highly correlated, e.g., due to traffic congestion propagation. Congestion can cause a significant variation in travel times, especially during peak hours, and thus impact fuel consumption and, consequently, greenhouse gas emissions. For companies offering home delivery services, delays due to congestion directly affect the quality of the customer's experience and their costs. I present a unifying modeling framework to address travel time correlations and show how different stochastic, robust, and distributionally robust optimization models can be represented in this form. I will also show how to use data analytics techniques to improve the tractability of underline optimization algorithms.

Bio: Borzou is an assistant professor of Operations and Decisions Sciences at Lazaridis School of Business and Economics, Wilfrid Laurier University. He will join the University of Alberta as the CPA chair of business analytics in July 2022. Before joining Laurier, Borzou was a postdoctoral researcher at Polytechnique Montreal and the Technical University of Dortmund, Germany. He holds a PhD in Information Technology from the Polytechnic University of Milan, Italy. His research interests include optimization under interactions and uncertainty and decomposition methods for large-scale mixed-integer nonlinear optimization with applications in supply chain management, transportation and logistics.

Title: Mixed-Projection Optimization - A new framework for modeling rank constraints

Abstract: Over the past decades, mixed-integer optimization (MIO) has emerged as a scalable framework to model and solve problems with non-convex constraints (e.g., cardinality) and logical relationships. Unfortunately, some practically relevant constraints, such as rank constraints, cannot be modeled using MIO. In this talk, I will present a new framework where projection matrices -- matrix analog of binary variables -- are used to capture the rank of a matrix. The resulting optimization problems, which we call mixed-projection optimization (MPO) problems, appear as a generalization of MIO. Indeed, we have extended some of the theoretical and computational tools of MIO to MPO: Using spatial branch-and-bound, we solved to provable (near) optimality problems with matrices with up to 30 (resp. 600) rows/columns. We also extended the perspective reformulation technique to derive tighter convex relaxations. Joint work with Dimitris Bertsimas and Ryan Cory-Wright.

Bio: Jean is an Assistant Professor of Management Science and Operations at London Business School. His research focuses on large-scale discrete optimization, robust optimization, and machine learning, with applications to healthcare operations. His work has been published in the likes of Operations Research, Mathematical Programming, and M&SOM, and recognized by many awards, including the INFORMS Pierskalla, George E. Nicholson, and Computing Society best student paper awards. Jean received a Ph.D. in Operations Research from MIT and a Diplôme d'ingénieur from Ecole Polytechnique (Paris). 

Title: Procurement Mechanisms with Post-Auction Pre-Award Cost-reduction Investigations 

Abstract: A buyer seeking to outsource production may be able to find ways to reduce a potential supplier's cost, e.g., by suggesting improvements to the supplier's proposed production methods. We study how a buyer could use such “cost-reduction investigations" by proposing a three-step supplier selection mechanism: First, each of several potential suppliers submits a price bid for a contract. Second, for each potential supplier, the buyer can exert an effort to see if she can identify how the supplier could reduce his cost to perform the contract; the understanding is that if savings are found, they are passed on to the buyer if the supplier is awarded the contract. Third, the buyer awards the contract to whichever supplier has the lowest updated bid (the supplier's initial bid price minus any cost-reduction the buyer was able to identify for that supplier). For this proposed process, we characterize how the buyer's decision on which suppliers to investigate cost reductions for in step 2 is affected by the aggressiveness of the suppliers' bids in step 1. We show that even if the buyer does not share the cost savings she identifies in step 2, ex ante symmetric suppliers are actually better off (ex ante) in our proposed mechanism than in a setting without such cost-reduction investigations, resulting in a win-win for the buyer and suppliers. When suppliers' cost and cost-reduction distributions become very heterogeneous, the win-win situation may no longer hold, but every supplier still has an incentive to allow the buyer to investigate him in step 2 because it increases his chance of winning the contract. Using an optimal mechanism analysis, our numerical studies show that our proposed Bid-Investigate-Award mechanism helps the buyer achieve near-optimal performance, despite its simplicity.

Bio: Qi (George) Chen is an Assistant Professor of Management Science and Operations at London Business School. He earned his doctoral degree from the Stephen M. Ross School of Business at the University of Michigan, and his bachelor degree from the Department of Automation at Tsinghua University. Prior to joining LBS, he has taught at the University of Texas at Dallas. His primary research interests include revenue management and pricing, supply chain management and strategic sourcing, and online marketplaces. 

Title: Fair Exploration for Online Learning

Abstract: Exploration is often necessary in online learning to maximize long-term reward, but it comes at the cost of short-term 'regret'. We study how this cost of exploration is shared across multiple groups. For example, in a clinical trial setting, patients who are assigned a sub-optimal treatment effectively incur the cost of exploration. When patients are associated with natural groups on the basis of, say, race or age, it is natural to ask whether the cost of exploration borne by any single group is 'fair'.

So motivated, we introduce the 'grouped' bandit model. We leverage the theory of axiomatic bargaining, and the Nash bargaining solution in particular, to formalize what might constitute a fair division of the cost of exploration across groups. On the one hand, we show that any regret-optimal policy strikingly results in the least fair outcome: such policies will perversely leverage the most 'disadvantaged' groups when they can. More constructively, we derive policies that are optimally fair and simultaneously enjoy a small 'price of fairness'. We illustrate the relative merits of our algorithmic framework with a case study on contextual bandits for warfarin dosing.

Link:https://openreview.net/forum?id=GEKTIKvslP

Bio: Jackie Baek is currently a PhD candidate in the Operations Research Center at MIT, and she will start as an assistant professor in the Department of Technology, Operations, and Statistics at NYU Stern in 2023. Her research interests are in the areas of machine learning, algorithmic fairness, and healthcare. Her work has been awarded finalists in the George Nicholson Student Paper Competition and the RMP Student Paper Award.

Title: Optimal Dynamic Allocation: Simplicity through Information Design

Abstract: We study dynamic nonmonetary markets where objects are allocated to unit-demand agents with private types. An agent’s value for an object is supermodular in her type and the quality of the object, and her payoff is quasilinear in her waiting cost. We identify the welfare-maximizing mechanism in the class of direct-revelation mechanisms that elicit agents’ types and assign them to objects over time. When the social planner can design the information disclosed to the agents about the objects, this mechanism can be implemented by a first-come first-served wait-list with deferrals. The optimal disclosure policy pools adjacent object types. Moreover, the hazard rate of the distribution of the agents’ types determines the structure of the optimal disclosure policy, when the agents’ utility function is separable. A single-peaked (single-dipped) hazard rate leads to the optimality of a lower censorship (upper censorship) policy.

Link:  https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3610386

Bio: Afshin Nikzad is a microeconomic theorist at University of Southern California. He is interested in market and mechanism design, with a particular focus on matching markets. He received his PhD in Economics from Stanford in 2018. 

