Abstract: The utilization of data is becoming paramount in addressing modern optimization problems characterized by increasing complexity. In this talk, I’ll present my vision of key aspects relating to data-driven optimization: (i) should data be used to reconstruct the unknown distribution of uncertainty, or should we rather directly optimize to achieve the final design? (ii) is it possible to endow data-driven designs with certifications of quality that hold beyond any assumed characteristics of the underlying distribution? 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 responses to the questions posed earlier.

Bio: 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: stochastic optimization, inductive methods, data-driven decision-making and the fundamentals of probability theory.

RENATA PEDRINI

Abstract: The long-term generation scheduling (LTGS) problem presented important developments over the years. In this work, we analyze different risk aversion strategies for handling the impact of climate change in the LTGS problem in hydro-dominated power systems using stochastic dual dynamic programming (SDDP). Despite the advances in modeling and solution strategy, there are important issues related to the distributions of inflows (and other random variables), especially with climate change. First, even though time series or other statistical methods are used to model these random parameters, the true probability distribution is never fully known. Furthermore, historical values used to devise the statistical models may no longer be valid as processes shift. We focus on solutions to mitigate these issues protecting the system from deviations from the inflow scenario distribution, achieved by a Distributionally Robust Optimization (DRO) framework. Instead of a single distribution, DRO considers all distributions that are sufficiently close to this nominal distribution and optimizes a worst-case expected (or risk-averse) objective, where the expectations concern all the considered distributions. DRO is more realistic because it explicitly considers existing data while recognizing that forecasts may contain errors. The DRO policies are tested against risk-neutral (expected value minimization) and CVaR risk-averse approaches using data from the Brazilian power system. Policies are compared as well as the practical application of different algorithms. The results indicate that incorporating DRO improves the out-of-sample performance of policies. 

(Authors: R. Pedrini, G. Bayraksan, E. C. Finardi, F. Beltrán)

Bio: Renata is a fifth-year Ph.D. candidate in Electrical Engineering at Brazil's Federal University of Santa Catarina (UFSC), where she works under the guidance of Prof. Erlon Finardi. Her research is centered on addressing challenges in power system operation and planning, with a particular emphasis on leveraging innovative algorithms and methodologies. Recently, Renata completed a 10-month collaboration with Prof. Guzin Bayraksan at The Ohio State University, focusing on developing strategies for power system operation in response to the impacts of climate change. Her collaborative efforts extend beyond academia, as she actively collaborates with energy companies to gain insights into the real-world challenges faced by the power system. She currently integrates CEPEL, the Center for Electrical Energy Research in Brazil working with inflow scenario generation.

ARAS SELVI

Abstract: The recent advent of data-driven and end-to-end decision-making across different areas of operations management has led to an ever closer integration of prediction models from machine learning and optimization models from operations research. A key challenge in this context is the presence of estimation errors in the prediction models, which tend to be amplified by the subsequent optimization model – a phenomenon that is often referred to as the Optimizer's Curse or the Error-Maximization Effect of Optimization. A contemporary approach to combat such estimation errors is offered by distributionally robust problem formulations that consider all data-generating distributions close to the empirical distribution derived from historical samples, where 'closeness' is determined by the Wasserstein distance. While those techniques show significant promise in problems where all input features are continuous, they scale exponentially when binary and/or categorical features are present. This work demonstrates that such mixed-feature problems can indeed be solved in polynomial time. We present a practically efficient algorithm to solve mixed-feature problems, and we compare our method against alternative techniques both theoretically and empirically on standard benchmark instances. 

(Authors: Reza Belbasi, Aras Selvi, Wolfram Wiesemann)

Bio: Aras (https://www.arasselvi.com/) is a doctoral candidate at Imperial College Business School, supervised by Professor Wolfram Wiesemann. He is affiliated with the Computational Optimization Group and the Data Science Institute of Imperial College London and has recently completed PhD research internships at The Alan Turing Institute and J.P. Morgan AI research. His research interests are the theory of data-driven decision making under uncertainty and its applications in machine learning, privacy, and fairness. In his recent works, he has been working on designing optimal privacy mechanisms, developing efficient algorithms for robust machine learning, as well as approximating hard decision making problems via robust optimization.

Abstract: We study a stochastic combinatorial optimization problem arising from applications in industrial engineering and manufacturing: the uncapacitated single-item lot-sizing problem under uncertain demand and costs. This problem consists in determining the timing and quantity of production of a product on a machine so as to satisfy the customer demand while minimizing the costs for the manufacturer.

