The TutORials in Operations Research series is published annually by INFORMS as an introduction to emerging and classical subfields of operations research and management science. These chapters are designed to be accessible for all constituents of the INFORMS community, including current students, practitioners, faculty, and researchers. The publication allows readers to keep pace with new developments in the field and serves as augmenting material for a selection of the tutorial presentations offered at the INFORMS Annual Meeting.
Nanoretail Operations in Developing Markets
Across much of the developing world, family-operated nanostores provide daily grocery needs to billions of poorly paid consumers. This highly fragmented retail channel is of critical importance to consumer brands as in many markets this is the largest retail channel. We characterize the empirical context in which these stores operate, as well as the intricate operations that manufacturers and distributors put in place to supply them with their goods. We then elaborate on modeling operations execution and operations strategies to expose critical tradeoffs that are distinct from those in organized retail in developed markets. We discuss research results that have demonstrated why many manufacturers choose to serve this market directly and at high frequency, why manufacturers invest considerable effort in deploying sales agents networks, how this channel manages to remain competitive with modern organized retail such as convenience store chains, and how digitization and novel financing solutions provide a further competitive advantage to the nanoretail channel. Finally, we discuss how to conduct research in this retail segment and provide examples of novel contexts and business models that may open up new areas of nanoretail research.
Speakers: Jan C. Fransoo, Rafael Escamilla, and Jiwen Ge
Humanitarian Operations and Earmarked Funding
Earmarked funding, also referred to as restricted funding, is one of the main characteristics of humanitarian operations. Earmarked funding can be defined as the donors’ contributions to humanitarian organizations to be used for a specific purpose. This is in contrast to flexible funding, which can be used for any purpose. This tutorial introduces the trade-off between total donations and operational performance due to humanitarian earmarking. The tutorial explains why allowing donors to earmark their contributions helps organizations increase fundraising effectiveness. It also explains why earmarking hurts organizations’ per-dollar performance. Because humanitarian organizations’ utility increases in total donations and per-dollar (or any other currency) performance, the best fundraising strategy for the organizations, collecting earmarked or flexible funds, is not apparent. Moreover, this tutorial argues that earmarked funding is here to stay and discusses models that reduce the negative effect of earmarked funding. Then, it proposes that a thoughtful mix of earmarked and flexible donations may be the best way for humanitarian organizations to fund their operations. The balance between earmarked and flexible funds depends on reducing the negative operational consequences of earmarked funds. The tutorial concludes by identifying areas for future research on humanitarian operations and earmarked funding.
Speaker: Alfonso J. Pedraza-Martinez
Nonconvex, Nonsmooth, and Nonregular Optimization: A Computational Framework 9299
Algebraic modeling languages presently lack the ability to effectively support the
formulation and solution of nonconvex and nonsmooth optimization problems. Since
an arbitrary problem of this kind is intractable, any hope to achieve practically useful
solutions would rely on means to convey specific problem structure to an algorithm.
In this tutorial, we present a framework for specifying nonconvex, nonsmooth, and
nonregular problems within an algebraic modeling language that makes available the
key structural properties to an algorithm. It also facilitates experimentation with
different model formulations and algorithmic approaches. The framework entails a
change of mindset away from the traditional formulation of objective and constraint
functions, and instead asks the analyst to specify a basic feasible set, a basic objective
function, one or more monitoring functions, and several performance functions. Eleven
examples ranging from goal programming to variational inequalities and engineering
risk analysis illustrate the practical implications of the framework.
Speaker: Michael C. Ferris, Olivier Huber, and Johannes O. Royset
Experimental Design for causal Inference Through an Optimization Lens
AThe study of experimental design offers tremendous benefits for answering causal ques tions across a wide range of applications, including agricultural experiments, clinical trials, industrial experiments, social experiments, and digital experiments. While valu able in such applications, the costs of experiments often drive experimenters to seek more efficient designs. Recently, experimenters have started to examine such efficiency questions from an optimization perspective, as experimental design problems are fun damentally decision-making problems. This perspective offers a lot of flexibility in leveraging various existing optimization tools to study experimental design problems. thus aims to examine the foundations of experimental design problems in the context of causal inference as viewed through an optimization lens.