Title: Discovering Causal Models with Optimization: Confounders, Cycles, and Feature Selection

Abstract: We propose a new method for learning causal structures from observational data, a process known as causal discovery. Our method takes as input observational data over a set of variables and returns a graph in which causal relations are specified by directed edges. We consider a highly general search space that accommodates latent confounders and feedback cycles, which few extant methods do. We formulate the discovery problem as an integer program, and propose a solution technique that leverages the conditional independence structure in the data to identify promising edges for inclusion in the output graph. In the large-sample limit, our method recovers a graph that is equivalent to the true data-generating graph. Computationally, our method is competitive with the state-of-the-art, and can solve in minutes instances that are intractable for alternative causal discovery methods. We demonstrate our approach by showing how it can be used to examine the validity of instrumental variables, which are widely used for causal inference. In particular, we analyze US Census data from the seminal paper on the returns to education by Angrist and Krueger (1991), and find that the causal structures uncovered by our method are consistent with the literature.

Bio: Auyon Siddiq is an Assistant Professor of Decisions, Operations & Technology Management at the UCLA Anderson School of Management. His research draws from optimization, statistics, and game theory, often with a focus on using optimization methods to estimate structural models from data. His work has been published in Management Science, Operations Research, and M&SOM, and in 2020 he was named a "Best 40 Under 40 Business School Professor" by Poets&Quants. He received a Ph.D. from the University of California, Berkeley, an M.A.Sc. from the University of Toronto, and a B.Eng. from Dalhousie University. He is a native of Halifax, Canada.

Title: On the Joint Inventory and Pricing Control for a One-Warehouse Multi-Store Problem with Lost Sales: Spiraling Phenomena and Near-Optimal Heuristics 

Abstract: We consider a joint inventory and pricing problem with one warehouse and multiple stores, in which the retailer makes a one-time decision on the amount of inventory to be placed at the warehouse at the beginning of the selling season, followed by periodic joint replenishment and pricing decisions for each store throughout the season. Unmet demand at each store is immediately lost. The retailer incurs the usual variable ordering costs, inventory holding costs and lost sales costs, and his objective is to maximize the expected total profits. The optimal policy for this problem is unknown and numerically challenging to compute. We first analyze the performance of two popular and simple heuristics that directly implement the solution of a deterministic approximation of the original stochastic problem. We show that simple re-optimization of the deterministic problem may yield a very poor performance by causing a ``spiraling down" movement in price trajectory, which in turn yields a ``spiraling up" movement in expected lost sales quantity (i.e., the expected  lost sales quantity continues to increase as we re-optimize more frequently). Our findings caution against a naive use of simple re-optimizations in the joint inventory and pricing setting with lost sales without first understanding the dynamics of the model. We then propose two heuristics based on the optimal solution of a deterministic relaxation of the original stochastic problem. Our first heuristic computes static prices and order-up-to levels for warehouse and stores, and then replenish each store at the beginning of each batch of periods. Our second heuristic builds on the first heuristic and dynamically changes prices over time. The dynamic pricing mechanism combines the idea of linear rate adjustment and random error averaging. We show that, under proper choices of control parameters, both heuristics have near-optimal performance when demand is Poisson and the annual market size is large, with the second heuristic outperforming the first one.

Bio: Murray Lei is an Assistant Professor of Management Analytics at Smith School of Business in Kingston, Canada. Murray holds a Ph.D. in Business Administration from the University of Michigan, and a B.Eng in Automation from Tsinghua University. He is interested in designing effective data-driven policies that are easily implementable in a dynamic business environment. His recent work lies in the area of pricing and revenue management, service operations, and supply chain management, and is motivated by real business problems that arise in the context of e-commerce/omnichannel retail and service platforms.

Title: Operating A Three-sided Marketplace: Pricing, Spatial Staffing and Routing in Food Delivery Systems 

Abstract: Motivated by the proliferation of food delivery platforms, such as DoorDash, Uber Eats and Meituan, that match restaurants, customers and delivery drivers over a geographically dispersed network, we study the platform’s joint pricing, staffing and routing problem under endogenous participation of all three sides. Using a state-dependent queueing model where the service rate depends on the imbalance of the three sides due to spatial frictions, we study the equilibrium behavior of a large system in heavy traffic and show through asymptotic analysis how the platform controls balance capacity utilization and service quality. We show the platform’s value is threefold: (i) increased market output as the platform boosts demand for restaurants and offers faster delivery; (ii) delivery resource pooling that saves the restaurants’ logistic costs and increases deliverer utilization; (iii) efficient network routing that reduces cross-location pickups, hence customer waiting and deliverer idleness.  

Bio: Zhe Liu is an Assistant Professor of Operations Management at Imperial College London. His research lies in revenue management, dynamic pricing and supply chain management. He is the recipient of the Finalist in George Nicholson Student Paper Competition, 2nd Place in POMS Student Paper Competition, and 2nd Place in POMS-HK Best Student Paper Competition. Zhe obtained his PhD in Operations Management from Columbia Business School and B.S. in Industrial Engineering from Tsinghua University. 

Title: Revenue maximization under unknown private values with non-obligatory inspection

Abstract: We consider the problem of selling a single item to n unit-demand buyers to maximize revenue, where the buyers' values are independently distributed (not necessarily identical) according to publicly known distributions but unknown to the buyers themselves, with the option of allowing buyers to inspect the item at a cost. This problem can be interpreted as a revenue maximizing variant of Weitzman's Pandora's problem with non-obligatory inspection. We present an approximation mechanism that achieves 1/2. The proposed mechanism generalizes to the case of selling k units of an item to unit-demand buyers, obtaining 1-1/\sqrt{k+3} of the optimal revenue in expectation. The mechanism is sequential and has a simple implementation that works in an online setting where buyers arrive in an arbitrary unknown order, yet achieving the aforementioned approximation with respect to the optimal offline mechanism.

Link: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3855810

Bio: Ali Makhdoumi is an Assistant Professor in the Decision Sciences area at Fuqua School of Business, Duke University. He has received a BSc in Electrical Engineering and a BSc in Mathematics from Sharif University of Technology, and Ph.D. in Electrical Engineering and Computer Science from MIT. His research interests include optimization, game theory, algorithm design, and learning theory with applications to operations research and operations management.

Title: Online Algorithms for Matching Platforms with Multi-Channel Traffic

Abstract: Two-sided platforms rely on their recommendation algorithms to help their visitors successfully find a match. However, on platforms such as VolunteerMatch – which has facilitated millions of connections between volunteers and nonprofits – a sizable fraction of website traffic arrives at a nonprofit’s page via an external link, thus bypassing the platform’s recommendation algorithm. We study how such platforms should account for this external traffic in the design of their recommendation engines, given the goal of maximizing successful matches. We model the platform’s problem as a special case of online matching with stochastic rewards, where (using VolunteerMatch terminology) volunteers arrive sequentially and (probabilistically) match with one opportunity, each of which has finite need for volunteers. In our framework, external traffic is interested only in their targeted opportunity; in contrast, internal traffic may be interested in many opportunities, and the platform’s online algorithm selects which opportunity to recommend. In evaluating the performance of different algorithms, we refine the notion of competitive ratio by parameterizing it based on the amount of external traffic. We introduce a new algorithm – Adaptive Capacity (AC) – which accounts for matches differently based on whether they originate from internal or external traffic. We establish a lower bound on AC’s competitive ratio that is increasing in the amount of external traffic, and we demonstrate that (in certain parameter regimes) AC is near-optimal regardless of the amount of external traffic, even though it does not know this amount a priori. Our analysis utilizes a path-based, pseudo-rewards approach, which we further generalize to settings where the platform can recommend a ranked set of opportunities. Beyond our theoretical results, we demonstrate the strong performance of AC in a case study motivated by VolunteerMatch data.

