Archive

Adaptive robust optimization problems have received significant attention in recent years, but remain notoriously difficult to solve when recourse decisions are discrete in nature. In this talk, we propose new reformulation techniques for adaptive robust binary optimization (ARBO) problems with objective uncertainty. Our main contribution revolves around a collection of exact and approximate network flow reformulations for the ARBO problem, which we develop by building upon ideas from the decision diagram literature. Our proposed models can generate feasible solutions, primal bounds and dual bounds, while their size and approximation quality can be precisely controlled through user-specified parameters. Furthermore, these models are easy to implement and can be solved directly with standard off-the-shelf solvers. We show that our models can generate high-quality solutions and dual bounds in significantly less time than popular benchmark methods.

Ian Zhu is an Assistant Professor in the Department of Analytics and Operations at NUS Business School. His research interests broadly lie in the design of algorithms for solving various optimization problems. His research has been published in journals such as Operations Research, MSOM, and INFORMS Journal on Computing, and has been recognized in several paper awards, including the Pierskalla Best Paper Award, Innovative Applications in Analytics Award, and the MSOM Student Paper Competition.

Linear-Quadratic-Gaussian (LQG) control is a fundamental control paradigm that is studied in various fields such as engineering, computer science, economics, and neuroscience. It involves controlling a system with linear dynamics and imperfect observations, subject to additive noise, with the goal of minimizing a quadratic cost function for the state and control variables. In this work, we consider a generalization of the discrete-time, finite-horizon LQG problem, where the noise distributions are unknown and belong to Wasserstein ambiguity sets centered at nominal (Gaussian) distributions. The objective is to minimize a worst-case cost across all distributions in the ambiguity set, including non-Gaussian distributions. Despite the added complexity, we prove that a control policy that is linear in the observations is optimal for this problem, as in the classic LQG problem. We propose a numerical solution method that efficiently characterizes this optimal control policy. Our method uses the Frank-Wolfe algorithm to identify the least-favorable distributions within the Wasserstein ambiguity sets and computes the controller's optimal policy using Kalman filter estimation under these distributions.

Bahar is an Assistant Professor of Operations Management at the University of Chicago Booth School of Business. In 2024, she completed her PhD in the Risk Analytics and Optimization Lab at École polytechnique fédérale de Lausanne (EPFL) in Switzerland. In 2018, she obtained her Bachelor of Science degree in Electrical and Electronics Engineering from the Middle East Technical University.

This talk presents two approaches to addressing multistage optimization problems under uncertainty: multistage robust optimization (RO) and multistage distributionally robust optimization (DRO).

In the first part of the talk, we focus on multistage robust optimization problems of the minimax type, where the uncertainty set is modeled as a Cartesian product of compact stagewise sets [1]. To manage the inherent intractability of these problems, we propose an approximation strategy based on random sampling of the uncertainty space. This leads to two main questions: (1) how to quantify the error introduced by this sampling approach as a function of the sample size, and (2) whether the optimal value of the sampled problem converges to the true robust optimal value as the number of samples increases. We provide probabilistic guarantees on constraint violations and develop a sequence of lower bounds for the original robust problem. Numerical results on a multistage inventory management problem show that the method is effective for problems with a small number of stages, while highlighting computational limitations for larger-scale instances.

The second part of the talk focuses on multistage mixed-integer distributionally robust optimization. In this setting, uncertainty is represented through conditional ambiguity sets defined on scenario trees, constrained to remain close to nominal conditional distributions according to some measure of similarity/distance. 

We propose a new bounding technique based on scenario grouping using the ambiguity sets associated

with various commonly used φ-divergences and the Wasserstein distance [2]. Our approach does not rely on specific structural assumptions such as linearity, convexity, or stagewise independence. The resulting bounds are applicable to a wide variety of DRO problems - multistage or two-stage, with or without integer variables, convex or nonconvex. Numerical results on a multistage production planning problem illustrate the effectiveness of the approach across different ambiguity sets, robustness levels, and partition strategies.

This talk aims to provide theoretical insights and practical methods for tackling uncertainty in multistage optimization, highlighting key aspects of both robust and distributionally robust formulations.

References:

[1] Maggioni, F., Dabbene, & Pflug, G. (2025) Multistage robust convex optimization problems: A sampling based approach, Annals of Operations Research, https://doi.org/10.1007/s10479-025-06545-4.

[2] Bayraksan, G., Maggioni, F., Faccini, D. & Yang, A. (2024) Bounds for Multistage Mixed-Integer Distributionally Robust Optimization, Siam Journal on Optimization, 34(1), 682-717.