The problem is modeled as a multi-stage stochastic mixed-integer linear program in which the evolution of the uncertain parameters is represented by a scenario tree. To solve it, we propose a new extension of the stochastic dual dynamic integer programming algorithm (SDDiP). This extension aims at being more computationally efficient in the management of the expected cost-to-go functions involved in the model, in particular by reducing their number and by exploiting the current knowledge on the polyhedral structure of the stochastic uncapacitated lot-sizing problem. The algorithm is based on a partial decomposition of the problem into a set of stochastic subproblems, each one involving a subset of nodes forming a sub-tree of the initial scenario tree. We then introduce a cutting-plane generation procedure that iteratively strengthens the linear relaxation of these sub-problems and enables the generation of additional strengthened Benders' cut, which improves the convergence of the method.  Our numerical results carried out on randomly generated large-size instances show that the proposed algorithm significantly outperforms the SDDiP algorithm.

Bio: Céline Gicquel is an Associate Professor in Operations Research at the Université Paris Saclay in France. She obtained a PhD in Industrial Engineering  from the Ecole Centrale Paris (France) in 2008 and an accreditation to supervise research from the Université Paris Saclay (France) in 2021. Her research interests involve modeling and solving combinatorial optimization problems coming from applications in manufacturing, energy and transportation. She worked among others on the development of stochastic integer programming approaches for lot-sizing and facility location problems.

Abstract: Multistage stochastic programming is a powerful tool allowing decision-makers to revise their decisions at each stage based on the realized uncertainty. However, in various practical settings, organizations are not able to be fully flexible, as decisions cannot be revised too frequently due to their high organizational impact. Consequently, decision commitment becomes crucial to ensure that initially made decisions remain unchanged for a certain period. To this end, in this talk, we introduce partially adaptive multistage stochastic programming approaches that strike an optimal balance between decision flexibility and commitment by determining the best stages to revise decisions depending on the allowed level of flexibility. We introduce a novel mathematical formulation and theoretical properties eliminating certain constraint sets. Furthermore, we develop a decomposition method that effectively handles mixed-integer adaptive multistage programs by adapting the integer L-shaped method and Benders decomposition. We further provide analytical results over stylized problems depending on the revision time. Computational experiments on stochastic lot-sizing and generation expansion planning problems show substantial advantages attained through optimal selections of revision times when flexibility is limited, while demonstrating computational efficiency of the proposed properties and solution methodology. Optimizing revision times in a less flexible case can outperform arbitrary selection in a more flexible case, achieving performance levels comparable to fully flexible settings. 

Bio: Beste Basciftci is an Assistant Professor at the Department of Business Analytics at the Tippie College of Business at the University of Iowa. She obtained her PhD degree in Operations Research from the H. Milton Stewart School of Industrial and Systems Engineering at Georgia Institute of Technology, with a minor in Statistics. She received her bachelor's degrees in industrial engineering and computer engineering from Bogazici University with High Honors. She also holds a master's degree in industrial engineering from Bogazici University. She is broadly interested in data-driven decision-making problems under uncertainty. Methodologically, her research focuses on developing mixed-integer, stochastic programming and distributionally robust optimization approaches to address modelling and computational challenges in operations research and management related problems. The application areas of these problems mainly involve i) energy systems and sustainability, ii) supply chains and facility location problems, and iii) smart city operations (such as sharing systems and emerging transportation systems). She is also interested in game theoretic frameworks where the decisions of multiple entities can be considered as part of these optimization approaches. She is honored to receive prestigious awards with her research, including the IISE Transactions Best Paper Award in Focus Issue of Operations Engineering & Analytics, INFORMS ENRE (Energy, Natural Resources and the Environment Section) Best Student Paper Award, University of Iowa Early Career Scholar Award, and Georgia Tech ISyE Alice and John Jarvis Research Award. 

Abstract: In this talk we discuss our experience with an ongoing  project which aims at building stochastic optimization models for the transition to clean energy in Chile. Many challenges arise in this type of project, from collecting data to properly modeling uncertainty, in addition of solving the problems efficiently and validating the models. We will discuss some of these challenges and how we have approached them.

Bio: Tito Homem-de-Mello is a Professor in the School of Business at Universidad Adolfo Ibañez, Santiago, Chile.  He obtained his Ph.D. in Industrial and Systems Engineering from Georgia Institute of Technology, an M.Sc. in Applied Mathematics from the same institution, and a B.Sc. degree in Computer Science from University of São Paulo, Brazil. His research focuses on optimization of systems under uncertainty. In particular, he studies theory and algorithms for stochastic optimization as well as applications of such methods in several areas such as energy, finance, and transportation. Dr. Homem-de-Mello was co-chair of the Program Committee of the XIV International Conference on Stochastic Programming, held in Brazil in 2016, and served for three years as a member of COSP, a steering committee for the stochastic optimization community. He is currently an Associate Editor for Computational Optimization and Applications.

Abstract: In this talk, we will discuss multi-stage stochastic programming (MSP) models and solution approaches for humanitarian relief logistics planning in natural disasters such as hurricanes. We consider logistics decision-making such as the relief item prepositioning, shelter planning, and contingency modality selection over multiple periods prior to the landfall of an impending hurricane. Using stochastic forecast information about the hurricane's attributes over time, we propose MSP models which provide optimal adaptive logistics decision policies. Our preliminary numerical results and sensitivity analyses demonstrate the value of MSP for hurricane relief logistics planning, as well as the trade-offs between policy flexibility, solution quality, and computational effort. 

