Program Overview
September 10, 2026 Thursday:
09:00 – 09:30: Registration
09:30 – 09:45: Welcome and Opening
09:45 – 10:45: Keynote 1 (Andreas Löhne)
10:45 – 11:15: Coffee Break
11:15 – 12:45: Scientific Program – Session 1
12:45 – 14:00: Lunch
14:00 – 15:30: Scientific Program – Session 2
15:30 – 16:00: Coffee Break and Group Photo
16:00 – 17:00: Scientific Program – Session 3
17:00: Transfer / Outing / Dinner
September 11, 2026 Friday:
09:30 – 10:30: Keynote 2 (Gülşah Karakaya)
10:30 – 11:00: Coffee Break
11:00 – 12:30: Scientific Program – Session 4
12:30: Lunch
The detailed program can be downloaded as a PDF file here.
List of speakers and talk details are available.
Scientific Program
September 10, 2026 Thursday:
09:00 – 09:30: Registration
09:30 – 09:45: Welcome
09:45 – 10:45: Keynote 1 (Chair: Serpil Sayın)
09:45 – 10:45: Andreas Löhne (Friedrich Schiller University Jena)
Set Linear Programming: Theory, Algorithms, and Applications
(Abstract)
Set linear programming, also known as polyhedral convex set optimization, extends classical multiobjective linear optimization by allowing the objective map to be set-valued. In this framework, the graph of the objective map is assumed to be a convex polyhedron. Set linear programs naturally arise in multistage decision-making processes of multiobjective linear programs, where only part of the decision variables is fixed in the first stage, while the remaining variables are determined in subsequent stages. In addition to the original objectives, the decision maker often values flexibility with respect to future decisions. We show that modeling such preferences for flexibility is a key advantage of set linear programming and can be naturally incorporated within its framework. This talk provides an introduction to the fundamental concepts of set linear programming and highlights its relationship to multiobjective linear programming. Particular emphasis is placed on applications motivated by a preference for flexibility. The talk further presents an overview of solution concepts and algorithmic methods developed for set linear programs. A central theoretical result shows that every set linear program can be represented by an equivalent multiobjective linear program. Although this reformulation is typically prohibitively large from a computational perspective, it reveals deep connections between the two problem classes and allows theoretical results to be transferred between them.
10:45 – 11:15: Coffee Break
11:15 – 12:45: Scientific Program – Session 1 (Chair: Stefan Ruzika)
11:15 – 11:45: Firdevs Ulus (Bilkent University)
On the convergence of outer approximation algorithms for convergence vector optimization problems
(Abstract)
Recently in [Ç. Ararat, F. U., M. Umer, SIAM Journal on Optimization, 34:3 (2024), pp 2169-3166] we studied the convergence rate of an outer approximation algorithm for solving bounded convex vector optimization problems. The algorithm solves a norm minimizing scalarization in each iteration and terminates after finitely many iterations ensuring that the Hausdorff distance between the outer approximation and the upper image is less than a given approximation error. The norm is arbitrary but fixed through the iterations. The strong convergence rate of $\mathcal{O}(k^{{2}/{(1-q)}})$ was proved for the Euclidean norm. We now prove the same convergence rate for any arbitrary fixed norm. Moreover, if the norm used in the scalarization changes through the iterations, then it is possible to obtain the same convergence rate under some assumptions on the set of norms used. We show that Pascoletti-Serafini scalarization can be seen as a norm-minimizing scalarization, where the norm depends on the parameters of the scalarization.
11:45 – 12:15: Mariagrazia Cairo (Sapienza University of Rome)
Parallelization Strategies for Bi-Objective Combinatorial Optimization Problems
(Abstract)
Computing the complete nondominated set of a bi-objective combinatorial optimization problem is often computationally demanding. Two-phase methods represent a well-established class of exact approaches for tackling this challenge. They first identify the supported nondominated points and then explore the remaining regions (triangles) of the objective space to recover the non-supported ones [1]. While there are dependencies between regions that prevent a straightforward parallelization of the second phase, their structure still offers opportunities to leverage parallel computing. In this work, we propose a general parallel framework for two-phase methods based on the concurrent exploration of triangles in the objective space. Different strategies for assigning triangles to be explored to parallel tasks are investigated to improve workload distribution and reduce computational imbalance among parallel processors. The framework is evaluated on the Bi-Objective Minimum Spanning Tree Problem through both distributed and shared memory implementations. Computational experiments over several graph families of different sizes and cost correlation coefficients show that the proposed shared memory implementation emerges as the most effective approach across the tested instances, yielding lower execution times and good scalability on larger and more challenging problems [2, 3]. The scalability analysis further highlights both the potential of the framework and the main factors affecting parallel efficiency.
