Activités

Centre de recherche sur l’intelligence² en gestion de systèmes complexes

Cerasela Crisan (Vasile Alecsandri University of Bacău): New strategy for parcel deliveries using a truck-and-drone system

Mercredi 22 avril 2026, 11h - 12h30
C-3150 (Pavillon de l'Entrepreneuriat et de l'innovation, 1250 rue Sanguinet)
Abstract du séminaire : Hybrid delivery systems—specifically the synergy between trucks and drones—emerge as an efficient solution for last-mile optimization, in the context of ecommerce development and urban congestions. By employing a 'truck-first, drone-second' hierarchical strategy, these models separate ground routing from aerial allocation, ensuring both scalability and clarity. Utilizing real-world road data and two-stage heuristics, the proposed approach balances computational speed with high-quality outcomes. The proposed drone integration significantly impacts delivery times and supports operational agility, offering a robust framework for sustainable modern logistics.

Gloria Cerasela Crişan received the degree in Informatics from the University of Bucharest, Romania, in 1986, and the Ph.D. degree in Informatics from the Alexandru Ioan Cuza University of Iaşi, Romania, in 2008. Since 2016, she has been an Associate Professor with the Department of Mathematics and Informatics, Faculty of Sciences, Vasile Alecsandri University of Bacău, Romania. Her research interests include solving combinatorial optimization problems, with applications in transportation and logistics problems.

Eliass Fennich (Laval) : Exact Solution of the Quadratic Knapsack Problem with Lifted Cover Inequalities

Jeudi 30 octobre 14h
Salle à confirmer
Abstract du séminaire : The Quadratic Knapsack Problem (QKP) is a fundamental combinatorial optimization problem with applications in portfolio selection, resource allocation, and related areas. It generalizes the classical Knapsack Problem (KP) by introducing quadratic interactions between items, which significantly increase the problem’s complexity. A key challenge for obtaining exact solutions to the QKP is the weakness of its linear programming relaxation. We propose to extend the classical lifted cover inequalities from the KP to the quadratic setting by using the interaction graph of the QKP to tighten the linear programming relaxation. Cover inequalities originate in early work on the knapsack polytope by Egon Balas, Laurence Wolsey, and Eitan Zemel. A cover inequality states that if a subset of items has a total weight exceeding the capacity, then not all of them can be chosen simultaneously, and lifting strengthens this inequality by adding coefficients for non-cover variables, extending it to a higher variable space while preserving validity. In the late 1980s, Zemel introduced a fast dynamic-programming-based, sequence-dependent lifting scheme. Building on that framework, we propose lifted edge-cover inequalities tailored to the QKP’s interaction graph and integrate them into a branch-and-cut algorithm. We define a natural notion of cover sets on the QKP interaction graph and show how it induces strong linear inequalities in the edge variable space. We generalize Zemel’s sequence-dependent lifting to the graph setting, providing an explicit procedure that lifts edge variables and ensures validity with respect to the quadratic interactions. We then develop a sequence-independent lifting framework that yields strong coefficients, avoids dependency on the lifting sequence, and remains efficiently computable through tight polynomial-time solution bounds for subproblems required in computing the proposed inequalities. Finally, we propose an efficient separation routine that generates heuristically the most violated inequalities. Computational experiments on both classic and newly generated benchmarks show that the proposed lifted inequalities substantially tighten relaxations, improving average optimality gaps by up to 90%. This is achieved while exploring less than half the number of nodes compared with CPLEX and incurring no significant overhead. Across multiple instance classes, our framework also outperforms state-of-the-art sequence-independent lifting schemes developed for the KP. Overall, the results present lifting as a clean and effective mechanism for embedding complex combinatorial structure into linear inequalities that are necessary for efficient integer programming algorithms.

Eliass Fennich est étudiant au doctorat à l'université Laval, Québec.

