Plenaries and Semi-Plenaries

This talk explores the critical divide between academic Operations Research and its application in complex industrial environments, with a focus on Amazon Web Services. In these settings, unprecedented scale and rapid delivery are not just desirable—they’re imperative. While academia often pursues algorithmic complexity and theoretical optimality, the industrial landscape demands a paradigm shift towards pragmatism and agility. Both aspects are neglected and often ignored in academia where complexity and problem-specific knowledge are sought for and valued over practical applications.

We confront the reality of multi-objective problems, soft constraints, and the ever-shifting sands of business requirements that characterize real-world scenarios. These factors necessitate a radical rethinking of traditional OR approaches.

Our proposition is bold yet practical: embrace heuristic solvers and simplified modeling techniques that prioritize speed, adaptability, and ease of implementation. This approach is particularly crucial when dealing with estimated input data, where the pursuit of mathematical optimality may be not just impractical, but potentially misleading.

Through a series of compelling case studies—ranging from classical routing and scheduling problems to large-scale virtual machine placement in Amazon EC2—we demonstrate the power of pragmatic methods in addressing real-world challenges. These examples illustrate how seemingly “suboptimal” solutions can yield robust, maintainable, and highly effective outcomes that balance performance with operational efficiency.

Ultimately, this presentation challenges the OR community to reconsider the true meaning of optimality in industrial contexts. We argue that a marginal optimality gap is a small price to pay for the immense benefits of enhanced flexibility, reduced operational complexity, and the ability to swiftly adapt to changing business requirements. This talk aims to spark a dialogue about aligning academic research with the pressing needs of industry, potentially reshaping the future of OR in practice.

Wind farms provide free and sustainable electricity, hence they play a central role in the green transition. The talk will show examples of how Decision Science has contributed to significantly bring down costs of green energy. We study the five phases of the life cycle: Area Selection, Wind farm design, Installation of a wind farm, Operations and maintenance, and End of life. We focus on the most interesting optimization models and solution methods, to give the reader an introduction to this exciting and growing research area. Several examples from our collaboration with Vattenfall will be shown to illustrate the impact of Decision Science.

Sustainability is a major concern in supply chain management, driving the need to integrate sustainable practices into supply chain design and operations. One approach is the development of closed-loop supply chains, which focus on material recovery and circularity, along with renewable energy supply chains, such as those supporting hydrogen-based systems. However, supply chains are inherently complex, and this complexity increases in sustainable networks due to the interplay of economic, environmental, and social objectives. To address these challenges, Operations Research (OR) methodologies provide a foundation for developing decision-support tools, incorporating modeling and optimization to enhance the design and management of sustainable networks. Within this context, this talk explores how optimization techniques can support strategic and tactical decision-making in sustainable supply chains. Additionally, it discusses key research challenges and future opportunities to advance truly sustainable supply chains.

Traditionally, operations (management) research has very much focused on changing operations to impact performance. With the growth in attention on the UN Sustainable Development Goals in our research, the definition of performance has been widened to include outcomes related to climate change, poverty, and labor conditions. However, in terms of actors this primarily focuses on operational actors, such as companies, NGOs, or hospitals. This has separated us from economics research that typically studies policy questions by governments and governmental institutions such as central banks. In this presentation, I will argue that as operations management researchers, we have an opportunity and a need to think more explicitly about the policy impact of our research, with an angle that distinguishes us from the traditional angles chosen in economics research.

I will discuss several related research projects, broadly within emergency and urgent healthcare services. This includes working with ambulance providers and the Indonesian Government to help them make critical decisions on the optimal types, capacities and geographical locations of response vehicles. Such factors directly impact on the probability of patient survival, ability to respond to major disasters, and the overall quality of care provided. There are however many challenges faced in Indonesia, including vast geographical areas, traffic congestion, inadequate numbers of ambulances and a lack of a co-ordinated service.

Over many decades the fields of forecasting, lot sizing and inventory management have developed in parallel. These fields are the foundation for today’s planning systems as embedded in Advanced Planning and Scheduling (APS) systems. The field of forecasting has a very strong empirical basis, as its methodology is derived to a large extent from mathematical statistics. The field of inventory management is rooted in the mathematical analysis of stochastic processes. Lot sizing is a well-developed branch of combinatorial optimization and MILP. In text books on Operations Research and Operations Management, separate chapters are devoted to these fields. As the real-life problem concerns the use of forecasts, inventory control rules and lot sizing mechanisms for multiple items in a multi-echelon inventory system, heuristics have been implemented in APS systems. Typically, a wide range of forecasting methods are offered, some lot sizing heuristics, and basic safety stock formulas derived under some service measure assumption.