Francesca Maggioni is Professor of Operations Research at the Department of Management, Information and Production Engineering (DIGIP) of the University of Bergamo (Italy). Her research interests concern both methodological and applicative aspects for optimization under uncertainty. From a methodological point of view, she has developed different types of bounds and approximations for stochastic, robust and distributionally robust multistage optimization problems. She applies these methods to solve problems in logistics, transportation, energy and machine learning. On these topics she has published more than 80 scientific articles featured in peer-reviewed operations research journals. In 2021 her research has been supported as principal investigator by a grant from the Italian Ministry of Education for the project “ULTRA OPTYMAL Urban Logistics and sustainable TRAnsportation: OPtimization under uncertainTY and MAchine Learning”. 

She is currently coordinating the research group CQIIA-Matnet for teaching of mathematics and its applications of the University of Bergamo and is Deputy Director of the PhD program in Applied Economics & Management of the Universities of Bergamo and Pavia.

She currently chairs the EURO Working Group on Stochastic Optimization and the AIRO Thematic Section of Stochastic Programming. She has been secretary and treasurer of the Stochastic Programming Society. She is Associate Editor of the journals Transportation Science, Computational Management Science (CMS), EURO Journal on Computational Optimization (EJCO), TOP, International Transactions in Operational Research (ITOR), Networks, RAIRO Operations Research and guest editor of several special issues in Operations Research and Applied Mathematics journals.

The utilization of data is becoming paramount in addressing modern, increasingly complex, optimization problems. In this talk, I will present my vision of key aspects relating to data-driven robust optimization: (i) should data be used to first construct a description of uncertainty, or should we rather use data directly in the optimization domain? (ii) is it possible to endow data-driven designs with robustness certificates that hold without any a-priori knowledge? In the second part of the talk, I will introduce “Scenario Optimization”, a collection of data-driven optimization techniques that can provide algorithmic and theoretical answers to these questions.  

Marco C. Campi is Professor of Control and Data-Driven Methods at the University of Brescia, Italy. He has held visiting and teaching appointments with Australian, USA, Japanese, Indian, Turkish and various European universities, besides the NASA Langley Research Center in Hampton, Virginia, USA. He has served as Chair of multiple IFAC Technical Committees and has been in diverse capacities on the Editorial Board of top-tier control and optimization journals. In 2008, he received the IEEE CSS George S. Axelby award for the article “The Scenario Approach to Robust Control Design”. He has delivered plenary and semi-plenary addresses at major conferences including OPTIMIZATION, SYSID, and CDC, besides presenting various talks as IEEE distinguished lecturers for the Control Systems Society. Marco C. Campi is a Fellow of IEEE and a Fellow of IFAC. His interests include: data-driven decision-making, stochastic optimization, inductive methods, and the fundamentals of probability theory. 

Counterfactual explanations (CEs) are advocated as being ideally suited to providing algorithmic recourse for subjects affected by the predictions of machine learning models. While CEs can be beneficial to affected individuals, recent work has exposed severe issues related to the robustness of state-of-the-art methods for obtaining CEs. Since a lack of robustness may compromise the validity of CEs, techniques to mitigate this risk are in order. In this talk we will begin by introducing the problem of (lack of) robustness, discuss its implications and present some recent solutions we developed to compute CEs with robustness guarantees.

Francesco is a Research Fellow affiliated with the Centre for Explainable Artificial Intelligence at Imperial College London. His research focuses on safe and explainable AI, with special emphasis on counterfactual explanations and their robustness. Since 2022, he leads the project “ConTrust: Robust Contrastive Explanations for Deep Neural Networks”, a four-year effort devoted to the formal study of robustness issues arising in XAI. More details about Francesco and his research can be found at https://fraleo.github.io/.

Sequential decision making often requires dynamic policies, which are computationally not tractable in general. Decision rules provide approximate solutions by restricting decisions to simple functions of uncertainties. In this paper, we consider a nonparametric lifting framework where the uncertainty space is lifted to higher dimensions to obtain nonlinear decision rules. We propose two nonparametric liftings, which derive the nonlinear functions by leveraging the uncertainty set structure and problem coefficients. Both methods integrate the benefits from lifting and nonparametric approaches, and hence provide scalable decision rules with performance bounds. More specifically, the set-driven lifting is constructed by finding polyhedrons within uncertainty sets, inducing piecewise-linear decision rules with performance bounds. The dynamics-driven lifting, on the other hand, is constructed by extracting geometric information and accounting for problem coefficients. This is achieved by using linear decision rules of the original problem, also enabling one to quantify lower bounds of objective improvements over linear decision rules. Numerical comparisons with competing methods demonstrate superior computational scalability and comparable performance in objectives. These observations are magnified in multistage problems with extended time horizons, suggesting practical applicability of the proposed nonparametric liftings in large-scale dynamic robust optimization.

