Archive

Here you find the list of talks in the MODUS Seminar from past semesters. The list goes back to 2023, for earlier talks please see the MODUS elearning course.

Winter Semester 2025/2026

29.10.2025, 12:15h
Marie Schmidt
Professorin für Optimierung unter Ressourcenbeschränkungen, Universität Würzburg
Robust multi-objective optimization

Abstract: When modeling real-world challenges as optimization problems, we often encounter uncertainty in problem parameters; as well as the coexistence of multiple goals which are difficult to trade-off against each other. Robust optimization is an approach that addresses the challenge of parameter uncertainty, aiming to find a solution that is feasible under all scenarios (parameter realizations ) and best in the worst-case. Multi-objective optimization addresses the challenge of multiple objective functions by introducing the concept of efficiency (or Pareto optimality), which says that a solution x is worth to look at, if there is no other solution which is at least as good as x with respect to all considered goals, and better in at least one of them.
Several ideas have been proposed to combine these two concepts into the concept of robust efficiency, and we briefly illustrate these before we turn to solution approaches for robust multi-objective problems. There are (at least) two ways to design solution approaches for computing robust efficient solutions: we can try to generalize algorithms for (single-objective) robust optimization to the multi-objective case, or generalize algorithms for (deterministic) multi-objective optimization. Though most of these approaches are applicable for a wider class of problems, for the presentation we focus on biobjective combinatorial problems with bounded uncertainty in both objective functions using the concept of point-wise robust efficiency.

12.11.2025, 12:15h
Yonatan Gutman
Professor and Head of the Department of Dynamical Systems, Polish Academy of Sciences, Warsaw, Poland
Prediction and Reconstruction of Dynamical Systems - A Mathematical Point of View

Abstract: A fundamental problem in experimental sciences is the problem of reconstructing and predicting the future of an observed dynamical system. In recent years together with Krzysztof Barański and Adam Śpiewak, we have studied this problem from a mathematical point of view within the framework of delayed observations. I will discuss various aspects of our research, including the probabilistic Takens theorem, the Schroer-Sauer-Ott-Yorke conjectures, the prediction delay dimension threshold and, if time permits, algorithmic aspects.

03.12.2025, 12:30h
Vedran Perić
Professor and Head of the Chair of Intelligent Energy Management, Universität Bayreuth
Data-Driven Modeling in Power Systems

Abstract: The talk will discuss modeling requirements in modern power systems, especially in light of growing renewable generation, reduced system inertia, and increasing operational uncertainty. A main focus will be on using system-identification and prediction-error methods to monitor electromechanical oscillations.The special focus will be on assessing uncertainty of the obtained mathematical models and factors that influence this uncertainty. Furthermore, the talk will address strategies for experiment design in system identification using prediction-error methods, with an emphasis on obtaining models with minimal uncertainty while respecting the constraints present in real power grids.Finally, the role of machine-learning methods in data- driven modeling will be examined, including where they are effective, where they have limitations, and how they can complement classical identification techniques in power-system applications.

10.12.2025, 12:30h
Jiří Outrata
Institute of Information Theory and Automation, The Czech Academy of Sciences, Prague, Czech Republic
On the numerical solution of a class of EPECs via the Gauss-Seidel method for computation of Nash equilibria
(jointly with the Oberseminar Numerics, Optimization, and Dynamical Systems)

Abstract: The talk is focused on a class of multi-leader multi-follower games for which we use the acronym EPEC. One of the prominent applications of this modeling framework is the deregulated electricity market, where one has to do with only one Follower. Recently, in [2], a variant of the implicit programming approach (ImP) to a class of MPECs has been suggested, based on the usage of a bundle method along with a special semismooth derivative. It seems that this variant might be used in the framework of the Gauss-Seidel method from [1] which, under suitable assumptions, would then converge to a stationary point of the considered EPEC. In this way we intend to enrich the current, rather modest arsenal of numerical techniques, capable to solve some types of EPECs. The approach will be illustrated by academic examples constructed via a modification of Nash-Cournot equilibria, where one firm takes over the role of the follower, selecting its strategy in dependence on the strategies, applied by its concurrents. The talk is based on joint work with H.Gfrerer, T. Roubal, and J. Valdman.
Literature:
[1] Ch. Kanzow, A.Schwartz: Spieltheorie, Birkhäuser 2018
[2] H. Gfrerer, M. Kočvara, J.V. Outrata: On the role of semismoothness in the implicit programming approach to selected nonsmooth optimization problems, arXiv:2412.05953