Bio: Scott Rodilitz is a postdoctoral fellow at Stanford Graduate School of Business, working with Daniela Saban, and he will be joining the DOTM group at UCLA's Anderson School of Management in Summer 2022 as an Assistant Professor. His research aims to amplify the impact of nonprofit and governmental organizations by leveraging powerful tools from the sharing economy, such as crowdsourcing and platform-based markets. Scott's work has been recognized as a winner/finalist for several INFORMS awards, including the MSOM Student Paper Competition, the Rothkopf Junior Researcher Prize, and the Public Sector OR Best Paper Award. His dissertation, which was completed at the Yale School of Management under the supervision of Vahideh Manshadi, received honorable mention for the Dantzig Dissertation Award.

Title: The Value of Information in Optimal Online Learning  

Abstract: The optimization of online learning requires the right balance between the value and cost of information. The existing knowledge about the value of information is limited mostly to offline learning, where information is gathered to improve a single-shot decision. In this talk, we try to understand the value of information in online settings which involve the trade-off between exploration and exploitation. We adopt a Bayesian framework and formulate the problem as a partially observable Markov decision process. The decision maker repeatedly selects from multiple actions whose utilities depend on an unknown state of nature. Different actions generate different imperfect information about the unknown state. The goal of the decision maker is to maximize the expected total utility accumulated over a finite horizon. Our central question is whether the value of information increases with the remaining time (time-to-go). One might expect the information to be more valuable when there are more decision opportunities for capitalizing on the information. While this is true for perfect information, it is generally false for imperfect information. Using a novel analysis, we establish conditions that guarantee the value of information to be increasing in the remaining time, and establish the monotonicity of optimal policies. These results provide useful insights for both the providers and users of imperfect information in an online environment. 

Bio: Jue Wang is an Assistant Professor and Distinguished Faculty Fellow of Management Analytics at Smith School of Business, Queen’s University. He conducts methodological research in sequential decision making under imperfect information, such as Bayesian online classification, change detection, and Bayesian bandit problems. He is especially  interested in understanding the structure of optimal policies for partially observable Markov decision processes, and using the structure to develop efficient solution algorithms. The applications of his research include personalized medicine, revenue management, reliability, and ecology. He received his PhD in Industrial Engineering from the University of Toronto. He also worked as a researcher at GE Healthcare and the Hospital for Sick Children in Toronto. He is actively collaborating with Scotiabank in customer analytics research. Jue has won the New Researcher Achievement Award from Smith School of Business, Queen’s University, the second place of Canadian Operations Research Society student paper competition, and the University of Toronto early stage technology award from MaRS Discovery District.

Title: Cooperative Security Against Interdependent Risks

Abstract: Firms in inter-organizational networks are exposed to a variety of interdependent risks. Interdependent risks are risks that are transferable across partner firms, such as contamination in food supply chains or data breaches in technology networks. They can be decomposed into intrinsic risks a firm faces from its own operations and extrinsic risks transferred from its partners. Firms broadly have access to two security strategies: either they can independently eliminate both intrinsic and extrinsic risks by securing their links with partners, or alternatively, firms can cooperate with partners to eliminate sources of intrinsic risk in the network. First, we develop a graph-theoretic model of interdependent risk and demonstrate that the network-optimal security strategy can be computed in polynomial time via a weighted min-cut network flow algorithm. Then, we use cooperative game-theoretic tools to examine whether and when firms can sustain the network-optimal security strategy via cost-sharing mechanisms that are stable (i.e., individually and coalitionally rational), fair, computable, and implementable via a series of bilateral cost-sharing arrangements. By analyzing commonly employed allocation mechanisms, we uncover a trilemma, that is, it is, in general, challenging to identify cost-sharing mechanisms that are stable, fair, and implementable. We then design a novel cost-sharing mechanism, the agreeable allocation, which is a restricted variant of the well-known Shapley value. It is easy to compute, bilaterally implementable, belongs to the core, and is fair in a well-defined sense. However, the agreeable allocation need not always exist. Interestingly, we find that in networks with homogeneous cost parameters, the presence of locally dense clusters of connected firms precludes the existence of the agreeable allocation, while the absence of sufficiently dense clusters (formally, k-cores) guarantees its existence. Finally, using the SDC Platinum database, we consider all inter-firm alliances formed in the food manufacturing sector from 2006 to 2020 and examine the practical feasibility of identifying bilateral security cost sharing arrangements in real-world alliances that can sustain network wide cooperative security against interdependent risks.

Link: https://arxiv.org/abs/2201.04308 

Bio: Sanjith Gopalakrishnan is an Assistant Professor of Operations Management at the Desautels Faculty of Management, McGill University. His main research interests are in environmental and socially sustainable operations management. He is also interested in the interplay of information, incentives, and fairness in multi-agent environments and networks. He obtained his PhD in Operations Management from the Sauder School of Business, University of British Columbia.

Title: Matching Patients with Surgeons: Heterogeneous Effects of Surgical Volume on Surgery Duration

Abstract: We study the heterogeneous effects of surgical volume on surgery duration

and address the challenges of matching patients with surgeons to improve hospitals’ operational efficiency. Using abdominal surgery as the clinical setting, we first provide empirical evidence that the effect of surgical volume on surgery duration is heterogeneous across patients. We then apply the causal forest approach to generate patient-specific information that captures the heterogeneous volume effects for different patients. We find the effect of surgical volume varies widely across different patients. More specifically, the surgical volume has a significant effect on surgery duration for 88% of patients but not for the remaining 12%. Among the patients with significant effects, the largest effect is around 10 times larger than the smallest effect. Finally, we develop an optimization model to compare the same hospital’s operational efficiency with and without patient-specific information. We find patient-specific information can reduce the total duration of surgeries by 3% to 18%, depending on the mix of patients and surgeons. The results of this study are useful to researchers and policymakers because we provide empirical evidence that the effect of surgical volume is heterogeneous and address the challenges of estimating the heterogeneous effects for different patients. The results are also useful to hospital administrators because we show a hospital can improve its operational efficiency by using patient-specific information to match patients with surgeons.

Link: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3699215

Bio: Behrooz Pourghannad is an Assistant Professor of Operations and Decision Sciences at the Lazaridis School of Business & Economics, Wilfrid Laurier University and a research collaborator at Mayo Clinic’s Data Science Pillar. Behrooz’s main research focuses on healthcare with an emphasis on patient-centric care and business model innovation with an emphasis on platform economy and sustainability. His research draws on several methodologies including machine learning, game theory, dynamic programing, and laboratory experiments.