Eojin Han is an assistant professor of IT, Analytics and Operations in Mendoza College of Business at University of Notre Dame. His research focuses on developing robust optimization methods to address operational challenges arising from uncertainties where accurate predictions are not available, with applications to healthcare, supply chains, and service systems. His work has been published in leading academic journals such as Operations Research and Management Science. Before joining Notre Dame, he served as an assistant professor at Southern Methodist University. Dr. Han holds a Ph.D. in Industrial Engineering and Management Sciences from Northwestern University and a B.S. in Mathematics and Electrical Engineering from Seoul National University. 

In bilevel and robust optimization we are concerned with combinatorial min-max problems, for example from the areas of min-max regret robust optimization, network interdiction, most vital vertex problems, blocker problems, and two-stage adjustable robust optimization.

Even though these areas are well-researched for over two decades and one would naturally expect many (if not most) of the problems occurring in these areas to be complete for the classes \Sigma_2 or \Sigma_3  from the polynomial hierarchy, almost no hardness results in this regime are currently known.

However, such complexity insights are important, since they imply that no polynomial-sized integer program for these min-max problems exist (unless the polynomial hierarchy collapses), and hence conventional IP-based approaches fail.

We address this lack of knowledge by introducing over 70 new \Sigma_2-complete and \Sigma_3-complete problems. I particular, we explore a meta-theorem, which reveals an underlying pattern behind the computational hardness of combinatorial min-max problems. Our meta-theorem applies to many classic combinatorial problems (like TSP, clique, vertex cover, facility location, knapsack, etc...) in conjunction with min-max regret, network interdiction, or two-stage adjustable robust optimization.

Lasse Wulf is a postdoctoral researcher at the Technical University of Denmark (DTU). He obtained a double master's degree in mathematics and computer science from Karlsruhe Institute of Technology (KIT) and a PhD from the Technical University of Graz. He is recipient of the best student paper award at IWOCA 2021, and the best dissertation award from the Austrian Society of Operations Research (ÖGOR). His research in robust optimization is centered around the computational complexity of combinatorial min-max optimization, for example in min-max regret, network interdiction, bilevel, and adjustable/recoverable multi-stage robust optimization. In particular, he is interested in cases where such problems are even harder than NP-hard.

In this talk, we first briefly review a general class of data-driven decision-making and highlight the role of convexity in addressing three research questions that arise in this context. We then focus on a particular setting of such problems known as inverse optimization (IO), where the goal is to replicate the behavior of a decision-maker with an unknown objective function. We discuss recent developments in IO concerning convex training losses and optimization algorithms. The main message of this talk is that IO is a rich supervised learning model that can capture a complex (e.g., discontinuous) behavior while the training phase is still a convex program. We motivate the discussion with applications from control (learning the MPC control law), transportation (the 2021 Amazon Routing Problem Challenge), and robotics (comparison with the state-of-the-art in the MuJoCo environments).    

Peyman Mohajerin Esfahani is an Associate Professor at the Delft Center for Systems and Control. He joined TU Delft in October 2016, and prior to that, he held several research appointments at EPFL, ETH Zurich, and MIT between 2014 and 2016. He received the BSc and MSc degrees from Sharif University of Technology, Iran, and the PhD degree from ETH Zurich. He currently serves as an associate editor of Mathematical Programming, Operations Research, Transactions on Automatic Control, and Open Journal of Mathematical Optimization. He was one of the three finalists for the Young Researcher Prize in Continuous Optimization awarded by the Mathematical Optimization Society in 2016, and a recipient of the 2016 George S. Axelby Outstanding Paper Award from the IEEE Control Systems Society. He received the ERC Starting Grant and the INFORMS Frederick W. Lanchester Prize in 2020. He is the recipient of the 2022 European Control Award. 

In uncertain optimization problems with decision-dependent information discovery, the decision-maker can influence when information is revealed, unlike the classic setting where uncertain parameters are revealed according to a prescribed filtration. This work examines two-stage robust optimization problems with decision-dependent information discovery, focusing on uncertainty in the objective function. We introduce the first exact algorithm for this class of problems and enhance the existing K-adaptability approximation by strengthening its formulation. Our approaches are tested on robust orienteering and shortest-path problems. We study the interplay between information and adaptability, demonstrating that the exact solution method often outperforms the K-adaptability approximation. Moreover, our experiments highlight the benefits of the strengthened K-adaptability formulation over the classical one, even in settings with decision-independent information discovery.