References:
[1] Ulungu, E., Teghem, J., 1994. The two phases method: An efficient procedure to solve bi-objective combinatorial optimization problems. Foundations of computing and decision sciences 20, 2, 149–165.
[2] Steiner, S., Radzik, T., 2008. Computing all efficient solutions of the biobjective minimum spanning tree problem. Computers & Operations Research 35, 1, 198–211.
[3] Amorosi, L., Puerto, J., 2022. Two-phase strategies for the bi-objective minimum spanning tree problem. International Transactions in Operational Research 29, 6, 3435–3463.
12:15 – 12:45: Gabriele Eichfelder (Ilmenau University of Technology)
Supportedness in Quadratic Multiobjective Optimization
(Abstract)
For single-objective quadratic optimization problems, much is known about the existence of optimal solutions and the topological properties of the image set. For instance, in the unconstrained case, the image set is always a closed convex interval, and either a minimizer exists or the problem is unbounded from below. This situation changes substantially as soon as two or more objective functions are considered: the image set may be neither closed nor open, and it need not be convex. Moreover, a multiobjective problem may be unbounded from below and still have efficient solutions. The aim of this talk is to contribute to a better understanding of quadratic multiobjective optimization problems. One tool in this analysis is provided by conic lifting-and-relaxation techniques, through which certain nonconvex quadratic problems can be reformulated exactly as convex problems over suitable cones. We use such ideas to identify a class of quadratic multiobjective problems for which all nondominated points are supported, that is, they can be obtained by weighted-sum scalarization. We also discuss further properties and characterizations of the image set of quadratic multiobjective problems, consequences for the existence of optimal and supported solutions, and, if time allows, the role of potentially redundant objective functions.
12:45 – 14:00: Lunch
14:00 – 15:30: Scientific Program – Session 2 (Chair: Lavinia Amorosi)
14:00 – 14:30: Diclehan Tezcaner Öztürk (Hacettepe University)
Multi-Objective Route Planning of a Coordinated Truck-Drone Pair for Post-Disaster Reconnaissance
(Abstract)
Obtaining timely and accurate information about the affected areas after a natural disaster is critical for effective emergency response. Traditional reconnaissance methods rely on ground-based inspections, which are rather slow and limited due to damaged infrastructure and restricted accessibility. Recent studies have started considering the use of air vehicles in addition to the ground vehicles to overcome such challenges. We consider the route planning problem of a truck and drone pair tasked with the reconnaissance of a region after a disaster. The truck may move to nodes that are accessible in the region and the drone, that travels with the truck, visits inaccessible nodes from the air. The information that can be collected at each location is time-dependent. The objectives are to maximize the total information collected through drone visits while minimizing the overall mission completion time. We formulate the problem as a mixed integer programming (MIP) model, and develop a column generation approach to reduce its computational burden. Our approach is a two-stage decomposition consisting of a master problem and pricing-based subproblems. Our computational tests show that our approach can generate the real nondominated frontier in considerably shorter durations than the MIP model and large instances can be solved to optimality.