Speaker: Jinglong Zhao
Machine Learning Methods for Large Population Games
In this tutorial, we provide an introduction to machine learning methods for finding Nash equilibria in games with large number of agents. These types of problems are important for the operations research community because of their applicability to real life situations such as control of epidemics, optimal decisions in financial markets, electricity grid management, or traffic control for self-driving cars. We start the tutorial by introducing stochastic optimal control problems for a single agent, in discrete time and in continuous time. Then, we present the framework of dynamic games with finite number of agents. To tackle games with a very large number of agents, we discuss the paradigm of mean field games, which provides an efficient way to compute approximate Nash equilibria. Based on this approach, we discuss machine learning algorithms for such problems. First in the context of discrete time games, we introduce fixed point based methods and related methods based on reinforcement learning. Second, we discuss machine learning methods that are specific to continuous time problems, by building on optimality conditions phrased in terms of stochastic or partial differential equations. Several examples and numerical illustrations are provided along the way.
Speakers: Gökçe Dayanikli and Mathieu Laurière
Food Bank Operations: A U.S. Perspective on Humanitarian Food Assistance
Food banks are non-profit organizations with the primary mission of providing food assistance to the communities they serve. This assistance occurs through the collection, storage, and distribution of food to people in need through a complex supply chain network of donors and charitable agencies. Distribution of food is challenging in this environment, given the many resource constraints experienced at the financial, human, and material levels. Furthermore, the COVID-19 pandemic highlighted the important role these organizations play as demand for food assistance surged, while sources of supply were constrained. This tutorial provides an overview of food bank operations from a supply chain perspective. We specifically characterize the key stakeholders, product, and information flows within the food bank supply chain and draw from our prior experience with several U.S. food banks to delineate structural differences that exist among these supply chain networks. We further elucidate the influence of the supply chain network design on the organizations’ operational decisions and strategic direction with respect to equity, efficiency, effectiveness, diversity and inclusion. We present several studies that illustrate the role of descriptive, predictive, and prescriptive analytics in improving the distribution of food and reducing food waste, as well as provide insights for future research in this area.
Speakers: Lauren Davis, Irem Sengul Orgut, Steven Jiang, Eric Aft, Charlie Hale, Larry Morris, Jean Rykaczewski
Combining Large Language Models and OR/MS to Make Smarter Decisions
Operations Research/Management Science (OR/MS) capabilities can provide tremendous value in helping enterprises and individuals make smarter decisions. However, the creation and deployment of OR/MS-based decision-making solutions require significant time and expertise, making their widespread application challenging. Large language models (LLMs), exemplified by models such as ChatGPT, Gemini, and Claude, are deep neural network models encompassing billions to trillions of parameters. These models are pre-trained with a vast scope of general knowledge and are quickly adaptable to many downstream tasks. Beyond core capabilities like document summarization and code generation, LLMs exhibit emerging capabilities, such as learning new tasks from a few natural language examples. Generative AI technologies like LLMs, therefore, have transformative potential for many fields and professions. In this work, we explore the potential of LLMs and their capabilities to significantly improve the creation of OR/MS based decision-making solutions. After providing relevant technical background on LLMs, we show this potential through three concrete and detailed examples: using LLMs to significantly reduce the time required to create decision-making applications while improving their quality; using LLMs to extract structured information from unstructured text without the need to create new natural language processing models, a capability important, for example, for demand forecasting; and using LLMs to drive natural language-based interfaces enabling a business user to easily and flexibly interact with analytical models used for decision making. As LLMs are a new technology that introduces new risks, this work also provides guidelines for their productive, ethical and responsible use and describes ongoing developments relevant to the OR/MS profession. We hope this work will serve as a starting point both for the application of LLMs to the OR/MS use cases we described and as the starting point for exploring new and exciting integrations between LLMs and OR/MS, ultimately enabling the use of OR/MS to make smarter decisions in a much more widespread manner.