14.1.2026, 12:30h
Christian Fiedler
Chair of Applied Numerical Analysis (Prof. M. Fornasier), Technische Universität München
Statistical learning theory for kernel methods with distributional inputs and two-stage sampling

Abstract: In a range of machine learning applications, from medical diagnostics to causality, distributions appear as inputs, and these distributions might not even be directly accessible, but only through samples thereof. Kernel-based methods are well-suited for this problem setting by first embedding the distributions into a Hilbert space and then using a standard kernel on this Hilbert space. This strategy has received particular attention for the case of regression with distributions as inputs, called distributional regression, and by now efficient algorithms as well as substantial theory are available. However, for other variants of learning with distributional inputs there has been much less work, most notably for the case of distributional classification. Motivated by this latter fact, we present recent work on advancing statistical learning theory for kernel methods with distributional inputs, covering important settings like binary classification with support vector machines.

21.1.2026, 12:30h
Tomáš Roubal
Department of Decision-Making Theory, Institute of Information Theory and Automation, The Czech Academy of Sciences, Prague, Czech Republic
On Ranges of Set-Valued Mappings and Solvability of the Load-Flow Problem

Abstract: Motivated by range-type questions arising in power-flow (load-flow) analysis, we study quantitative lower estimates of the range of set-valued mappings. We derive sufficient conditions ensuring that the image of a feasible set covers a prescribed neighborhood in the output space, phrased via openness and covering properties for nonsmooth set-valued mappings. The approach uses generalized derivative concepts (including Clarke-type and Páles–Zeidan-type Jacobians) and yields Lyusternik–Graves and Pourciau-style results, together with stability under Lipschitz perturbations.
We apply these statements to constrained load-flow models. Viewing the load-flow equations as a mapping from admissible network states to power injections under operational limits. The resulting range estimates provide certified solvability regions for injections and robustness margins guaranteeing existence of feasible load-flow solutions under perturbations of operating conditions.

28.1.2026, 12:30h
Veronika Riedl
Lehrstuhl für Wissenschaftliches Rechnen, Universität Bayreuth
Coupled 1D/2D Modeling of Coastal Ocean Circulation with the Discontinuous Galerkin Method

Abstract: Shallow water equations (SWE) are the main mathematical model used in coastal and regional ocean modeling. Realistic coastal ocean domains often include features with predominantly one-dimensional structure such as shipping channels or rivers. Accurately meshing and simulating in 2D requires very fine mesh resolution negatively affecting the CFL condition and resulting in significant computational overhead.
In this talk, a coupled 1D/2D modeling approach is presented in which 1D elements are embedded within the 2D domain as edges of the 2D elements while possessing a distinct bathymetry. This increases model flexibility and improves computational efficiency by allowing for larger time steps. Using the Discontinuous Galerkin (DG) method, we validated our model with idealized and realistic test cases while addressing the main challenge of conservative and stable coupling between 1D and 2D flows.

Summer Semester 2025

21.05.2025, 12:15h
Andreas Löhne
Professur Mathematische Optimierung, Universität Jena
Two Applications of Multi-Objective Linear Programming

Abstract: We explore two less conventional applications of multi-objective linear programming (MOLP).
In the first application, we employ a MOLP solver to compute geometric operations on polyhedra, including the Minkowski sum, intersection, and the convex hull of their union.
The second application addresses multi-objective optimization problems involving two sequential decision makers. The first decision maker controls a subset of variables and acts initially, while the second controls the remaining variables and acts subsequently. This setting gives rise to a multi-stage decision problem, which can be interpreted as an optimization problem with a set-valued objective function—commonly referred to as a set optimization problem. We examine this framework in detail, discussing solution methodologies, the underlying decision-making structure, and practical applications. Furthermore, we highlight the equivalence of such problems to classical multi-objective linear programs.