Behrooz’s research has been recognized with several awards including INFORMS IBM Service Science Section's Best Student Paper Award, POMS College of Behavior in Operations Management Junior Scholar Paper Competition and INFORMS Aviation Application Section Best Student Paper Competition. His papers have been published in several highly regarded journals, including Production and Operations Management and Transportation Science. His more recent works are under revision at Manufacturing & Service Operations Management.

Before joining the Lazaridis School, Behrooz was a postdoctoral fellow at the Institute for Mathematics and its Application (University of Minnesota) and Kern Center for the Science of Health Care Delivery (Mayo Clinic). Behrooz received a B.S. degree in Industrial Engineering from Azad University, an M.A. in Mathematics from Sabanci University, Istanbul, Turkey, an M.A. in Business Administration from the University of Michigan, and a Ph.D. in Industrial Engineering from the University of Minnesota.

Title: Analytics for COVID-19: Accelerating Vaccine Trials and Optimizing Vaccine Distribution

Abstract: The COVID-19 pandemic has caused devastating humanitarian impact worldwide. In this talk, we demonstrate the successful development and implementation of data-driven analytical tools that combine epidemiology, machine learning, and optimization to save thousands of lives during the COVID-19 pandemic. We first introduce a novel epidemiological model DELPHI that incorporates many COVID-19 specific effects, including an explicit modeling of societal interventions. We applied DELPHI to over 200 regions worldwide since early April 2020 with consistently high predictive power, and is a key contributor to the CDC ensemble forecast.

We then illustrate two life-saving applications of DELPHI. First, we worked with Janssen Pharmaceuticals to combine DELPHI with robust optimization to select optimal Phase III trial locations for the Janssen/J&J single-dose COVID-19 vaccine Ad26.Cov2.S, accelerating the trial by 8 weeks and allowing millions earlier access to a life-saving vaccine. The early approval of the Janssen/J&J vaccine is conservatively estimated to have already saved over 3,000 lives from COVID-19 death. Then, we collaborated with the Federal Emergency Management Agency (FEMA) to optimize the Federal COVID-19 vaccine rollout. We incorporated the nonlinear DELPHI dynamics in an optimization model to minimize future pandemic deaths, and developed a novel algorithm to effectively solve the nonlinear problem at scale, showing that an optimized allocation can save more than 4,000 lives in 3 months. The model informed and guided FEMA in allocating vaccines to the 20+ Federal COVID-19 vaccination centers. 

Link: https://mlli.mit.edu/projects/mathematical-modeling/modeling-covid-19-epidemic

Bio: Michael Lingzhi Li is a final-year doctoral candidate at the MIT Operations Research Center, advised by Prof. Dimitris Bertsimas. The goal of his research is to use data-driven analytics and optimization to create real-world impact in healthcare and public health. To achieve this goal, he works on both new optimization algorithms and evaluation frameworks as well as important collaborations with corporations, hospitals, and governments worldwide. He is the recipient of awards including the 2021-2022 INFORMS Edelman Finalist, 2021 INFORMS Doing Good with Good OR Finalist, the 2021 Innovative Applications in Analytics Award, the 2020 INFORMS Pierskalla Award and the 2019 MSOM Best Student Paper Finalist Award.

Title: Revenue Maximization and Learning in Product Ranking

Abstract: We consider the revenue maximization problem for an online retailer who plans to display a set of products differing in their prices and qualities and rank them in order. The consumers have random attention spans and view the products sequentially before purchasing a ``satisficing'' product or leaving the platform empty-handed when the attention span gets exhausted. Our framework extends the cascade model in two directions: the consumers have random attention spans instead of fixed ones and the firm maximizes revenues instead of clicking probabilities. We show a nested structure of the optimal product ranking as a function of the attention span when the attention span is fixed and design a $1/e$-approximation algorithm accordingly for the random attention spans. When the conditional purchase probabilities are not known and may depend on consumer and product features, we devise an online learning algorithm that achieves $\tilde{\cO}(\sqrt{T})$ regret relative to the approximation algorithm, despite of the censoring of information: the attention span of a customer who purchases an item is not observable. Numerical experiments demonstrate the outstanding performance of the approximation and online learning algorithms.

Link: https://arxiv.org/abs/2012.03800

Bio: Dr. Ningyuan Chen is currently an assistant professor at the Department of Management at the University of Toronto, Mississauga, and cross-appointed at the Rotman School of Management, University of Toronto. Before joining the University of Toronto, he was an assistant professor at the Hong Kong University of Science and Technology. Prior to that, he was a postdoctoral fellow at the Yale School of Management. He received his Ph.D. from the Industrial Engineering and Operations Research (IEOR) department at Columbia University in 2015. He is interested in various approaches to making data-driven decisions in business applications such as revenue management. His studies have been published in Management Science, Operations Research, Annals of Statistics, and other journals. His research is supported by the UGC of Hong Kong and the Discovery Grants Program of Canada.

Title: Share or Solo? Individual and Social Choices in Ride-Sharing

Abstract: Ride-hailing platforms offer riders pooling services to share rides with other riders. Shared rides mitigate the driver shortage and reduce rider wait times, especially in rush hours. We study a queueing model in which riders wait for drivers and choose whether to join the queue and if joining, whether to take a shared or solo ride. When a shared ride is chosen, the rider may need to wait for another fellow rider to carpool. We analyze and compare the choices by decentralized riders and the centralized social planner of joining rates and sharing probabilities. We discover that, under the FIFO discipline, self-interested riders in equilibrium always under-share, compared to the socially optimal solution. This leads to under-join behavior by riders due to thinner effective system capacity compared to that of the socially optimal system. In contrast, under the PFS discipline, riders may over-share to gain priority over solo riders, regardless of the negative sharing externality. Nonetheless, in equilibrium, the social planner can induce decentralized riders to achieve the socially optimal joining rate by charging a toll and to achieve the socially optimal sharing probability by appropriate social, monetary, or priority schemes. Lastly, we conduct a numerical study with the ride-hailing data of Chicago in August 2019.

Link: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3675050

Bio: Hengda Wen is a PhD student in Operations Management from Rotman School of Management, University of Toronto. His primary research interest lies in the sharing economy, ride-hailing economy, and queueing theory. In the sharing economy, the platform coordinates self-interested agents to maximize its revenue or the social welfare. His research on ride-hailing economy focuses on designing pricing and priority mechanisms to maximize the social welfare or profit in the ride-sharing market. Hengda’s current works focus on the real-world sharing economy, such as ride-sharing and instant delivery service. He is also working on the crypto sharing economy, such as batched transactions on the blockchain.