Rosario Paradiso is an Assistant Professor in the Department of Operations Analytics at Vrije Universiteit Amsterdam in the Netherlands. He earned his PhD from the Department of Mathematics and Computer Science at the University of Calabria in Italy in 2020. Rosario's research primarily focuses on exact algorithms for combinatorial optimization problems, both deterministic and under uncertainty, with applications in supply chain management.

Problem uncertainty typically limits how well decisions can be tailored to the problem at hand but often can not be avoided. The availability of large quantities of data in modern applications however presents an exciting opportunity to nevertheless make better informed decisions. Distributional robust optimization (DRO) has recently gained prominence as a paradigm for making data-driven decisions which are protected against adverse overfitting effects. We justify the effectiveness of this paradigm by pointing out that certain DRO formulations indeed are statistically efficient. Furthermore, DRO formulations can also be tailored to protect decisions against overfitting even when working with messy corrupted data. Finally, we discuss a hitherto hidden relation between DRO and classical robust statistics.

Bart obtained his PhD in control theory at ETH Zurich in 2016 under the supervision of Prof. Manfred Morari. He was a SNSF postdoctoral researcher at the Operations Research Center at MIT from 2018 to 2020 and Assistant Professor at the MIT Sloan School of Management from 2018 to 2024. As of 2024 Bart is a tenured senior researcher in the Stochastics Group at CWI Amsterdam. Bart's research is located on the interface between optimization and machine learning.  He particularly enjoy developing novel mathematical methodologies and algorithms with which we can turn data into better decisions. Although most of his research is methodological, he does enjoy applications related to problems in renewable energy.

We design primal and dual bounding methods for general multistage adaptive robust optimization (MSARO) problems. From the primal perspective, this is achieved by applying decision rules that restrict the functional forms of only a certain subset of decision variables resulting in an approximation of MSARO as a two-stage adjustable robust optimization (2ARO) problem. We leverage the 2ARO literature in the solution of this approximation. From the dual perspective, decision rules are applied to the Lagrangian multipliers of a Lagrangian dual of MSARO, resulting in a two-stage stochastic optimization problem. As the quality of the resulting dual bound depends on the distribution chosen when developing the dual formulation, we define a distribution optimization problem with the aim of optimizing the obtained bound, and we develop solution methods tailored to the nature of the recourse variables. Computational experiments illustrate the potential value of our bounding techniques compared to the existing methods. 

(This is a joint work with Maryam Daryalal and Ayse N. Arslan.)

Merve Bodur is an Associate Professor in the School of Mathematics at the University of Edinburgh. She obtained her Ph.D. from the University of Wisconsin-Madison, her B.S. in Industrial Engineering and her B.A. in Mathematics from Bogazici University, Turkey. Her main research area is optimization under uncertainty, primarily for discrete optimization problems, with applications in a variety of areas such as scheduling, transportation, healthcare, telecommunications, and power systems. She serves on the editorial boards of Operations Research Letters, Omega, and INFOR. She is currently the Vice Chair/Chair-Elect of the INFORMS Computing Society, serves on the Committee on Stochastic Programming, and is a former Vice Chair of the INFORMS Optimization Society.

In the shortest path problem, we are given a directed graph G with nonnegative arc costs, and we seek an s-t path in G of the smallest total cost. This fundamental combinatorial problem can be solved efficiently in general digraphs. In some applications, a path computed in the first stage can be modified to some extent after the costs in G have been changed. Therefore, knowing the new cost scenario, we compute another (second-stage) path in some prescribed neighborhood of the first-stage path, so that the total cost of both paths is minimal. Typically, the second-stage arc costs are uncertain while the first-stage path is computed, and we seek a solution that minimizes the total cost under the worst second-stage cost realization.  We use interval uncertainty sets to model the uncertain second-stage arc costs, which is very common in robust optimization. The recoverable robust shortest path problem under the interval uncertainty has been extensively studied since the work of Busing (2011, 2012). In this talk, some new results in this area are presented. We will characterize the complexity of the problem, depending on the class of the input digraph (general, acyclic, arc series-parallel), type of the neighborhood (arc inclusion, arc exclusion, arc symmetric difference) and structure of the interval uncertainty set (discrete budgeted, continuous budgeted). We will also state some open problems in this area.