In this lecture we discuss the caveats of this pragmatic approach, typically resulting in a discrepancy between target customer service and actual customer service. We provide an explanation for this phenomenon and propose a framework under which under the assumption of linear holding and penalty costs, we can derive the optimal parameters of all end-item control policies with only two long-run simulations, irrespective of the forecasting method and lot sizing mechanisms. We discuss possible extensions to setting parameters for control of upstream items in the multi-item multi-echelon setting.

Abstract will be submitted later

We investigate Bayesian optimization for mixed-feature spaces using both tree kernels and graph kernels for Gaussian processes. With respect to trees kernels, our Bayesian Additive Regression Trees Kernel (BARK) uses tree agreement to define a posterior over sum-of-tree functions. With respect to graph kernels, our acquisition function with shortest paths encoded allows us to optimize over graphs, for instance to find the best graph structure and/or node features. We formulate both acquisition functions using mixed-integer optimization and show applications to a variety of challenges in molecular design, engineering and machine learning.

The tree kernel work is joint with Toby Boyne, Alexander Thebelt, Jose Folch, Calvin Tsay, Robert Lee, Nathan Sudermann-Merx, David Walz, and Behrang Shafei.

The graph kernel work is joint with Yilin Xie, Shiqiang Zhang, Jixiang Qing, and Calvin Tsay.

Efficient logistics planning and routing are critical for modern transportation and warehouse management, where complex constraints and large-scale operations must be handled effectively. This talk will explore various routing and planning problems, including classical and rich vehicle routing problems, order picking, and team orienteering. We will discuss real-world, large-scale problems in transportation and warehouse planning, highlighting the challenges that arise in practice.

Various algorithmic approaches have been developed for solving large-scale routing and related planning problems, ranging from classical heuristics to advanced metaheuristics. We will provide an overview of key methodologies, including large neighborhood search, evolutionary mechanisms, and hybrid techniques, emphasizing their efficiency and adaptability in different logistical settings. Beyond traditional optimization, handling uncertainties is essential for robust logistics planning. We will examine how uncertainties such as the amount of transported load and temporal aspects impact routing solutions and discuss strategies for adapting and extending existing efficient approaches.

In post-disaster response, relief items are delivered to disaster victims to meet immediate needs and alleviate suffering. At the initial stages of the disaster, it is important to allocate limited supplies equitably and ensure that they reach the people in need as soon as possible. We will present a study on planning vehicle routes from a distribution center to shelters while allocating limited relief supplies. To balance efficiency and equity, a bi-objective problem is defined. The objectives are to minimize a Gini-index-based measure of inequity in unsatisfied demand for fair distribution and to minimize total travel time for timely delivery. By deriving mathematical properties of the optimal solution, we introduce valid inequalities and design an algorithm for optimal delivery allocations given feasible vehicle routes. A branch-and-price (B&P) algorithm is developed to solve the problem efficiently. Computational tests on realistic datasets show that the B&P algorithm significantly outperforms commercial MIP solvers. Our bi-objective approach reduces aid distribution inequity by 34% without compromising efficiency.

Column Generation (CG) is a technique to solve Linear Programs with a very large number of variables. Instead of explicitly evaluating reduced costs, variables are dynamically generated by solving auxiliary optimization problems known as pricing subproblems. CG is one of the major optimization techniques, being also effective in integer programming, in algorithms like Branch-and-Price and Branch-Cut-and-Price. It has been successfully applied to many types of vehicle routing, cutting and packing, airline planning, timetabling, crew scheduling, graph coloring, clustering, lot sizing, and machine scheduling, among other problems. The talk provides an overview of the CG. The central question explored is: under what circumstances are CG-based algorithms likely to outperform other existing methods? The discussion draws on material from the recent book “Optimizing with Column Generation: advanced Branch-Cut-and-Price Algorithms (Part I)” available at https://optimizingwithcolumngeneration.github.io.