Title: Stratifying Online Field Experiments Using The Pigeonhole Design

Abstract: Practitioners and academics have long appreciated the benefits that experimentation brings to firms. For the online web-facing firms, however, it still remains challenging in handling heterogeneity when experimental units arrive sequentially in online field experiments. In this paper, we study a novel online experimental design problem, which we refer to as the “Online Stratification Problem”. In this problem, experimental units with heterogeneous covariate information arrive sequentially, and must be immediately assigned into either the control or the treatment group, with an objective of minimizing the total discrepancy between the two groups. To solve this problem, we propose a novel experimental design approach, which we refer to as the “Pigeonhole Design”. The pigeonhole design partitions the covariate space into smaller spaces, which we refer to as pigeonholes, and maintains the number of control and treatment units as balanced as possible in each pigeonhole. We analyze the theoretical performance of the pigeonhole design and show the effectiveness by comparing against two well-known benchmark designs: the match-pair design, and the completely randomized design.

Bio: Jinglong Zhao is an Assistant Professor of Operations and Technology Management at Questrom School of Business at Boston University. He works at the interface between optimization and econometrics, with applications in online platforms. His research leverages discrete optimization techniques to design field experiments that arise from the technology sector. Jinglong completed his PhD in Social and Engineering Systems and Statistics at Massachusetts Institute of Technology.

Title: Right to Repair: Pricing, Welfare, and Environmental Implications

Abstract: The "right to repair" (RTR) movement calls for government legislation that requires manufacturers to provide repair information, tools, and parts so that consumers can independently repair their own products with more ease. The initiative has gained global traction in recent years. Repair advocates argue that such legislation would break manufacturers’ monopoly on the repair market and benefit consumers. They further contend that it would reduce the environmental impact by reducing e-waste and new production. Yet, the RTR legislation may also trigger a price response in the product market as manufacturers try to mitigate the profit loss. This paper employs an analytical model to study the pricing, welfare, and environmental implications of RTR. We find that as the RTR legislation continually lowers the independent repair cost, manufacturers may initially cut the new product price and then raise it. This non-monotone price adjustment may further induce a non-monotone change in consumer surplus, social welfare, and environmental impact. Strikingly, the RTR legislation can potentially lead to a "lose-lose-lose" outcome that compromises manufacturer profit, reduces consumer surplus, and increases the environmental impact, despite repair being made easier and more affordable.

Link: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3516450

Bio: Luyi Yang is an assistant professor in the Operations and Information Technology Management Group at the University of California, Berkeley’s Haas School of Business. His research interests include service operations, business model innovation, digital marketplaces, smart mobility, sustainability, and operations-marketing interface. His research is published or forthcoming in leading journals such as Management Science, Operations Research, and Manufacturing & Service Operations Management and recognized by various research awards such as M&SOM Service Management SIG Best Paper Award, INFORMS Service Science Best Cluster Paper Award, INFORMS Minority Issues Forum Paper Competition, and INFORMS Junior Faculty Interest Group (JFIG) Forum Paper Competition (Honorable Mention). He has taught courses in business analytics, data mining, and operations management. Prior to joining Berkeley Haas, he was an assistant professor of operations management and business analytics at Johns Hopkins University’s Carey Business School. He received his Ph.D. and MBA from the University of Chicago, Booth School of Business, and his BS in Industrial Engineering and BA in English, both from Tsinghua University.

Title: Cold Start to Improve Market Thickness on Online Advertising Platforms: Data-Driven Algorithms and Field Experiments

Abstract: Cold start describes a commonly recognized challenge for online advertising platforms: with limited data, the machine learning system cannot accurately estimate the click-through rates (CTR) of new ads and, in turn, cannot efficiently price these new ads or match them with platform users. Traditional cold start algorithms often focus on improving the learning rates of CTR for new ads to improve short-term revenue, but unsuccessful cold start can prompt advertisers to leave the platform, decreasing the thickness of the ad marketplace. To address these issues, we build a data-driven optimization model that captures the essential trade-off between short-term revenue and long-term market thickness on the platform. Based on duality theory and bandit algorithms, we develop the Shadow Bidding with Learning (SBL) algorithms with a provable regret upper bound of O(T^{2/3}K^{1/3}(log T)^{1/3}d^{1/2}), where K is the number of ads and d captures the error magnitude of the underlying machine learning oracle for predicting CTR. Our proposed algorithms can be implemented in a real online advertising system with minimal adjustments. To demonstrate this practicality, we have collaborated with a large-scale video-sharing platform, conducting a novel, two-sided randomized field experiment to examine the effectiveness of our SBL algorithm. Our results show that the algorithm increased the cold start success rate by 61.62% while compromising short-term revenue by only 0.717%. Our algorithm has also boosted the platform's overall market thickness by 3.13% and its long-term advertising revenue by (at least) 5.35%. Our study bridges the gap between the bandit algorithm theory and the cold start practice, highlighting the value of well-designed cold start algorithms for online advertising platforms. 

Bio: Renyu (Philip) Zhang is an Assistant Professor of Operations Management at New York University Shanghai and a Visiting Associate Professor at The Chinese University of Hong Kong Business School. He is also an economist and Tech Lead at Kwai, one of the world’s largest online video-sharing and live-streaming platform. Philip’s recent research focuses on data-driven optimization and A/B testing, together with their applications to the recommendation and pricing strategies of large-scale online platforms. His research works have appeared in Operations Research and Manufacturing & Service Operations Management and have been recognized by INFORMS RMP Best Student Paper Competition, INFORMS Data Mining Section Best Paper Award, INFORMS Service Science Section Best Paper Award, and POMS College of Supply Chain Management Best Student Paper Competition. He has also developed data science and economics frameworks to evaluate and optimize the ecosystem of Kwai, especially its recommender system and advertising platform. Prior to joining NYU Shanghai, Philip obtained his PhD degree in Operations Management from Olin Business School, Washington University in St. Louis. 

Title: Supermodularity in Two-Stage Distributionally Robust Optimization

Abstract: We solve a class of two-stage distributionally robust optimization problems which have the property of supermodularity. We exploit the explicit worst-case expectation of supermodular functions and derive the worst-case distribution for the robust counterpart. This enables us to develop an efficient method to obtain an exact optimal solution of these two-stage problems. We also show that the optimal scenario-wise segregated affine decision rule returns the same optimal value in our setting. Further, we provide a necessary and sufficient condition to check whether the supermodularity property holds for any given two-stage optimization problem. We apply this framework to several classic problems, including multi-item newsvendor problems, facility location design problems, a lot-sizing problem on a network, appointment scheduling problems and assemble-to-order systems. While these problems are typically computationally challenging, they can be solved efficiently using our approach.

Bio: Aiqi Zhang is a postdoctoral researcher at Rotman School of Management, advised by Prof. Sheng Liu and Prof. Wei Qi. Her research interest includes robust optimization, machine learning, supply chain management, smart city operations and healthcare operations. Previously, she obtained her bachelor's degree in Statistics from University of Science and Technology of China, and a PhD degree in Industrial Engineering from the Chinese University of Hong Kong.