28.05.2025, 12:15h
Giovanni Fantuzzi
Chair for Dynamics, Control, Machine Learning and Numerics, FAU Erlangen-Nürnberg
Understanding transformers: hardmax attention, clustering, and perfect sequence classification

Abstract: Transformers are an extremely successful machine learning model, famously known for powering platforms such as ChatGPT. What distinguishes them from classical deep neural networks is the presence of "attention" layers between standard "feed-forward" layers. In this talk, I will discuss how simple geometrical rules can explain the role of the attention layers and, consequently, the outstanding practical performance of transformers. Specifically, by focussing on a simplified class of transformers with "hardmax" attention, I will first show that attention layers induce clustering of the transformer's input data. I will then use this clustering effect to construct transformers that can perfectly classify a given set of input sequences with arbitrary but finite length, modelling, for example, books to be classified by a library. Crucially, the complexity of this construction is independent of the sequence length. This is in stark contrast to classical deep neural networks, explaining (at least in part) the superior performance of transformers for sequence classification tasks.

Winter Semester 2024/2025

27.11.2024, 12:15h, also via zoom
Aditi Jain
Applied Mathematics, University of Bayreuth
Investigation of Complex Intracellular Dynamics through the lens of Cooperative Dynamical Systems

Abstract: Movement is an important part of life. For example, in a central and fundamental process known as gene expression, there is a movement of biological particles called RNA polymerases on the DNA strand to produce messenger RNA (mRNA). Understanding these complex transport phenomena has been a significant area of research in mathematics, biology, and physics. Over the years, the Ribosome Flow Model (RFM), obtained via a mean-field approximation of a stochastic model called the Totally Asymmetric Simple Exclusion Process (TASEP), has provided a rigorous mathematical framework for the analysis. It is a deterministic, continuous-time model for analyzing the flow of interacting particles, and its dynamics are described by ordinary differential equations (ODEs). The results of the RFM analysis can be used to model and engineer gene expression. In this talk, I rely on the framework of RFM to model and analyze the dynamical flow of particles along an ordered chain of sites encapsulating various biologically observed features. The presentation will focus on formulating a system of non-linear ordinary differential equations, where the densities of each site on a lattice serve as the state variables and understand their asymptotic behavior. Exploring cooperative irreducible systems of ODEs with a first integral exhibiting positive gradient, results are leveraged on the global phase portrait of such systems in the proposed models. These frameworks yield deeper insights into how parameters influence system dynamics, enhancing our comprehension of the underlying processes.

04.12.2024, 12:15h
Christoph Helbig
Ecological Resource Technology, University of Bayreuth
Modelling, Simulating, and Evaluating Global Material Cycles with Methods of Industrial Ecology

Abstract: The increasing demand for resources and the associated environmental impacts of extraction, processing, and consumption pose significant sustainability challenges from greenhouse gas emissions to land and water use. To address these challenges, it is essential to understand and manage global material cycles effectively. This presentation will introduce quantitative approaches to modelling, simulating, and evaluating global material cycles using industrial ecology methods. Combining material flow analysis, life cycle assessment, and raw material criticality assessment allows us to understand the complex interactions between human activities, natural systems, and material flows. The presentation will draw on my research experience in developing and applying innovative methods for studying mining, production, recycling, and dissipation of metal and mineral resources. The work is rooted in industrial ecology, which views human activities as part of the natural environment and seeks to optimise material and energy flows to minimise environmental impacts. Using case studies and examples, I will demonstrate how our Ecological Resource Technology group at the University of Bayreuth applies these methods to evaluate the sustainability of global material cycles, specifically focusing on strategically relevant technologies in energy, mobility, and electronics. The presentation will highlight the benefits of our approaches, such as their ability to provide a comprehensive understanding of material cycles, and their limitations, such as the complexity of data collection and analysis, and discuss future research directions for improving the understanding and management of global material cycles.

11.12.2024, 12:15h
Carsten Hartmann
Fachgebiet Stochastik und ihre Anwendungen, BTU Cottbus-Senftenberg
Duality of estimation and control, with application to the simulation of rare events

Abstract: A computational problem in statistics concerns the reduction of the variance of an estimator of some quantity of interest. A typical example is rare event simulation, for which the standard deviation of most Monte Carlo estimators can be orders of magnitude larger than the quantity of interest. In this talk, I will explain how the development of variance reduction algorithms for rare events can benefit from exploiting variational principles from statistical mechanics in combination with techniques from stochastic optimal control. A particular focus will be on stochastic differential equations and rare events that involve unbounded random stopping times, which pose a particular challenge because the computational complexity of the rare event simulation also depends on average simulation time per sample that is not controlled by the variance. I will present recent results that reveal an unexpected connection between different rare event simulation algorithms and discuss further applications of statistical mechanics principles in computational stochastics.