Santiago Mora Cruz (Tec de Monterrey / CRI2GS) : When Hospitals Run Out of Medicine: Optimization and Learning Approaches for Smarter Inventory Management

Mercredi 24 septembre 2025 11h - 12h
A-2625 (Pavillon Hubert-Aquin, 400 Sainte-Catherine E)
Abstract du séminaire : Healthcare providers face a persistent dilemma: keep too much medicine in stock and risk waste, or keep too little and face costly shortages. Traditional optimization methods, while elegant and practical, often assume a perfect foresight of demand and fail to capture the uncertainty and randomness of the real world. In this project, we explore three approaches to pharmacy inventory management: deterministic optimization, stochastic optimization, and reinforcement learning. Using synthetic demand data inspired by a university clinic, we benchmark each method on cost efficiency and resilience to uncertainty.

Santiago Mora Cruz is a third-year student at Tec de Monterrey and a visiting student at the CRI2GS, funded by a MITACS Globalink scholarship.

Asgard Mendoza (Tec de Monterrey / CRI2GS): Smarter Routes: A Machine Learning Approach to the Capacitated Arc Routing Problem

Mercredi 24 septembre 2025 10h - 11h
A-2625 (Pavillon Hubert-Aquin, 400 Sainte-Catherine E)
Abstract du séminaire : Every package you receive, every street swept, and every road gritted in the winter is part of a complex routing puzzle. While the Capacitated Vehicle Routing Problem (CVRP) is a well-studied topic, its lesser-known relative, the Capacitated Arc Routing Problem (CARP), is equally vital. This talk will explore what makes the CARP a unique and important challenge in Operations Research. We'll examine why both of these NP-Hard problems rely on heuristic algorithms for real-world solutions and how we can enhance these heuristics using Machine Learning. The impact of these improvements is far from theoretical: for example, a 1% increase in efficiency for waste pickup in Montreal could save the city more than one million Canadian dollars annually. This demonstrates how even small algorithmic advancements can translate into substantial economic and environmental gains.

Asgard Mendoza is a third-year student at Tec de Monterrey and a visiting student at the CRI2GS, funded by a MITACS Globalink scholarship.

Cerasela Crisan (Vasile Alecsandri University of Bacău): Efficient Deliveries with Truck-and-Drone Transportation Systems.

Vendredi 9 mai 2025, 11h – 12h30, DS-3650 (Salle départementale AOTI)
Lien Zoom:
Abstract du séminaire : The final segment of the parcel delivery process, commonly referred to as last-mile delivery, is characterized by logistical complexity and has a significant influence on customer satisfaction. Traditionally, this stage is executed using conventional modes of transportation such as cars, trucks, boats or bicycles. In recent years, Unmanned Aerial Vehicles (UAVs), commonly known as drones, have emerged as a novel alternative. Unlike traditional transport modes, drones operate independently of road infrastructure and offer several distinct advantages, including low weight, minimal energy consumption, high velocity, and rapid deployment capabilities. However, these benefits are counterbalanced by inherent limitations such as reduced payload capacity, limited operational range, and increased susceptibility to adverse weather conditions—contrasting sharply with the robustness and scalability of traditional delivery vehicles. Hybrid delivery systems, combining trucks and drones, have been under active investigation and practical deployment for over a decade. These systems have demonstrated substantial improvements in terms of cost-efficiency, environmental sustainability, and customer satisfaction, thereby delivering value across individual, organizational, and societal dimensions. This presentation aims to provide a comprehensive overview of both theoretical models and practical implementations related to integrated truck-and-drone delivery networks, highlighting their potential and challenges within the evolving logistics landscape.

Gloria Cerasela Crişan received the degree in Informatics from the University of Bucharest, Romania, in 1986, and the Ph.D. degree in Informatics from the Alexandru Ioan Cuza University of Iaşi, Romania, in 2008. Since 2016, she has been an Associate Professor with the Department of Mathematics and Informatics, Faculty of Sciences, Vasile Alecsandri University of Bacău, Romania. Her research interests include solving combinatorial optimization problems, with applications in transportation and logistics problems.