Title: Who Is Next: Patient Prioritization Under Emergency Department Blocking 

Abstract: Upon arrival at emergency departments (EDs), patients are classified into different triage levels indicating their urgency. Using data from a large hospital in Canada, we find that, within the same triage level, the average waiting time (time from triage to initial assessment by a physician) of patients who are discharged is shorter than that of patients who are admitted for middle- to low-acuity patients, suggesting that the order in which patients are served deviates from first-come, first-served, and to a certain extent, discharged patients are prioritized over admitted patients. This observation is intriguing as, among patients of the same triage level, admitted patients—who need further care in the hospital—should be deemed no less urgent than discharged patients who only need treatment at the ED. To understand how ED decision makers choose the next patient for treatment, we estimate a discrete-choice model and find that ED decision makers apply urgency-specific delay-dependent prioritization. Moreover, we find that, when the ED blocking level is sufficiently low, admitted patients are prioritized over discharged patients for high-acuity patients, whereas disposition does not affect the prioritization of middle- to low-acuity patients. When the ED blocking level becomes sufficiently high, decision makers start to prioritize discharged patients in an effort to avoid further blocking the ED. We then analyze a stylized model to explain the rationale behind the change in decisionmakers’ prioritization behavior as the ED blocking level increases. Using a simulation study, we demonstrate how policies inspired by our findings improve ED operations by reducing the average patient waiting time and length of stay, resulting in significant cost savings for hospitals. We

also show how to leverage our findings to improve the accuracy of ED waiting time predictions. By testing and highlighting the central role of decision makers’ patient prioritization behavior, this paper advances our understanding of ED operations and patient flow.

Bio: Wenhao Li is currently a postdoctoral fellow at the University of Toronto. He received his bachelor’s degree in Automation from Zhejiang University and Ph.D. in Management Sciences from the City University of Hong Kong. His research interests include healthcare analytics and contextual bandits.

Title: Approximation Algorithms for Matching Under Uncertainty

Abstract: The presence of uncertainty is a common challenge when it comes to designing algorithms for matching markets such as organ exchange markets, labor markets, and dating platforms. This uncertainty is often due to the stochastic nature of the data or restrictions that result in limited access to information. In this talk, I will discuss my works on the stochastic matching problem. In this problem, the goal is to find a large matching of a graph whose edges are uncertain. Particularly, we only know the existence probability of each edge but to verify their existence, we need to perform costly queries.  For instance, in labor markets, the existence of an edge between a freelancer and an employer represents their compatibility to work with one another, and a query translates to an interview between them which is often time-consuming. I will present an algorithm we recently developed, showing that despite the uncertainty in the graph, one can find a nearly maximum size matching (weighted and unweighted) by conducting only a constant number of queries per vertex. This significantly improves upon prior algorithms which could only achieve 2/3-approximation for unweighted graphs and only slightly better than 0.5-approximation for weighted graphs.

Bio: Mahsa Derakhshan is currently a FODSI postdoctoral fellow at UC Berkeley. She received her Ph.D. in computer science from the University of Maryland and was later a postdoctoral researcher at Princeton University. Her primary research interest is in algorithmic mechanism design and algorithmic game theory. Particularly, the focus of her research is on designing market algorithms in the presence of uncertainty and strategic behavior.  She is also interested in the design and analysis of big-data algorithms, especially in distributed and dynamic settings.

Title: Parallel Queues with Discrete-Choice Arrival Pattern: Empirical Evidence and Asymptotic Characterization

Abstract: We consider a parallel-queue system in which each queue is served by a dedicated service provider. The arrival process is driven by a discrete choice model, that is, customers observe the queue length for each service provider and choose one to join upon arrival. We assume that a customer’s utility is the difference between the service reward and the waiting cost, both of which are heterogeneous. Empirical analysis of the vehicle queues at the U.S.-Canada border-crossing port of entry supports our model setting. We show that with such a choice model, the arrival rate function satisfies certain properties, which allow us to characterizes the fluid and diffusion limit of the queue-length process. In particular, we show that even without the well-used Lipschitz-continuity assumption, the fluid limit process is unique and is attracted to a unique equilibrium. The diffusion limit process is a reflected multi-dimensional Ornstein-Uhlenbeck process centered at that equilibrium. We prove that the stationary distribution of the diffusion limit is a truncated multivariate Gaussian and interchange of limits holds.

Bio: Yichuan Ding is currently an assistant professor at the Desautels Faculty of Management, McGill University. His research interests lies in operations research and data analytics, and their application in healthcare delivery systems such as kidney allocation and exchange. Yichuan’s research works have appeared in top tier operations management journals including Operations Research, Manufacturing and Service Operations Management, and Mathematics of Operations Research.

Title: The value of knowing drivers' reservation wage in Ride Sharing systems

Abstract:

Consider a ride sharing platform, and a large population of strategic potential drivers, heterogeneous in terms of their reservation wages (i.e., income goals). Drivers choose whether or not to work for the platform. The platform is assumed to be knowledgeable of the different drivers' reservation wages. How can the platform implement a matching policy that leverages this information in order to improve system efficiency? Can such improvement be quantified? 

In this work we adopt a mean-field approach to model the dynamics of drivers' spatial location, revenue, and availability status. We compare drivers' equilibrium participation under two different matching policies: an oblivious one that treats drivers symmetrically, and a reservation-aware one that prioritizes drivers according to their reservation wage. We show that the reservation-aware policy outperforms the oblivious, and can do up to twice as good, both in attracting driver supply and in responding to passenger demand. We demonstrate by simulation that the mean-field model provides an accurate approximation for a corresponding (stochastic) discrete model, in which the discussed improvement is observed empirically.

Bio: Ran I. Snitkovsky is a postdoctoral fellow at Columbia Business School and Shenzhen Research Institute of Big Data, CUHK Shenzhen. In 2020 he received his PhD in Operations Research from the School of Mathematical Sciences, Tel Aviv University, during which he also spent several months as a short-term visiting scholar at Tuck School of Business.  Ran's research revolves around the modeling and methodology of strategic, social and behavioral interactions in queueing systems, with a strong emphasis on economic and managerial insights.  Prior to his PhD, he worked for 5 years in developing Command-and-Control algorithms for the Israeli security system. 

Title:  Epsilon-greedy global optimization under radial basis function interpolation.

Abstract: We study a global optimization problem where the objective function can be observed exactly at individual design points with no derivative information. We suppose that the design points are determined sequentially using an epsilon-greedyalgorithm, i.e., by sampling uniformly on the design space with a certain probability, and otherwise sampling in a local neighborhood of the current estimate of the best solution. We study the rate at which the estimate converges to the global optimum, and derive two types of bounds: an asymptotic pathwise rate, and a concentration inequality measuring the likelihood that the asymptotic rate has not yet gone into effect. Both bounds vanish more rapidly when the width of the local search neighborhood is made to shrink over time at a suitably chosen speed. 

Bio: I am a new postdoc fellow here in the area of Operations Management and Statistics. My previous research involves convergence rate theory of simulation optimization, mainly from a stochastic perspective. My current focus is on discovering statistical properties of models in operations research. Prior to this postdoc position, I earned PhD degree in applied math at University of Maryland. I am happy to meet new faces and I am still exploring new focuses so you are always welcomed to chat with me.

