Denis Sauré

Dynamic assortment learning under multinomial logit (MNL) demand couples revenue and information through a shared choice denominator. A Thompson sampling policy must decide how much of the item-level likelihood to carry. Filtering it to product-versus-outside comparisons yields one conjugate law per product, revisable after every customer, but discards the periods in which a displayed product lost the sale to another, precisely those that couple the attractions. Retaining the complete likelihood preserves that coupling, at a computational cost. We study this information-filtration trade-off at both ends and in between. At the filtered end, the probability that several sampled coordinates are optimistic at once can decay exponentially in the display capacity K, which is why existing analyses correlate the draws. Such correlation is not necessary: in linear-fractional form, assortment optimism is a single signed linear inequality in the attractions, so favorable draws offset unfavorable ones, and this assortment-level optimism yields gap-free frequentist regret O(K √ N T) for independent, variance-inflated draws. At the complete end, we construct a Gibbs sampler for the joint posterior and prove that exact draws attain Bayesian regret O(√ N T), uniformly in K. To our knowledge, both are the first guarantees of their kind. A computational study places a mean-field policy between the ends, separating posterior dependence from correlation imposed across draws.

We study a sequential stochastic shortest-path interdiction problem in which an interdictor repeatedly blocks arcs to increase an evader's routing cost. Each period, a cost vector is drawn from a fixed, unknown distribution, and the evader observes the realized costs before routing; the interdictor observes only the evader's chosen path. We derive an instance-dependent logarithmic regret lower bound, computed through a semi-infinite information-allocation problem that admits an exact finite exponential-cone reformulation under regularity conditions. This informational censoring creates observational equivalence: distinct cost distributions can induce identical path observations under some interdiction actions. The resulting dominated information property means the actions needed to distinguish confounded environments are inferior in expected value, so pure posterior sampling and value-optimistic rules can fail to learn all relevant alternatives and incur linear regret. We propose Identifying-Exploration Greedy policies, which deliberately sample an identifying action set and achieve logarithmic regret on identifiable instances with a positive optimality gap. Experiments on constructed instances, generated networks, and a border-infiltration case study show when path-only learning succeeds, when it fails, and why deliberate information acquisition is essential for logarithmic regret.

Durante décadas, el Virus Respiratorio Sincicial (VRS) ha representado una de las principales causas de morbilidad y mortalidad en lactantes en Chile y el mundo. La experiencia del año 2023, marcada por un brote anticipado y severo, visibilizó las limitaciones del sistema sanitario frente a este virus, generando un consenso transversal respecto a la necesidad de una estrategia más robusta, preventiva e intersectorial. En este contexto, la conformación de la Coalición VRS –un esfuerzo articulado entre ex autoridades sanitarias, sociedades científicas, universidades, expertos clínicos y técnicos– permitió diseñar e impulsar una serie de recomendaciones que fueron importantes en la toma de decisiones de las autoridades sanitarias de la época, materializándose el 2024, con la incorporación pionera del anticuerpo monoclonal Nirsevimab al Programa Nacional de Inmunizaciones. Chile se convirtió así en el primer país del hemisferio sur en ejecutar una estrategia nacional y universal de inmunización contra el VRS, alcanzando altas coberturas y reducciones sustantivas en hospitalizaciones. Este artículo busca documentar los resultados positivos de esa implementación, como también reflexionar críticamente sobre los factores que la hicieron posible, identificar brechas aún pendientes y en especial proyectar las decisiones necesarias para evitar retrocesos y avanzar hacia una política de prevención sostenible, equitativa y basada en evidencia frente al VRS y otras infecciones respiratorias infantiles.

During the COVID-19 crisis, the Chilean Ministry of Health and the Ministry of Sciences, Technology, Knowledge and Innovation partnered with the Instituto Sistemas Complejos de Ingeniería (ISCI) and the telecommunications company ENTEL, to develop innovative methodologies and tools that placed operations research (OR) and analytics at the forefront of the battle against the pandemic. These innovations have been used in key decision aspects that helped shape a comprehensive strategy against the virus, including tools that (1) provided data on the actual effects of lockdowns in different municipalities and over time; (2) helped allocate limited intensive care unit (ICU) capacity; (3) significantly increased the testing capacity and provided on-the-ground strategies for active screening of asymptomatic cases; and (4) implemented a nationwide serology surveillance program that significantly influenced Chile’s decisions regarding vaccine booster doses and that also provided information of global relevance. Significant challenges during the execution of the project included the coordination of large teams of engineers, data scientists, and healthcare professionals in the field; the effective communication of information to the population; and the handling and use of sensitive data. The initiatives generated significant press coverage and, by providing scientific evidence supporting the decision making behind the Chilean strategy to address the pandemic, they helped provide transparency and objectivity to decision makers and the general population. According to highly conservative estimates, the number of lives saved by all the initiatives combined is close to 3,000, equivalent to more than 5% of the total death toll in Chile associated with the pandemic until January 2022. The saved resources associated with testing, ICU beds, and working days amount to more than 300 million USD. Funding: This work was supported by the ANID PIA/APOYO [Grant AFB180003 and AFB220003], used to hire research assistants for developing solutions, information systems, data management, and training of field staff. Fieldwork was funded by the Ministry of Health of the Government of Chile. M. Goic acknowledges the financial support of Fondecyt [Project 1221711] and the Institute for Research in Market Imperfections and Public Policy [Grant IS130002 ANID]. D. Sauré acknowledges the financial support of Fondecyt [Project 1211407]. Finally, G. Weintraub and A. Carranza thank the Stanford RISE COVID-19 Crisis Response Faculty Seed Grant Program for helpful financial support.