Kenza Alioui (UQAM): Production routing problem with simultaneous pickup and delivery

Jeudi 1 mai 2025, 14h – 14h30, DS-3650 (Salle départementale AOTI)
Lien Zoom:
Abstract du séminaire :
This work presents a new Production Routing Problem (PRP) variant in a two-echelon supply chain with multiple production plants, distribution centers (DCs), and retailers. We integrate the reverse logistics by collecting recyclable packaging from retailers to plants, aiming to minimize total costs, including production, inventory holding, and transportation costs, over a multi-period horizon. The problem involves two transportation layers: from plants to DCs and from DCs to retailers. Each layer is modeled as a Vehicle Routing Problem with Simultaneous Pickup and Delivery (VRPPD) using a heterogeneous vehicle fleet, where vehicles distribute products and collect packaging. Retailers’ demand and packaging return quantities are predetermined, allowing for proactive planning of production, inventory, and transportation. Inventory management is considered at all stages of the supply chain, with capacity and holding costs applied to all facilities. Each plant produces a single product, allowing DCs to be visited multiple times per period, whereas retailers are visited at most once per period. In this study, we aim to develop a mixed integer linear programming (MILP) model and solution methods to address this complex problem, integrating production, inventory, distribution, and routing problem with simultaneous pickup and delivery and reverse logistics.

Kenza Alioui est étudiante au doctorat au département d'Analytique, Opérations et Technologies de l'Information de l'École des Sciences de la Gestion de l'Université du Québec à Montréal (ESG-UQAM).

Frédéric Quesnel (UQAM): The crew split problem for team crew rostering.

Jeudi 10 avril 2025, 14h – 15h30, DS-2220
Lien Zoom:
Abstract du séminaire : The Crew Pairing Problem (CPP) is an optimization problem at the heart of the aviation industry, focused on creating the minimum-cost set of crew pairings. These pairings are sequences of flights and rest periods that together form the work schedule for a crew over several days. The goal is to cover all flights within a given period while minimizing costs. Once the optimal pairings are identified, they are assigned to individual crews in a next step. Airlines often benefit when cabin crews work together as a team for entire pairings. This approach not only boosts team morale and efficiency but also enhances the robustness of the flight schedule. However, varying crew requirements for different flights—such as the number of crew members and their specific roles on the plane—can make it challenging to keep a team intact across consecutive flights. In such cases, forming a core team to cover most of the requirements and supplementing it with additional crew members can be advantageous. Despite being recognized as a significant issue, finding the optimal way to form these teams has not been systematically addressed until now. In this presentation, we introduce the Crew Split Problem (CSP), a novel approach to determine the composition of core teams for each flight and identify the complementary crews that will join them. The CSP is a multi-objective optimization problem that seeks to maximize the size of each core team, minimize connection times, and reduce deadhead costs. Additionally, our model incorporates the concept of downranking, allowing higher-ranked crew members to work in lower-ranked positions if necessary. We demonstrate the effectiveness of the Crew Split Problem through a large-scale real-world application, highlighting its potential benefits for the airline industry.
Frédéric Quesnel est professeur de recherche opérationnelle au département d'Analytique, Opérations et Technologies de l'Information de l'École des Sciences de la Gestion de l'Université du Québec à Montréal (ESG-UQAM). Il a obtenu un doctorat en mathématiques appliquées à Polytechnique Montréal. Ses travaux portent sur la résolution de problèmes combinatoires de grande taille, principalement dans le domaine de l'aviation. Il s'intéresse également au développement d'algorithmes primaux pour la programmation en nombres entiers. Récemment, Frédéric a commencé à explorer de nouvelles applications de l'apprentissage automatique pour améliorer les algorithmes de branchement-et-prix utilisés pour résoudre ces types de problèmes.