Bio: Dr. Yongjia Song is an associate professor in the Department of Industrial Engineering at Clemson University. Dr. Song received his Ph.D. degree in industrial and systems engineering from University of Wisconsin-Madison in 2013. Dr. Song’s research interests include optimization under uncertainty, integer programming, and applications of optimization in transportation and logistics, networks, and health care. Dr. Song is a recipient of the NSF CAREER award in 2021, and his research has been supported by several federal funding agencies, such as the NSF, DOE, ONR, among others. 

Abstract: Bilevel is a powerful tool for modeling hierarchical decision making processes between interdependent decision makers. Compared to single-level optimization models under data uncertainty, bilevel programs further involve decision uncertainty, where the upper-level decision maker may not be certain about the lower-level decision maker’s reaction when they have multiple alternative optimal solutions. In this talk, we present recent results of bilevel programming models with uncertainty under ambiguously known distribution considering (i) a pessimistic or (ii) an optimistic upper-level decision maker whose decision is binary. For (i), we study a pessimistic bilevel program with moment information of uncertainty, construct upper bounds based on 0-1 semidefinite programming (SDP) approximations and develop cutting plane algorithms to solve 0-1 SDPs. We conduct numerical studies to demonstrate the effectiveness and efficiency on a facility location problem. For (ii), we consider an optimistic bilevel program using Wasserstein metrics. We show that favorable data-driven properties including out-of-sample guarantee, asymptotic consistency, and tractability are still preserved in the bilevel setting. A case study of a demand response problem on a 33-bus distribution network is conducted, and our results draw attention to the need of considering equity constraints with heterogeneous households.

Bio: Yiling Zhang is an Assistant Professor in the Department of Industrial and Systems Engineering at the University of Minnesota. She received her Ph.D. in Industrial and Operations Engineering from the University of Michigan. Her research interests include stochastic programming, integer programming, and nonlinear programming. Her research has applications to various complex service systems, including transportation systems, power systems, and healthcare operations.

Abstract: In ride-hailing systems, pickup time refers to the time that elapses from the moment a vehicle is dispatched to pick up a rider until the rider is picked up. A fundamental phenomenon in ride-hailing systems is that there is a trade-off between pickup time and the time that a vehicle waits for a dispatch. In short, if vehicles spend little time idle waiting for a dispatch, then few vehicles are available when a rider makes a request, and thus the mean distance between a rider and the closest available vehicle is long, which means that pickup time is long.

This phenomenon is of great importance in ride-hailing. In spite of this, the existing literature on price optimization for ride-hailing, and on repositioning optimization for ride-hailing, ignores pickup time. We consider a Markov Decision Process formulation of the problem of making pricing decisions when a potential customer requests a ride, as well as vehicle repositioning decisions when a rider is dropped off. The Markov Decision Process models the trade-off between pickup time and the time that a vehicle waits for a dispatch. The Markov Decision Process is intractable, and therefore we consider the associated fluid optimization problem. Typical fluid optimization problems replace random variables with their means. In this case, such an approach results in an intractable fluid optimization problem. Instead, we consider a finite mixture approximation of the pickup time distribution, that results in a tractable conic optimization problem.

The usual approach when using fluid optimization problems is to use its optimal solution to control the original stochastic process. We show that in this case the fluid system may have two stable fixed points, one that is optimal and one that is suboptimal (as well as an unstable fixed point). The consequence is that the usual approach may result in poor performance, with the stochastic system spending too much time close to the suboptimal fixed point. We show how the fluid optimal solution can be used to construct policies that avoid this trap, and that perform much better in simulations than the policies proposed in previous papers.

Bio: Anton Kleywegt is a faculty member in the Stewart School of Industrial and Systems Engineering at the Georgia Institute of Technology. His research interests include stochastic optimization, with applications in transportation and revenue management.

Abstract: The presentation will begin by describing smooth bootstrap applied to confidence intervals for the optimality gap associated with a solution to a stochastic program. This extends work where stochastic programming is used based on a sample from an unknown probability distribution. The second part of the talk will describe a software architecture for computing solutions with upper and lower bounds in an asynchronous parallel environment.

Bio: Prof. David Woodruff's research concerns computational aspects of optimal decision making. He is particularly interested in problems with a mix of discrete and continuous choices with multiple time stages when there is significant uncertainty. His research includes solution algorithms, problem representation and modeling language support. He has worked on applications in operations, logistics, science, and has been involved recently in a number of applications in electrical energy planning and scheduling. From 2013 to 2019 he was editor-in-chief of the INFORMS Journal on Computing, which is a publication of the Institute for Operations Research and Management Science.