08.01.2025, 12:15h
Dominik Kamp
Wirtschaftsmathematik, University of Bayreuth
A Delta-Debugger for Mixed-Integer Programming Solvers

Abstract: Recent performance improvements of mixed-integer programming (MILP) solvers went along with a significantly increased complexity of their source codes. This poses challenges in investigating solver behavior, especially if something goes wrong due to implementation errors (aka bugs). In this talk, an open-source delta-debugger for MILP solvers is presented. Delta-debugging is a general trial-and-error approach to isolate the cause of a software failure by simplifying the input data for an implemented algorithm. In practice, applying this tool within the development of the open-source MILP solver SCIP contributed to an increase of approximately 71 % more bugfixes than in the year before. As highlighted in case studies, instances which trigger bugs in reasonable time could usually be reduced to a few variables and constraints in less than an hour. The resulting input instance then makes it significantly easier to discover the root cause of the solver behavior of interest. This way, even performance analysis could already benefit of this automized approach to generate simple performance-adversarial examples.

22.01.2025, 12:15h
Philipp Braun
School of Engineering, Australian National University Canberra
Orchestrating control laws for reach-and-avoid problems: Lyapunov based approaches

Abstract: Control design for robotic systems guaranteeing safety and convergence properties in cluttered environments is intrinsically challenging due to their potentially conflicting objectives. While several research streams tackle the problem from different angles, a general solution for nonlinear dynamical systems is still out of reach. In this presentation we discuss difficulties, solution concepts and tools in controllers designs for reach-and-avoid problems (i.e., simultaneous target set stabilization and obstacle avoidance). While various approaches exist, we focus on the presentation of related Lyapunov methods.

29.01.2025, 12:15h
Michael Baumann
Applied Mathematics, University of Bayreuth
On fair outcomes and how to detect them - Rabin'93 revisited

Abstract: Rabin fairness was published in 1993. We explain the concept, discuss questions, and clarify some notable points. Further, we show whether and how fairness equilibria can be calculated via Python/SymPy.

05.02.2025, 12:15h
Ronan Richter
Wirtschaftsmathematik, University of Bayreuth
Methods and measurements for robust adversarial examples

Abstract: After demonstrating their capabilities in everyday tasks, this machine learning models will get increasingly prevalent in critical applications as well. However, their vulnerability to adversarial examples is one of the longest known traits, that raises concerns about their reliability. Building on a previous talk in this seminar, this talk will focus on the creation and analysis of robust adversarial examples for deep neural networks. These kinds of adversarial examples are of particular interest, since they may abstract away from one specific AI model towards a more general class of DNNs. Thus, robust adversarial examples can give insights to the broader limitations of AI approaches and thus help assess the trustworthiness of AI systems. Particularly, in this talk, various methods, from literature and newly designed ones, for generating these adversarial examples will be presented. In this context, we also see, how several notions of adversarial examples exist in different context. Furthermore, will discuss measurements for the robustness of an adversarial example and evaluate the various methods on small examples based on these measurements.

Tuesday, 18.02.2025, 10:00h, Room S74, NWII
Giulia Giordano
University of Trento, Italy
Optimal control of compartmental epidemic models with an arbitrary number of infected and non-infected compartments
(jointly with the Oberseminar Numerics, Optimization, and Dynamical Systems)