Title: Improving Human Decision-Making with Machine Learning

Abstract: A key aspect of human intelligence is their ability to convey their knowledge to others in succinct forms. However, despite their predictive power, current machine learning models are largely blackboxes, making it difficult for humans to extract useful insights. Focusing on sequential decision-making, we design a novel machine learning algorithm that conveys its insights to humans in the form of interpretable "tips". Our algorithm selects the tip that best bridges the gap in performance between human users and the optimal policy. We evaluate our approach through a series of randomized controlled user studies where participants manage a virtual kitchen. Our experiments show that the tips generated by our algorithm can significantly improve human performance relative to intuitive baselines. In addition, we discuss a number of empirical insights that can help inform the design of algorithms intended for human-AI interfaces. For instance, we find evidence that participants do not simply blindly follow our tips; instead, they combine them with their own experience to discover additional strategies for improving performance. 

Link: https://arxiv.org/abs/2108.08454 

Bio: Park Sinchaisri is an Assistant Professor of Operations & IT Management at UC Berkeley Haas School of Business. His primary research interests center around combining tools from operations, economics, machine learning, and behavioral sciences to study how to manage the future of work and the human-AI interface. He received a PhD in Operations, Information & Decisions and an MA in Statistics from Wharton, an SM in Computational Science & Engineering from MIT, and an ScB in Computer Engineering & Applied Mathematics-Economics from Brown. Growing up in Bangkok, Park hopes to expand his research to make a positive social impact, from solving urban problems to helping the marginalized work populations.

Title: Differential Privacy in Personalized Pricing with Nonparametric Demand Models

Abstract: In the recent decades, the advance of information technology and abundant personal data facilitate the application of algorithmic personalized pricing. However, this leads to the growing concern of potential violation of privacy due to adversarial attack. To address the privacy issue, this paper studies a dynamic personalized pricing problem with \textit{unknown} nonparametric demand models under data privacy protection. Two concepts of data privacy, which have been widely applied in practices, are introduced: \textit{central differential privacy (CDP)} and \textit{local differential privacy (LDP)}, which is proved to be stronger than CDP in many cases. We develop two algorithms which make pricing decisions and learn the unknown demand on the fly, while satisfying the CDP and LDP gurantees respectively. In particular, for the algorithm with CDP guarantee, the regret is proved to be at most $\tilde O(T^{(d+2)/(d+4)}+\varepsilon^{-1}T^{d/(d+4)})$. Here, the parameter $T$ denotes the length of the time horizon, $d$ is the dimension of the personalized information vector, and the key parameter $\varepsilon>0$ measures the strength of privacy (smaller $\varepsilon$ indicates a stronger privacy protection). On the other hand, for the algorithm with LDP guarantee, its regret is proved to be at most $\tilde O(\varepsilon^{-2/(d+2)}T^{(d+1)/(d+2)})$, which is near-optimal as we prove a lower bound of $\Omega(\varepsilon^{-2/(d+2)}T^{(d+1)/(d+2)})$ for any algorithm with LDP guarantee.

Bio: Sentao Miao is an Assistant Professor in Bensadoun School of Retail Management & Desautels Faculty of Management at McGill University. His research interests are mainly in developing efficient learning and optimization algorithms with various applications in Operations Management. For methodologies, Sentao Miao focuses on statistical and machine learning algorithms such as online learning, multi-arm bandit problem, reinforcement learning; he is also interested in approximation algorithms with provable performance. For applications, he mainly works on operations management problems such as dynamic pricing, assortment selection, inventory control, etc. Sentao Miao obtained his PhD degree in Department of Industrial and Operations Engineering at University of Michigan.

Title: Matchmaking Strategies for Maximizing Player Engagement in Video Games

Abstract: Managing player engagement is an important problem in the video game industry, as many games generate revenue via subscription models and microtransactions. We consider a class of online video games whereby players are repeatedly matched by the game to compete against one another. Players have different skill levels which affect the outcomes of matches, and the win-loss record influences their willingness to remain engaged. The goal is to maximize the overall player engagement over time by optimizing the dynamic matchmaking strategy. We propose a general but tractable framework to solve this problem, which can be formulated as an infinite linear program. We then focus on a stylized model where there are two skill levels and players churn only when they experience a losing streak. The optimal policy always matches as many low-skilled players who are not at risk of churning to high-skilled players who are one loss away from churning. In some scenarios, when there are too many low-skilled players, high-skilled players are also matched to low-skilled players that are at risk of churning.

Regarding the power of the optimal policy, we compare it to the industry status quo that matches players with the same skill level together (skill-based matchmaking). We prove the benefit of optimizing the matchmaking system grows linearly with the number of skill levels. We then use our framework to analyze two common but controversial interventions to increase engagement: adding AI bots and a pay-to-win system. We show that optimal matchmaking may reduce the number of bots needed significantly without loss of engagement. The pay-to-win system can influence player engagement positively when the majority of players are low-skilled. Surprisingly, even non-paying low-skilled players may be better off in some scenarios. Finally, we conduct a case study with real data from an online chess platform. We show the optimal policy can improve engagement by 4-6% or reduce the percentage of bot players by 15% in comparison to skill-based matchmaking.

Bio: Xiao Lei is a fourth-year PhD of Operations Research at Columbia University. His research interests include video game analytics, revenue management and pricing and social operations management. His work has been awarded INFORMS Service Science Best Student Paper Award and CSAMSE Best Paper Award. In the summer of 2020, he worked as a data science intern at Activision Blizzard.

Title:  An Empirical Study of Time Allotment and Delays in E-commerce Delivery

Abstract: Our paper examines how different facets of eCommerce delivery data can be used to develop and improve delay prediction models. We apply a mix of causal inference and machine learning (e.g., random forest) models to a comprehensive, large-scale dataset spanning user, delivery, and order information from JD.com, one of the largest eCommerce companies in China. In doing so, we first analyze how duration of each leg of the delivery, time allotted for each leg, and probability of delay relate to each other. Then, we fit random forest models to predict delays and identify primary predictors for such delays. Testing random forest models with different feature sets shows that including information about the earlier leg or warehouse package load can significantly improve the accuracy of the prediction model. Our prediction models suggest that managers can leverage various operational data to identify delays early on to prevent the orders from being delayed.

Bio: MoonSoo is a recent PhD graduate of Harvard Business School's Tech & Operations Management Unit and currently works at Walmart eCommerce as an Advanced Analytics Manager. His research interests span consumer protection, retail analytics, and service management. Prior to receiving his PhD, he worked as an operations analyst at various organizations, including UC-Berkeley Admissions Office, Disneyland Resort, startups, and a consulting firm for gaming industry.

Title:  Restaurant Ranking in Food Delivery Platforms with Unknown Demand

Abstract: Expansion of food delivery platforms in the past few years has created new interest in studying their service operations, which has enriched the field with new and challenging problems. In this work, we consider a food delivery platform that needs to make decisions over discrete time horizon. At each time the platform should jointly pick and display (possibly different) restaurant rankings to its users in different neighborhoods and route the drivers to serve the demand. The generated demand for a restaurant is specified based on both its positions in the displayed restaurant rankings and an unknown \emph{attraction parameter} associated with that restaurant's popularity; therefore, knowing the ranking we can use the number of online orders as feedback information for learning of these parameters. Being equipped with this mechanism, our main goal is to design a polynomial time online learning algorithm for the platform to maximize its profit---i.e., the generated revenue net routing cost. Besides conceptual modeling contributions, our main technical contribution is an algorithm that implicitly learns the attraction parameters and achieves a near-optimal multiplicative and vanishing additive regret for the aforementioned objective. We also evaluate the performance of our algorithm using experimental results over real-world data sets and observe that it also performs well in practice.