 Adam Kasperski is a Professor at the Wroclaw University of Science and Technology.  His scientific interests focus on robust discrete optimization problems. His work aims to better understand the computational complexity of different approaches in this area. He is currently a head of research project entitled "Decision making under risk and uncertainty - concepts, methods, and applications", at the Wroclaw University of Science and Technology.

We present prescriptive analytics paradigms for data-driven decision-making for complex systems and environments. Conventional data-driven decision-making methods assume access to plentiful, well-structured data to generate a prediction model, and subsequently, an optimization model is used to create a decision. However, real data can be incomplete (sparse, missing), scarce, partially observable, and time-varying.  We present new learning and optimization frameworks to tackle emerging challenges caused by the complexity and limitations of data and difficulties inherent in data-system integration. Our approaches are based on amalgams of machine learning, distributionally robust/stochastic optimization, and combinatorial optimization.

 Hoda Bidkhori is an Assistant Professor at George Mason University, where her research focuses on the theory and applications of data analytics and data-driven optimization, with applications in logistics, supply chain management, and healthcare. Previously, she served as an Assistant Professor at the University of Pittsburgh. Prior to that, she was a postdoctoral researcher in Operations Research at MIT, where she also earned her Ph.D. in Applied Mathematics. 

We study decision problems under uncertainty, where the decision-maker has access to K data sources that carry biased information about the underlying risk factors. The biases are measured by the mismatch between the risk factor distribution and the K data-generating distributions with respect to an optimal transport (OT) distance. In this situation the decision-maker can exploit the information contained in the biased samples by solving a distributionally robust optimization (DRO) problem, where the ambiguity set is defined as the intersection of K OT neighborhoods, each of which is centered at the empirical distribution on the samples generated by a biased data source. We show that if the decision-maker has a prior belief about the biases, then the out-of-sample performance of the DRO solution can improve with K—irrespective of the magnitude of the biases. We also show that, under standard convexity assumptions, the proposed DRO problem is computationally tractable if either K or the dimension of the risk factors is kept constant. 

Daniel Kuhn is Professor of Operations Research at the College of Management of Technology at EPFL, where he holds the Chair of Risk Analytics and Optimization (RAO). His current research interests are focused on data-driven optimization, the development of efficient computational methods for the solution of stochastic and robust optimization problems and the design of approximation schemes that ensure their computational tractability. His work is primarily application-driven, the main application areas being engineered systems, machine learning, business analytics and finance. Before joining EPFL, Daniel Kuhn was a faculty member in the Department of Computing at Imperial College London (2007-2013) and a postdoctoral research associate in the Department of Management Science and Engineering at Stanford University (2005-2006). He is the editor-in-chief of Mathematical Programming. 

In this work we propose a novel polyhedral uncertainty set for robust optimization: the smooth uncertainty set. This set imposes bounds on the difference of pairs of uncertain parameters. Smooth uncertainty sets arise naturally in applications where there is a strong connection between the generation of these parameters due to vicinity in space/time, or due to physical constraints. Thus,  for such applications, smooth uncertainty sets can be constructed based on expert estimation or existing data. 

Under probabilistic assumptions on the uncertain parameters' correlations, we derive a construction of the uncertainty set that provides probabilistic guarantees for the resulting solution of the robust problem. We show that the size of this set decreases as the correlation between the parameters increases.

In terms of tractability, for specially structured problems, we prove the robust counterpart can be reformulated in a compact way, with no additional variables and few additional constraints. For a more general problem structure, we construct a column generation algorithm tailored for the smooth uncertainty set for solving the robust problem.

Joint work with Michael Poss (CNRS, LIRMM) and Noam Goldberg (Bar-Ilan University)

Shimrit Shtern is a senior lecturer in the Faculty of Data and Decision Sciences at the Technion - Israel Institute of Technology. She received a Ph.D. in Optimization and an MSc in Operations Research, both from the Technion, and was a post-doctoral fellow at the Operation Research Center at MIT. Her research focuses on theory and applications of optimization under uncertainty as well as development and convergence analysis of first order methods for structured optimization problems. She is a recipient of research grants from both the Israeli Science Foundation (ISF) and The Deutsche Forschungsgemeinschaft (DFG, German Research Foundation).