Philippe Goulet Coulombe (UQAM): Dual Interpretation of Machine Learning Forecasts

Mardi 11 mars 2025, 15h – 16h30, DS-3650 (salle départementale AOTI)
Lien Zoom:
Abstract du séminaire : Machine learning predictions are typically interpreted as the sum of contributions of predictors. Yet, each out-of-sample prediction can also be expressed as a linear combination of in-sample values of the predicted variable, with weights corresponding to pairwise proximity scores between current and past economic events. While this dual route leads nowhere in some contexts (e.g., large cross-sectional datasets), it provides sparser interpretations in settings with many regressors and little training data-like macroeconomic forecasting. In this case, the sequence of contributions can be visualized as a time series, allowing analysts to explain predictions as quantifiable combinations of historical analogies. Moreover, the weights can be viewed as those of a data portfolio, inspiring new diagnostic measures such as forecast concentration, short position, and turnover. We show how weights can be retrieved seamlessly for (kernel) ridge regression, random forest, boosted trees, and neural networks. Then, we apply these tools to analyze post-pandemic forecasts of inflation, GDP growth, and recession probabilities. In all cases, the approach opens the black box from a new angle and demonstrates how machine learning models leverage history partly repeating itself.
Philippe Goulet Coulombe est professeur adjoint d'économie à l'Université du Québec à Montréal (UQAM). Ses recherches portent sur l'économétrie, l'apprentissage automatique, la macroéconomie, la finance et les changements climatiques, avec un accent particulier sur le développement de nouveaux algorithmes d'apprentissage qui améliorent l'efficacité statistique et l'interprétabilité dans l'analyse des séries temporelles. Il est titulaire d'un doctorat en économie de l'Université de Pennsylvanie et est rédacteur en chef adjoint de l'International Journal of Forecasting.

Mark Purdon (UQAM): Capacité de planification et autorité de gouvernance pour la décarbonisation des transports régionaux métropolitains : une comparaison entre la Californie et le Québec

Vendredi 14 février 2025, 14h – 15h30, DS-3650 (salle départementale AOTI)
Lien Zoom:
Abstract du séminaire : Dans cet article, nous présentons une étude comparative de la décarbonisation des systèmes de transport des grandes régions métropolitaines de Montréal et de Los Angeles. La Californie a adopté le Sustainable Communities and Climate Protection Act, connu sous le nom de « SB375 », qui vise explicitement à réduire les émissions de gaz à effet de serre provenant des transports dans les régions métropolitaines. S'appuyant sur un cadre institutionnel existant aux États-Unis pour la planification régionale des transports et la qualité de l'air, SB375 exige que les organisations de planification régionale démontrent que les mesures de transport prévues permettront aux régions métropolitaines d'atteindre les « objectifs climatiques du plan régional » grâce à l'utilisation d'une modélisation technico-économique sophistiquée. Toutefois, des questions se posent quant à l'efficacité de SB375 et à la pertinence de son transfert à d'autres juridictions. La comparaison avec les efforts de décarbonisation des transports dans la région du Grand Montréal permet d'identifier les défis en matière de gouvernance. Notre étude est basée sur l'examen des documents politiques disponibles ainsi que sur des entretiens réalisés en 2020-2021 avec vingt-trois experts et praticiens connaissant bien la politique et la modélisation des transports dans les régions métropolitaines de Los Angeles et de Montréal. Les résultats sont interprétés à l'aide d'un cadre analytique dans lequel nous distinguons la capacité de planification (y compris la capacité administrative et la modélisation technico-économique) ainsi que l'autorité de gouvernance sur le financement des systèmes de transport et l'utilisation des sols.
Mark Purdon est un expert dans le domaine émergent de la politique comparée du changement climatique, qui combine des éléments de politique comparée, de politique publique comparée et de relations internationales. Il s'intéresse particulièrement à la relation entre la décarbonisation et l'économie politique, et possède une vaste expérience de recherche dans les pays développés et en développement dans les domaines du financement climatique, des énergies renouvelables, des transports, de la foresterie et de l'utilisation des terres. Il a rejoint le Département de stratégie, responsabilité sociale et environnementale de l'ESG UQÀM en tant que professeur en 2018 où il est titulaire de la Chaire de décarbonisation depuis 2021.