Speakers:
Segev Wasserkrug, Leonard David Jean Boussioux, and Wei Sun
Interventions for Patients with Complex Medical and Social Needs
Patients with multiple chronic conditions and social needs represent a small percentage of the population but have a disproportionate impact on healthcare costs and utilization. Organizations around the United States have created programs – often referred to as complex care interventions – to improve the health and well-being of such patients and reduce avoidable hospital and emergency department use. In this tutorial, we focus on two emerging themes in the field: (1) identifying clinically meaningful subgroups in complex care populations through unsupervised learning methods; and (2) describing the key operational features of interventions with an emphasis on staffing needs and the impact on patient outcomes. The material presented in this tutorial draws on the research of the Healthcare Operations Research Lab at the University of Massachusetts, Amherst, and its collaborating partners. To illustrate these themes and contextualize the details of complex care delivery, we use a range of patient-level examples, visualizations, descriptive summaries, case studies, and results from the clinical literature.
Speakers: Hari Balasubramanian, Sindhoora Prakash, Ali Jafari, Arjun Mohan, and Chaitra Gopalappa
Randomized Rounding Approaches to Online Allocation, Sequencing, and Matching
Randomized rounding is a technique that was originally used to approximate hard offline discrete optimization problems from a mathematical programming relaxation. Since then it has also been used to approximately solve sequential stochastic optimization problems, overcoming the curse of dimensionality. To elaborate, one first writes a tractable linear programming relaxation that prescribes probabilities with which actions should be taken. Rounding then designs a (randomized) online policy that approximately preserves all of these probabilities, with the challenge being that the online policy faces hard constraints, whereas the prescribed probabilities only have to satisfy these constraints in expectation. Moreover, unlike classical randomized rounding for offline problems, the online policy’s actions unfold sequentially over time, interspersed by uncontrollable stochastic realizations that affect the feasibility of future actions. This tutorial provides an introduction for using randomized rounding to design online policies,
Speaker: Will Ma
Measuring the Efficacy of Amazon’s Recommendation Systems
Amazon’s Fulfillment By Amazon (FBA) program provides assistance to Selling Partners (“sellers,” for short) in the form of information sharing, recommendations guiding seller actions (e.g., restock quantity recommendations, excess inventory recommendations), and delegated actions (e.g., automated removals of aged inventory). Amazon’s vision is to help sellers make better decisions and achieve better business outcomes.
In this tutorial, we consider the sophisticated optimization models Amazon employs to generate recommendations. For example, if a seller has excess inventory, Amazon recommends actions to increase their sell-through rate, such as creating a sale or Sponsored Product ad. We demonstrate how we measure the efficacy of these recommendation systems on seller-product outcomes (e.g., revenue, units shipped, and customer clicks on product listings, or “glance views”). Measuring such outcomes is a causal inference problem because we only observe each seller-product’s “factual” and not their “counterfactual” outcome. We employ causal machine learning methodologies such as double machine learning, causal forest, and doubly-robust forest to separate selection bias from a comparison of “treatment” and “control” sellers. For example, we find that aligning with the restock and excess inventory recommendations, on average, improves several seller-salient outcomes. We also present methods for measuring heterogeneity in the efficacy of these recommendations across seller and product segments, and estimate personalized benefits for each seller-product. Finally, through A/B testing, we find that sharing quantified efficacy information with sellers increases their adoption of Amazon recommendations. Sellers are responding to this messaging, and the duty to them is to rigorously identify causal estimates.
Speakers:
Ozalp Ozer, Serdar Simsek, Xiaoxi Zhao, Ethan Dee, and Vivian Yu
Digital Transformation in Transportation Systems: Navigating User Behavior and System Efficiency
This tutorial explores the impact of digitalization on today’s transportation systems, focusing on how emerging information technologies and pricing schemes are reshaping the travel behavior in congestible networks. Our focuses include: static and dynamic routing games, the impact of asymmetric information on network efficiency, and the design of incentives for system efficiency and equity. Through a combination of theoretical insights and empirical studies, the tutorial offers an in-depth analysis of models and tools for analyzing strategic user behavior, the role of information and incentive mechanisms in promoting socially desirable outcomes, and the application of these theories in real-world transportation systems.
Speakers: Haripriya Pulyassary and Manxi Wu