We consider a framework for sequential bilevel linear programming in which a leader and a follower interact over multiple time periods. In each period, the follower observes the actions taken by the leader and reacts optimally, according to the follower’s own objective function, which is initially unknown to the leader. By observing various forms of information feedback from the follower’s actions, the leader is able to refine the leader’s knowledge about the follower’s objective function and, hence, adjust the leader’s actions at subsequent time periods, which ought to help in maximizing the leader’s cumulative benefit. We show that greedy and robust policies adapted from previous work in the max-min (symmetric) setting might fail to recover the optimal full-information solution to the problem (i.e., a solution implemented by an oracle with complete prior knowledge of the follower’s objective function) in the asymmetric case. In contrast, we present a family of greedy and best-case policies that are able to recover the full-information optimal solution and also provide real-time certificates of optimality. In addition, we show that the proposed policies can be computed by solving a series of linear mixed-integer programs. We test policy performance through exhaustive numerical experiments in the context of asymmetric shortest path interdiction, considering various forms of feedback and several benchmark policies.

In this paper, we describe our work in scheduling Argentina’s First Division professional football league, the Superliga. Following existing work in sports scheduling, we develop an integer programming model for the Superliga season schedule and then, solve it using a decomposition approach. Unlike previous work, this scheme is based on the creation and assignment of cluster patterns, which take advantage of the model’s geographically driven handling of sporting fairness. We also model the assignment of matches to specific dates and time slots while simultaneously considering various conditions relating to or imposed by game broadcasters, the government, and international tournament calendars. Our work was implemented to schedule the Superliga’s 2018–2019 and 2019–2020 seasons, achieving clear improvements on a number of criteria over the previous approach.

During the current COVID-19 pandemic, active testing has risen as a key component of many response strategies around the globe. Such strategies have a common denominator: the limited availability of diagnostic tests. In this context, pool testing strategies have emerged as a means to increase testing capacity. The efficiency gains obtained by using pool testing, derived from testing combined samples simultaneously, vary according to the spread of the SARS-CoV-2 virus in the population being tested. Motivated by the need for testing closed populations, such as long-term care facilities (LTCFs), where significant correlation in infections is expected, we develop a probabilistic model for settings where the test results are correlated, which we use to compute optimal pool sizes in the context of two-stage pool testing schemes. The proposed model incorporates the specificity and sensitivity of the test, which makes it possible to study the impact of these measures on both the expected number of tests required for diagnosing a population and the expected number and variance of false negatives. We use our experience implementing pool testing in LTCFs managed by SENAMA (Chile’s National Service for the Elderly) to develop a simulation model of contagion dynamics inside LTCFs, which incorporates testing and quarantine policies implemented by SENAMA. We use this simulation to estimate the correlation of test results among collected samples when following SENAMA’s testing guidelines. Our results show that correlation estimates are high in settings representative of LTCFs, which validates the use of the proposed model for incorporating correlation in determining optimal pool sizes for pool testing strategies. Generally, our results show that settings in which pool testing achieves efficiency gains, relative to individual testing, are likely to be found in practice. Moreover, the results show that incorporating correlation in the analysis of pool testing strategies both improves the expected efficiency and broadens the settings in which the technique is preferred over individual testing.

We study sequential shortest path interdiction, where in each period an interdictor with incomplete knowledge of the arc costs blocks at most [ ] arcs, and an evader with complete knowledge about the costs traverses a shortest path between two fixed nodes in the interdicted network. In each period, the interdictor, who aims at maximizing the evader’s cumulative cost over a finite time horizon, and whose initial knowledge is limited to valid lower and upper bounds on the costs, observes only the total cost of the path traversed by the evader, but not the path itself. This limited information feedback is then used by the interdictor to refine knowledge of the network’s costs, which should lead to better decisions. Different interdiction decisions lead to different responses by the evader and thus to different feedback. Focusing on minimizing the number of periods it takes a policy to recover a full information interdiction decision (that taken by an interdictor with complete knowledge about costs), we show that a class of greedy interdiction policies requires, in the worst case, an exponential number of periods to converge. Nonetheless, we show that under less stringent modes of feedback, convergence in polynomial time is possible. In particular, we consider different versions of imperfect randomized feedback that allow establishing polynomial expected convergence bounds. Finally, we also discuss a generalization of our approach for the case of a strategic evader, who does not necessarily follow a shortest path in each period.