Multi-stage mixed-integer robust optimization, in which decisions are taken sequentially as new information becomes available about the uncertain problem parameters, is a very versatile yet computationally challenging paradigm for decision making under uncertainty. In this talk, I will propose solution methods for  two-stage and multi-stage versions of such problems. For the first, we will construct piecewise constant decision rules by adaptively partitioning the uncertainty set. The partitions of this set are iteratively updated by separating so-called critical scenarios. We discuss how to identify such critical scenarios based on the branch-and-bound procedure for solving the current corresponding static problem. For multi-stage robust mixed-integer programs, we will consider a finite adaptability scheme, which allows us to decompose the multi-stage problem into a large number of much simpler two-stage problems. We discuss how these two-stage problems can be solved both exactly and approximately, and we report numerical results for route planning and location-transportation problems.

Ward Romeijnders is a full professor within the Department of Operations at the University of Groningen. His research is focused on developing exact and approximate solution methods for integer optimization problems under uncertainty. This class of problems can be used to support decision making under uncertainty for a wide range of applications in, e.g., energy, healthcare, logistics, and finance. Ward publishes his research in journals such as Operations Research, Mathematical Programming, SIAM Journal on Optimization, European Journal of Operational Research, and INFORMS Journal on Computing, and he is Associate Editor of Mathematical Methods of Operations Research. He has received several research grants from the Netherlands Organisation for Scientific Research, the latest one for a project called "Discrete Decision Making under Uncertainty".

We present two machine learning methods to learn uncertainty sets in decision-making problems affected by uncertain data.  In the first part, we introduce mean robust optimization (MRO), a framework that constructs data-driven uncertainty sets based on clustered data rather than on observed data points directly. By varying the number of clusters, MRO balances computational complexity and conservatism, and effectively bridges robust and Wasserstein distributionally robust optimization.  In the second part, we present LRO, a method to automatically learn the shape and size of the uncertainty sets in robust optimization.  LRO optimizes the performance across a family of parametric problems, while guaranteeing a target probability of constraint satisfaction.  It relies on a stochastic augmented Lagrangian method that differentiates the solutions of robust optimization problems with respect to the parameters of the uncertainty set. Our approach is very flexible, and it can learn a wide variety of uncertainty sets commonly used in practice.  We illustrate the benefits of our two machine learning methods on several numerical examples, obtaining significant reductions in computational complexity and conservatism, while preserving out-of-sample constraint satisfaction guarantees.

Bartolomeo Stellato is an Assistant Professor in the Department of Operations Research and Financial Engineering at Princeton University. Previously, he was a Postdoctoral Associate at the MIT Sloan School of Management and Operations Research Center. He received a DPhil (PhD) in Engineering Science from the University of Oxford, a MSc in Robotics, Systems and Control from ETH Zurich, and a BSc in Automation Engineering from Politecnico di Milano. He is the developer of OSQP, a widely used solver in mathematical optimization. Bartolomeo Stellato's awards include the NSF CAREER Award, the Princeton SEAS Howard B. Wentz Jr. Faculty Award, the Franco Strazzabosco Young Investigator Award from ISSNAF, the Princeton SEAS Innovation Award in Data Science, the Best Paper Award in Mathematical Programming Computation, and the First Place Prize Paper Award in IEEE Transactions on Power Electronics. His research focuses on data-driven computational tools for mathematical optimization, machine learning, and optimal control.

We study network design problems for nonlinear and nonconvex flow models under demand uncertainties. To this end, we apply the concept of adjustable robust optimization to compute a network design that admits a feasible transport for all, possibly infinitely many, demand scenarios within a given uncertainty set. For solving the corresponding adjustable robust mixed-integer nonlinear optimization problem, we show that a given network design is robust feasible, i.e., it admits a feasible transport for all demand uncertainties, if and only if a finite number of worst-case demand scenarios can be routed through the network. We compute these worst-case scenarios by solving polynomially many nonlinear optimization problems. Embedding this result for robust feasibility in an adversarial approach leads to an exact algorithm that computes an optimal robust network design in a finite number of iterations. Since all of the results are valid for general potential-based flows, the approach can be applied to different utility networks such as gas, hydrogen, or water networks. We finally demonstrate the applicability of the method by computing robust gas networks that are protected from future demand fluctuations.

Johannes Thürauf is a postdoctoral researcher at Trier University since 11/2021. He studied Mathematics at the Friedrich-Alexander Universität (FAU) Erlangen-Nürnberg and received his M.Sc. degree in 2017. Afterward, he worked as a PhD student at the FAU Erlangen-Nürnberg and got his PhD at the end of 2021. His main research interests include robust and bilevel optimization (under uncertainty), for example with applications to networks. In particular, he is interested in the interplay of robust and bilevel optimization.