14:30 – 15:00: Tamara Ertl (Johannes Kepler University)
Multi-objective Vehicle Routing Considering Physical and Mental Driver Fatigue
(Abstract)
We integrate drivers’ physical and mental fatigue into the capacitated vehicle routing problem and study the resulting trade-offs between operational cost and driver well-being. Mental fatigue accumulates from uninterrupted driving and is relieved by breaks, whereas physical fatigue builds up nonlinearly from lifting tasks at customers, with diminishing marginal increase as fatigue approaches its upper bound. We develop a compact two-index mixed-integer linear programming formulation that tracks fatigue along routes and allows breaks to be scheduled at customer nodes. We consider three settings: a bi-objective problem with cost against the maximum mental-fatigue peak (with physical fatigue constrained), a bi-objective variant in which physical fatigue becomes the second objective, and a tri-objective problem minimizing cost, mental, and physical fatigue simultaneously. To tackle the bi-objective problem variants, we rely on the ε-constraint method. For the tri-objective problem, we implement the criterion-space search algorithm of Tamby and Vanderpooten (2021). Each scalarized subproblem is currently solved by branch-and-cut using rounded capacity cuts and newly derived physical capacity cuts. As an alternative exact approach, we explore the fatigue measures as resources in a branch-cut-and-price approach using the VRPSolver framework of Pessoa et al. (2020). Our preliminary experiments on CVRPLib instances indicate that reducing fatigue peaks requires additional cost, that scheduling breaks lowers and balances the mental-fatigue peak across drivers, and that the physical-fatigue threshold becomes binding only on larger instances. (with: Gerhard Hiermann, Sophie N. Parragh, and Francesco Pilati)
15:00 – 15:30: Michael Stiglmayr (University of Wuppertal)
Ordinal Shortest Paths to Compute Safe and Short Routes to School
(Abstract)
Ordinal costs can be used to model the quality of objects w.r.t. ordered categories whenever numerical (cost) values are not available. For example, the safety of a street segment can be ranked in ordered categories like safe, medium safe, and unsafe. We particularly focus on applications in pedestrian routing and consider the case of optimizing paths for children on their way to elementary school. The route planing problem then corresponds to an ordinal shortest path problem. It is known that ordinal optimization problems can be represented by a vector optimization problem, which can be (linearly) transformed into a standard multi-objective optimization problem. Based on a safety assessment in up to 16 categories we use the k-shortest path algorithm and an adapted filtering to compute practically relevant, non-dominated paths to school. We discuss the interrelation between ordinal optimization and associated multiobjective problem formulations at the particular case of shortest path problems. Modeling aspects with regard to consistency of preference choices and wrt. the avoidance of safe, yet overly long paths, are discussed. Associated solution concepts are suggested, and the problem is illustrated at a small real world instance. (with: Rabea Freese, Kathrin Klamroth, Julia Sudhoff Santos, Chiara Weuste)
15:30 – 16:00: Coffee Break and Group Photo
16:00 – 17:00: Scientific Program – Session 3 (Chair: Gökhan Kof)
16:00 – 16:30: Xavier Gandibleux (Nantes University)
The Paving as a Representation of Nondominated Points for the Bi-objective Uncapacitated Facility Location Problem
(Abstract)
A three-phase algorithm for computing the set of nondominated points for the bi-objective uncapacitated facility location problem has been proposed (https://optimization-online.org/?p=34431). The first two phases construct a paving—that is, a collection of boxes covering all nondominated points—using a breadth-first branch-and-bound algorithm. The boxes are then contracted to obtain the most compact paving possible. The resulting paving provides a representation that can already be presented to a decision-maker. The algorithm is implemented in Julia and is available in open-source (https://github.com/xgandibleux/biUFLP). It has been tested on instances from the literature (https://github.com/vOptSolver/vOptLib/tree/master/UFLP). The presentation will cover the algorithm and results, with a particular focus on the first two phases.
16:30 – 17:00: Renee Lamsfuß (University of Wuppertal)
Consistent Path Selection for Bi-objective Location Problems on Graphs
(Abstract)
We consider bi-objective location problems on graphs and want to compute not only the efficient locations, but also the corresponding routes to the existing facilities. We assume that a central decision maker wants to locate a new facility, e.g., a pizza delivery place, which uses bicycles for delivery. The two objective functions could be the traveling time and the number of left turns, as they are very risky. Route choices then depend on the preferences of the decision maker(s), and in general, there may not exist a unique optimal path between a customer and a new facility location. In this talk, we introduce the concept of consistent paths, where the choice of a path from the facility to a demand node implies a certain preference information w.r.t. both objective functions. We compute efficient facility locations, which are considered together with associated Pareto optimal facility-to-customer-routes. All paths of a solution are consistent if the different preference relations for the paths do not contradict each other. We present an algorithm that computes all efficient solutions with consistent path choices and illustrate the algorithm with an ordinal location problem. (with: Kathrin Klamroth, Michael Stiglmayr, Julia Sudhoff Santos)
17:00: Transfer / Outing / Dinner
September 11, 2026 Friday:
09:30 – 10:30: Keynote 2 (Chair: Özlem Karsu)
09:30 – 10:30: Gülşah Karakaya (Middle East Technical University)
Distance-based Value Functions in Multi-objective Optimization
(Abstract)
In multi-objective optimization, it is important to converge towards preferred solutions of a decision maker (DM). This task is inherently challenging, as multiple conflicting objectives typically generate a diverse set of efficient solutions, each representing different tradeoffs. To identify preferred solutions, interactive approaches where the DM is engaged throughout the search process have been developed. These approaches typically assume an underlying value function that represents the preferences of the DM. They seek for preference information from the DM progressively, update the search space accordingly, and continue until converging to preferred solutions. The approaches exploit the properties of the assumed underlying value function in order to increase the efficiency of the search process. Linear, quasiconcave/quasiconvex, and general monotone functions are among the commonly assumed value functions. There is a tradeoff between the generality of the assumed value function and the speed of convergence. Typically, these approaches guarantee converging to preferred solutions provided that the assumed form of the value function represents the DM’s preferences well. In this talk, we address the family of distance-based value functions that can represent a wide variety of preference structures. We develop interactive approaches that assume such functions and we demonstrate that they work well.