Abstract: Epidemiological models describing the spread of infectious diseases within a population are fundamental to enable the design of control approaches that optimally curb the contagion, while coping with uncertainty in the parameter values. In the first part of the talk, we consider epidemic models formed by a positive compartmental system that includes various infected categories (e.g. asymptomatic, symptomatic, quarantined), fed back by a positive interaction representing contagion. Techniques from optimal and robust control theory allow us to assess the sensitivity of the model to uncertain parameter values. We also formulate an optimal control problem on a finite horizon and we discuss the peculiarities of its solution that, given the structure of the model, can be obtained in one iteration, without resorting to shooting procedures. On an infinite horizon, we show that the Hamilton-Jacobi-Bellman equation can be solved exactly. In the second part of the talk, we consider a general class of epidemic models with an arbitrary number of infected and non-infected (e.g. susceptible, vaccinated) compartments and we formulate an optimal vaccination control problem to minimize the number of infections and the cost of vaccination. We show that the sequential quadratic Hamiltonian (SQH) scheme, for which we offer rigorous global convergence guarantees, can solve the problem efficiently. As a main case study, we consider the dynamics of the COVID-19 pandemic, captured by the SIDARTHE model and its extensions.

Tuesday, 18.02.2025, 11:00h, Room S74, NWII
Rami Kats
University of Trento, Italy
Oscillations in strongly 2-cooperative systems and their applications in systems biology
(jointly with the Oberseminar Numerics, Optimization, and Dynamical Systems)

Abstract: The emergence of sustained oscillations (via convergence to periodic orbits) in high-dimensional nonlinear dynamical systems is a highly non-trivial question with important applications in systems biology, including the understanding of bio-molecular oscillators ruling cell life-cycle and metabolism, as well as circadian rhythms in hormone secretion, body temperature and metabolic functions. In systems biology, the mechanism underlying such widespread oscillatory biological motifs is still not fully understood. From a mathematical perspective, the study of sustained oscillations is comprised of two parts: (i) showing that at least one periodic orbit exists and (ii) studying the stability of periodic orbits and/or characterising the initial conditions which yield solutions that converge to periodic trajectories. In this talk we will focus on a specific class of nonlinear dynamical systems that are strongly 2-cooperative. Employing results from the theory of cones of rank k, the spectral theory of totally positive matrices and Perron-Frobenius theory, we will show that strongly 2-cooperative systems admit an explicit set of initial conditions of positive measure, such that every solution emanating from this set converges to a periodic orbit. We will further demonstrate our results using the n-dimensional Goodwin oscillator and a 4-dimensional biological oscillator based on RNA–mediated regulation.

Tuesday, 01.04.2025, 10:00h, Room S82, NWII
Vladimir Veliov
Technical University of Vienna, Austria
Combining in-host and epidemiological models
(jointly with the Oberseminar Numerics, Optimization, and Dynamical Systems)

Abstract: The dynamics of the epidemic diseases at population level is influenced by the disease status of the individuals, that is, by in-host processes. Epidemiological parameters/functions such as infectiousness, immunity level, rate of gain/wane of immunity, recovery rate, mortality rate, etc. represent aggregated values of corresponding in-host parameters.
The talk begins with a short review of in-host virus disease modelling with involvement of innate and adoptive immune system. In parallel, we consider a particular epidemic model with dynamic immunity, which we combine with a simple in-host model to obtain a joint immuno-epidemiological model with reinfection. The issue of data availability and parameter evaluation will be discussed.

02.04.2025, 12:15h, Room S82, NWII
Vedran Krčadinac
University of Zagreb, Croatia
On mosaics of projective planes

Abstract: In the first part of the talk I will give a gentle introduction to finite projective planes and their generalization, combinatorial designs. I will introduce mosaics of designs and give a short overview of the topic. I will then focus on a particular case where existence is an open problem: mosaics of projective planes of order four. The deceptively simple-looking problem is to find a 21 by 21 matrix with zeros on the diagonal and entries 1, 2, 3 and 4 elsewhere. Each nonzero entry should appear five times in every row and column and exactly once in the same position for every pair of rows or columns. I will talk about attempts to construct such an object by computer and seek advice from the MODUS audience on how to approach the problem computationally.

Summer Semester 2024

24.04.2024, 12:15h
Julia Slipantschuk
Dynamical Systems and Data, Universität Bayreuth
Transfer operators and applications

Abstract: A fruitful approach for providing effective statistical descriptions of chaotic dynamical systems is to lift the dynamics to linear evolution operators on infinite dimensional Banach spaces. Spectral data of these operators, known as Koopman or transfer operators, yield insight into the long-term behaviour of the underlying system. In this talk I will give a short introduction into the spectral theory of transfer operators associated to (mostly) low-dimensional discrete chaotic dynamical systems, and present a complete description of eigenvalues of these operators for a certain special class of non-linear maps. In general, such explicit descriptions of eigendata are rarely available, and in various applications algorithms or numerical schemes to approximate these are required. I will present strong spectral convergence and consistency results for a popular numerical approximation scheme -- extended dynamic mode decomposition (EDMD) -- applied to examples of chaotic systems given by one-dimensional expanding maps.