Amir M. Fathollahi-Fard (UQAM): The Scheduled Service Network Redesign Problem

Mercredi 22 janvier 2025, 10h30 – 12h, CIRRELT
Abstract du séminaire : Natural disasters, including earthquakes, floods, hurricanes, and landslides, have profound impacts on transportation service networks, particularly disrupting rail and road transportation modes. Such events often result in the destruction of critical infrastructure, such as terminals and their connections through bridges, tunnels, and rails, leading to prolonged service interruptions. To address these challenges in transportation service networks, this paper introduces an innovative extension of the classical Scheduled Service Network Design (SSND) problem, referred to as the Scheduled Service Network Redesign (SSNR) problem. The SSNR problem optimizes post-disaster networks by selecting demands and assigning them to available services, maximizing total profit. Comprehensive computational experiments using 30 test instances and sensitivity analyses are conducted to evaluate the effectiveness of the SSNR problem. The results offer actionable managerial insights for enhancing the resilience of transportation service networks against natural disasters.
Amir M. Fathollahi-Fard est chercheur postdoctoral à l'Université du Québec à Montréal, Montréal, Canada, et professeur adjoint à l'Université de Science et Technologie de Macao, Taipa, Macao. Il a obtenu son doctorat de l'École de Technologie Supérieure, Université du Québec, Montréal, Canada. Ses intérêts de recherche incluent la gestion de la chaîne d'approvisionnement, la logistique durable, la conception de réseaux, la planification et la gestion des opérations. Avec plus de 100 publications dans des revues renommées telles que Computers & Operations Research, Annals of Operations Research, International Journal of Production Research, et d'autres, Amir a été reconnu comme l'un des 1 % des chercheurs les plus cités en 2024, selon la base de données Web of Science.

Timo Gschwind (Kaiserslautern–Landau): Branch-and-Price-Based Solution of Two Set-Partitioning Problems with Knapsack-Constrained Sets

Jeudi 16 janvier 2025, 14h – 15h30, DS-3245
Lien Zoom: https://uqam.zoom.us/j/88558032616
Abstract du séminaire : We consider the effective solution of the order batching problem (OBP), a fundamental problem in warehouse operations, and the set-union bin packing problem (SUBP), which has manifold applications in flexible manufacturing and beyond, via branch and price (B&P). From a modelling point of view, the OBP and the SUBP both seek for the cost-optimal partition of a set of tasks into subsets that are only restricted by a knapsack constraint. In addition, both problems feature a specific non-trivial characteristic. In the OBP, the cost of a subset is a non-separable function in the comprised tasks. In the SUBP, a subset’s consumption of the knapsack capacity is a non-separable function of the comprised tasks. We propose a generic B&P approach that can – with modest adaptations – be applied to solve both the OBP and the SUBP as well as other related problems. The corresponding column-generation pricing problems are modeled as shortest path problems with resource constraints that can be adapted to handle the implications from nonrobust valid inequalities and branching decisions. They are solved by means of effective labeling algorithms that rely on strong completion bounds rather than the application of dominance rules. We demonstrate the effectiveness of the proposed approach in an extensive computational campaign on established benchmarks and solve to optimality hundreds of instances that were previously unsolved.

Nadia Tahmasbi (UQAM): Optimizing freight transport through intelligent match-making under uncertainty

Mardi 26 mars 2024, 13h – 14h30, DS-1525
Abstract du séminaire : The transportation sector is actively adapting to meet growing demands for higher efficiency, profitability, and a lower environmental footprint. At the core of this trend are two-sided logistics platforms like Uber Freight, which facilitate interactions between shippers and carriers. In our study, we address a digital platform for logistics service providers, defined as an integrated multi-stakeholder freight transportation system where the demand of many shippers (e.g., retailers, distributors, and manufacturers) are being matched to the capacity offer from many carriers (e.g., freight carriers, terminal managers and other logistics service providers) using one centralized intelligent decision support entity. From a system design perspective, this platform has a Many-to-One-to-Many structure, thus, we call this logistics platform an M1M system. The intelligent platform matches the demand and supply sides together in an efficient way while satisfying the demand timely, respecting the operational restrictions and generating profit for carriers and the platform. We investigate the tactical planning required for M1M systems under uncertainty of demand volume and the capacity of service offers.
Nadia Tahmasbi is a Ph.D. candidate in the business administration program jointly offered by Université du Québec à Montréal (UQAM), McGill University, Concordia University, and HEC Montreal.