Bio: Fatemeh (Niusha) Navidi is a principal researcher at the University of Chicago Booth School of Business. Her research involves studying sequential decision making processes in stochastic and online optimization. She has been working on design and analysis of approximation algorithms for ranking, routing and classification problems in this setting. Currently, her main focus is on online learning algorithms and analyzing their performance using both theoretical and computational approaches. Prior to joining Chicago Booth, she was a PhD candidate at the University of Michigan in Industrial and Operations Engineering working in the same field.

Title: Time and the Value of Data

Abstract:  In this presentation we investigate the effectiveness of time-dependent data in improving the quality of AI-based products and services. Time-dependency means that data loses its relevance to problems over time. This loss causes deterioration in the algorithm's performance and, thereby, a decline in created business value. We model time-dependency as a shift in the probability distribution and derive several counter-intuitive results. 

We, theoretically, prove that even an infinite amount of data collected over time may have limited relevance for predicting the future, and an algorithm that is trained on a current dataset of bounded size can attain a similar performance. Moreover, we prove that increasing data volume by including older datasets may put a company in a disadvantageous position. 

Having these results, we answer questions on how data volume creates a competitive advantage. We argue that time[1]dependency weakens the barrier to entry that data volume creates for a business. So much that competing firms equipped with a limited, but sufficient, amount of current data can attain better performance. This result, together with the fact that older datasets may deteriorate algorithms' performance, casts doubt on the significance of first-mover advantage in AI-based markets. 

We complement our theoretical results with an experiment. In the experiment, we empirically measure the value loss in text data for the next word prediction task. The empirical measurements confirm the significance of time[1]dependency and value depreciation in AI-based businesses. For example, after seven years, 100MB of text data becomes as useful as 50MB of current data for the next word prediction task.

Bio: Ehsan Valavi is a Ph.D. candidate in Technology and Operations Management at Harvard Business School. His research interest is at the interface of information systems, operations management, and strategy. He is currently interested in studying the growth of digital firms and the challenges they face in various business areas. His recent research has focused on the scalability of Artificial Intelligence (AI) based solutions and the value of data for digital firms.  He completed his undergraduate studies in Electrical Engineering (Telecommunications) at the University of Tehran and has a master's degree in communication systems from the Swiss Federal Institute of Technology at Lausanne (EPFL). He also holds another master's degree in Decision, Risk, and Operations Management from Columbia Business School.

Title: A New Approach for Vehicle Routing with Stochastic Demand: Combining Route Assignment with Process Flexibility

Abstract: We propose a new approach for the vehicle routing problem with stochastic demands for the case in which customer demands are revealed before vehicles are dispatched. Our approach combines ideas from vehicle routing and manufacturing process flexibility to propose overlapped routing strategies with customer sharing. We characterize the asymptotic performance of the overlapped routing strategies under probabilistic analysis. Using the characterization, we demonstrate that our overlapped routing strategies perform close to the theoretical lower-bound derived from the reoptimization strategy, and significantly outperform the routing strategy without overlapped routes. The effectiveness of the proposed overlapped routing strategies in non-asymptotic regimes is further verified through numerical analysis.

Bio: Hanzhang Qin is a Ph.D. candidate in Computational Science and Engineering under supervision of Professor David Simchi-Levi. He is affiliated with Laboratory for Information & Decision Systems and Center for Computational Science & Engineering at MIT. He holds two master's, one in EECS and one in Transportation both from MIT. Prior to attending MIT, Hanzhang received two bachelor degrees in Industrial Engineering and Mathematics from Tsinghua University, where he was advised by Professor Hai Jiang and Professor Liping Zhang for his undergraduate theses.

Title: Blockchain monitored debt and capital structure under moral hazard

Abstract: The unobservability of the borrower's action is a major reason that smaller enterprises face high financing costs and credit rationing. We show that bank loans fully monitored by blockchain allow poor firms with low working capital to eliminate this agency cost. Interestingly, this is achieved by financing all production using fully monitored debt and leaving all internal capital unused; because the use of private, unmonitored internal capital creates unobservability whereas all-debt financing provides full transparency of operations. In contrast, rich firms find it costly to eliminate moral hazard via transparency and they prefer a mix of internal capital and unmonitored debt to finance production. We identify the working capital level at which a firm is indifferent between using all-debt or mixed financing for production. A poor firm with working capital below this indifference level strictly prefers all-debt financing and a rich firm above the level strictly prefers a debt-equity mix. We extend our results to a supply chain and show that the entire supply chain benefits from the use of monitored debt by an individual firm. The "inefficient" bankruptcy cost can create value under blockchain by mitigating the deadweight loss due to decentralization.

Bio: Jie Ning, PhD, joined the Weatherhead School of Management in 2013 after receiving her doctoral degree from the University of Michigan. Her research focuses on how firms (should) make operational decisions in the presence of financial constraints and economic incentives. She is particularly interested in studying the dynamic coordination between the operational and financial decisions of a single firm and the interaction between multiple financially constrained firms. Ning’s research extends to empirical operations management; interfaces of operations, finance, and economics; risk management; and stochastic optimization. Ning has worked with a number of Fortune 500 companies on a combination of empirical and theoretical research projects.

Jie received her PhD in industrial and operations engineering in 2013. She holds a MS in physics, a MSE in industrial and operations engineering from University of Michigan, and a BS in physics from the University of Science and Technology of China. She instructs students in Operations Research and Supply Chain Management as well as other courses.

Title: Design and Analysis of Switchback Experiments

Abstract: Switchback experiments, where a firm sequentially exposes an experimental unit to random treatments, are among the most prevalent designs used in the technology sector, with applications ranging from ride-hailing platforms to online marketplaces. Although practitioners have widely adopted this technique, the derivation of the optimal design has been elusive, hindering practitioners from drawing valid causal conclusions with enough statistical power. We address this limitation by deriving the optimal design of switchback experiments under a range of different assumptions on the order of the carryover effect --- the length of time a treatment persists in impacting the outcome. We cast the optimal experimental design problem as a minimax discrete optimization problem, identify the worst-case adversarial strategy, establish structural results, and solve the reduced problem via a continuous relaxation. For switchback experiments conducted under the optimal design, we provide two approaches for performing inference. The first provides exact randomization based p-values, and the second uses a new finite population central limit theorem to conduct conservative hypothesis tests and build confidence intervals. We further provide theoretical results when the order of the carryover effect is misspecified and provide a data-driven procedure to identify the order of the carryover effect. We conduct extensive simulations to study the empirical properties of our results and conclude with practical suggestions.