10:30 – 11:00: Coffee Break
11:00 – 12:30: Scientific Program – Session 4 (Chair: Tamey Cansın Ekşi)
11:00 – 11:30: Cem Tekin (Bilkent University)
Black-box Vector Optimization
(Abstract)
Black-box optimization aims to optimize systems whose objective values can be evaluated only at high cost, without access to analytical models or gradients. While Bayesian optimization has become a powerful framework for sample-efficient optimization of single-objective black-box functions, extending it to vector-valued objectives introduces new challenges arising from conflicting objectives. In this talk, I will present our recent work on black-box vector optimization, beginning with algorithms for discrete design spaces and then extending them to continuous and correlated design spaces through Gaussian process surrogate models that efficiently exploit spatial correlations among designs. These methods provide principled approaches for identifying Pareto optimal solutions in a sample-efficient manner while accommodating general preference structures encoded by polyhedral ordering cones. Finally, I will introduce VOPy, our open-source Python library for black-box vector optimization, which provides a unified, extensible platform for developing, benchmarking, and applying algorithms.
11:30 – 12:00: Stephanie Riedmüller (Zuse Institute Berlin)
Minimizing Objective Coefficients in Multi-Objective Binary Programming
(Abstract)
For a given multi-objective binary program, we define the problem of finding the smallest integer objective coefficients without changing the efficient set. Finding such coefficients is essentially a dominance-preserving transformation that contracts the objective space. We present a novel exact method and compare it against simple scaling heuristics. The exact method formulates an integer program with exponentially many constraints and solves it via a cutting-plane algorithm. In a computational study, we discuss the applicability and potential benefits of the objective space contraction.
12:00 – 12:30: Joris Wenzel (RPTU in Kaiserslautern)
Topology of Pareto Sets in Thermodynamic Modeling
(Abstract)
Mathematical modeling of material properties is essential in many areas of engineering. Important model classes include equations of state and force-field models for molecular simulations. Their parameters are fitted to experimental data, but improving the description of one property often deteriorates the description of others, naturally motivating the use of multi-objective optimization. These optimization problems typically exhibit a characteristic structure: Pareto sets contain a pronounced Pareto knee region and long, often nearly linear branches connecting this region to the optima of the corresponding single-objective problems [1-2]. The Pareto knee is of particular interest because it represents solutions with favorable trade-offs between conflicting objectives and has been observed in many applications [3-4]. In this talk, we analyze the existence, location, and sharpness of Pareto knees, the observed topologies of the Pareto sets, and how both of them can be explained and identified in practice: A quadratic approximation of the objective functions is used to effectively parametrize equations of state and force-field models and may also be suitable for other thermodynamic optimization problems [5]. Within this framework, we study how variations in model parameters influence the emergence and prominence of Pareto knees and compare different mathematical definitions of the concept. We further examine the relationship between the topology of the Pareto set and the prominence of the knee region across different dimensions of the decision and objective spaces. This connects structural properties of multi-objective optimization problems with the identification and interpretation of Pareto knees. (with: F. Jirasek, S. Ruzika, H. Hasse)
References:
[1] K. Stöbener, P. Klein, S. Reiser, M. Horsch, K.-H. Küfer, H. Hasse, Fluid Phase Equilibria 373 (2014) 100–108.
[2] A. Kulkarni, M. Bortz, K.-H. Küfer, M. Kohns, H. Hasse, Journal of Chemical Theory and Computation 16.8 (2020) 5127–5138.
[3] C. Ramirez-Atencia, S. Mostaghim, D. Camacho, sKPNSGA-II: Knee point based MOEA with self-adaptive angle for Mission Planning Problems (2022).
[4] J. Tang, H. Wang, Y. Jin, Swarm and Evolutionary Computation 92 (2025) 101813.
[5] A. Kulkarni, M. Kohns, M. Bortz, K.-H. Küfer, H. Hasse, Optimization and Engineering 24 (2022) 1611–1632.
12:30: Lunch