We address a continuous-time continuous-space stochastic optimal control problem where the controller lacks exact knowledge of the underlying diffusion process. Instead, the controller relies on a finite set of historical disturbance trajectories. To tackle this challenge, we propose a novel approach called Distributionally Robust Path Integral control (DRPI). This method utilizes distributionally robust optimization (DRO) to make the resulting policy robust against the unknown diffusion process. Notably, the DRPI scheme bears resemblance to risk-sensitive control, allowing us to leverage the path integral control (PIC) framework as an efficient solution method. We establish theoretical performance guarantees for the DRPI scheme, which facilitates the determination of an appropriate risk parameter for the risk-sensitive control. Through extensive validation, we demonstrate the effectiveness of our approach and its superiority over risk-neutral and risk-averse PIC policies in scenarios where the true diffusion process is unknown.

Grani A. Hanasusanto is an Associate Professor of Industrial & Enterprise Systems Engineering at the University of Illinois Urbana-Champaign (UIUC). Before joining UIUC, he was an Assistant Professor at The University of Texas at Austin and a Postdoctoral Scholar at the College of Management of Technology at École Polytechnique Fédérale de Lausanne. He holds a Ph.D. in Operations Research from Imperial College London and an MSc in Financial Engineering from the National University of Singapore. His research focuses on developing tractable solution schemes for decision-making problems under uncertainty, with applications spanning operations management, energy systems, finance, data analytics, and machine learning. He is a recipient of the NSF CAREER award and the INFORMS Diversity, Equity, and Inclusion (DEI) Ambassadors Program award. He currently serves as a committee member of the INFORMS Diversity Community and an Associate Editor for Operations Research.

In this talk, we focus on static robust optimization problems where the uncertainty set can be controlled through the actions of the decision maker known as decision-dependent uncertainty. Particularly, we consider an uncertainty reduction paradigm where the decision-maker can reduce upper bounds of uncertain parameters as a result of some proactive actions. This paradigm was recently proposed in the literature and the resulting problems were shown to be NP-Hard.  In this talk, we pay particular attention to the special case of robust combinatorial optimization problems, and show that they are also NP-Hard under the uncertainty reduction paradigm. Despite this discouraging result, we show that under some additional assumptions polynomial-time algorithms can be devised to solve robust combinatorial optimization problems with uncertainty reduction. We additionally provide insights into possible MILP reformulations in the general case and illustrate the practical relevance of our results on the shortest path instances from the literature.

Ayse N. Arslan is a junior researcher at Inria Center of Bordeaux. Her research interests lie at the intersection of mixed-integer programming and optimization under uncertainty with a particular focus on stochastic and robust optimization and their applications. Some of her recent work includes decomposition algorithms for solving two-stage robust optimization problems and primal and dual decision rules for approximating multi-stage robust optimization problems.

In this talk, we will present our new inexact column-and-constraint generation (i-C&CG) method for solving challenging two-stage robust and distributionally robust optimization problems. In our i-C&CG method, the master problems only need to be solved to a prescribed relative optimality gap or time limit, which may be the only tractable option when solving large-scale and/or challenging problems in practice. We equip i-C&CG with a backtracking routine that controls the trade-off between bound improvement and inexactness. Importantly, this routine allows us to derive theoretical finite convergence guarantees for our i-C&CG method, demonstrating that it converges to the exact optimal solution under some parameter settings. Numerical experiments on a scheduling problem and a facility location problem demonstrate the computational advantages of our i-C&CG method over a state-of-the-art C&CG method.

Dr. Karmel S. Shehadeh is an Assistant Professor in the Department of Industrial and Systems Engineering at Lehigh University. Before joining Lehigh, she was a presidential and dean postdoctoral fellow at Heinz College of Information Systems and Public Policy at Carnegie Mellon University. She holds a doctoral degree in Industrial and Operations Engineering from the University of Michigan, a master's degree in Systems Science and Industrial Engineering from Binghamton University, and a bachelor's in Biomedical Engineering from Jordan University of Science and Technology. Her research interests revolve around stochastic and robust optimization with applications to healthcare operations and analytics, facility location, transportation systems, and fair decision-making. Professional recognition of her work includes an INFORMS Minority Issues Forum Paper Award (winner), Junior Faculty Interest Group Paper Prize (finalist), Service Science Best Cluster Paper Award (finalist), and Lehigh's Alfred Noble Robinson Faculty Award.

In combinatorial optimization problems under budget uncertainty the solution space is deterministic, but the objective is uncertain. By the seminal work of Bertsimas & Sim (2003, 2004), such problems can be solved in various ways: by separating constraints bounding the objective, by a compact reformulation, or by solving n+1 deterministic problems of the same type. Although the complexity of the problem does not increase in theory, in many cases the performance deteriorates significantly. 

In this talk, we show two ways to enhance the computation times of the compact reformulation: (i) recycling valid inequalities known for the deterministic problem yielding surprisingly strong valid inequalities for the robust problem and (ii) a tailored branch and bound algorithm performing a tree search on one of the dual variables. This is joint work with Timo Gersing and Christina Büsing.

Arie M.C.A. Koster is since 2009 a Professor of Mathematics at RWTH Aachen University. Before, he received his PhD from Maastricht University and worked at Zuse Institute Berlin and University of Warwick. His research interests include integer linear programming, polyhedral combinatorics, robust optimization, algorithmic graph theory, and network optimization. He has published over 100 research papers and is on the editorial board of Annals of Operations Research, INFORMS Journal on Computing, and Networks.

We study inventory control with volume flexibility: A firm can replenish using period-dependent base capacity at regular sourcing costs and access additional supply at a premium. The optimal replenishment policy is characterized by two period-dependent base-stock levels but determining their values is not trivial, especially for nonstationary and correlated demand. We propose the Lookahead Peak-Shaving policy that anticipates and peak shaves orders from future peak-demand periods to the current period, thereby matching capacity and demand. Peak shaving anticipates future order peaks and partially shifts them forward. This contrasts with conventional smoothing, which recovers the inventory deficit resulting from demand peaks by increasing later orders. Our contribution is threefold. First, we use a novel iterative approach to prove the robust optimality of the Lookahead Peak-Shaving policy. Second, we provide explicit expressions of the period-dependent base-stock levels and analyze the amount of peak shaving. Finally, we demonstrate how our policy outperforms other heuristics in stochastic systems. Most cost savings occur when demand is nonstationary and negatively correlated, and base capacities fluctuate around the mean demand. Our insights apply to several practical settings, including production systems with overtime, sourcing from multiple capacitated suppliers, or transportation planning with a spot market. Applying our model to data from a manufacturer reduces inventory and sourcing costs by 6.7%, compared to the manufacturer's policy without peak shaving.

This is a joint work with Joren Gijsbrechts, Robert N. Boute, Jan A. Van Mieghem.

Christina Imdahl is Assistant Professor of Machine Learning in Operations Management at the OPAC group of Eindhoven University of Technology (TU/e). Christina's research is focuses on decision-making in operations management. She develops models that are motivated by operational practice and explores how we can leverage data to make better decisions. In her research, she combines the use of stochastic and robust optimization with empirical analyses to gain insights into performance.

For more than a century, election officials across the United States have inspected voting machines before elections using a procedure called Logic and Accuracy Testing (LAT). This procedure consists of election officials casting a test deck of ballots into each voting machine and confirming the machine produces the expected vote total for each candidate.

In this talk, I will bring a scientific perspective to LAT by introducing the first formal approach to designing test decks with rigorous security guarantees. Specifically, we propose using robust optimization to find test decks that are guaranteed to detect any voting machine misconfiguration that would cause votes to be swapped across candidates. Out of all the test decks with this security guarantee, the robust optimization problem yields the test deck with the minimum number of ballots, thereby minimizing implementation costs for election officials. To facilitate deployment at scale, we developed a practical exact algorithm for solving our robust optimization problems based on mixed-integer optimization and the cutting plane method.

In partnership with the Michigan Bureau of Elections, we retrospectively applied our robust optimization approach to all 6928 ballot styles from Michigan's November 2022 general election; this retrospective study reveals that the test decks with rigorous security guarantees obtained by our approach require, on average, only 1.2% more ballots than current practice. Our robust optimization approach has since been piloted in real-world elections by the Michigan Bureau of Elections as a low-cost way to improve election security and increase public trust in democratic institutions.

This is joint work with Alex Halderman (Michigan) and Braden Crimmins (Stanford and Michigan), and our preprint is available at https://arxiv.org/pdf/2308.02306.pdf.

Brad Sturt an assistant professor of business analytics at the University of Illinois Chicago. His current research focuses on the application of robust optimization to revenue management and public policy. He is also broadly interested in using robust optimization to design tractable approaches for dynamic optimization problems. His research has received several recognitions, including second place in the INFORMS Junior Faculty Interest Group (JFIG) Paper Competition and second place in the INFORMS George Nicholson Student Paper Competition.