Monday, 6.05.2024, 10:30h, S102
Talk in the lecture series Scientific Computing, Coffee and Tea at 10:00h
Luca Bonaventura
Dipartimento di Matematica, Politecnico di Milano, Italy
A deeper look at shallow water models

Abstract: The derivation of the classical hydrostatic, vertically averaged shallow water equations will be presented and revisited to highlight typical conceptual errors. The choice of time scales determined implicitly by the identification of standard flow regimes will be discussed. Examples of non-hydrostatic shallow water models will be presented and the prospects for their rigorous derivation will be discussed.

8.05.2024, 12:15h
Jirka Vomlel
Institute of Information Theory and Automation, Academy of Sciences of the Czech Republic   and   Faculty of Arts, Charles University, Prague
Uncovering Relationships: Using Bayesian Networks to Analyze Open-Minded Thinking and Conspiracy Beliefs in Czech University Candidates

Abstract: I will begin with a brief introduction to one of the glass-box models of artificial intelligence called Bayesian Networks (BNs), a model named after the English statistician Rev. Thomas Bayes. BNs represent a probabilistic model that can help visualize relationships between variables. I will illustrate the use of Bayesian networks on data from a Czech university entrance exam. This data included a test of open-minded thinking designed by Jonathan Baron and also examined students' attitudes towards various conspiracies. Using BNs, we can identify conspiracy clusters and their relationships with open-minded thinking.

15.05.2024, 12:15h
April Herwig
AG Numerik komplexer Systeme, TU München
An introduction to pseudospectra and application to validated computational dynamics

5.06.2024, 12:05h, Kolloquium zur Masterarbeit
Florian Streifinger
Modellierung, Analyse und Simulation von Krankheiten

5.06.2024, 13:00h, Kolloquium zur Masterarbeit
Kilian Bernreuther
Analyse, Simulation und Optimale Steuerung eines Krankheitsmodells

12.06.2024, 12:15h
Lisa Jagau
Lehrstuhl Scientific Computing, Universität Bayreuth
Hydrodynamische Modellierung von Mikroplastik-Transport und Sedimentierung in Seen

Abstract: Mikroplastikpartikel (MP) sind für Organismen in der Hydrosphäre potenziell schädlich. Um die Exposition und das damit verbundene Risiko besser einschätzen zu können, muss das Transport- und Sedimentierungsverhalten von MP in aquatischen Systemen quantifiziert werden. Wir erstellen ein dreidimensionales hydrodynamisches Modell und validieren dieses anhand gemessener Daten für den Großen Brombachsee. Anschließend nutzen wir das Modell, um MP-Transport und Sedimentierung zu modellieren.
Unser Schwerpunkt ist die Modellierung von Polymeren mit unterschiedlichen Dichten und Partikelgrößen, um Muster der Partikelverweilzeit und Sedimentierung zu identifizieren. Die Verteilung von MP im Rechengebiet wird sowohl von der Partikeldichte als auch von der Partikelgröße stark beeinflusst. Kleinere, leichtere Partikel verteilen sich über die gesamte horizontale Ausdehnung des Sees und Partikel mit höherer Dichte oder größerem Durchmesser setzen sich in einem begrenzten Bereich um den Einströmungsort ab, was auf eine höhere Absetzgeschwindigkeit hindeutet.

19.06.2024, 12:15h
Ronan Richter
Lehrstuhl Wirtschaftsmathematik, Universität Bayreuth
Towards Analyzing small DNNs by Robust Adversarial Examples created with MILPs

Abstract: As DNN-based systems are gaining increasing popularity and governments begin to regulate AI systems, there is a growing demand for methods to analyze the trustworthiness of a DNN and the limits of its application. One classical method of showing weaknesses of DNNs are Adversarial Examples. These are slightly modified versions of input data, that lead a DNN into wrong classifications. As Fischetti and Jo (2018) have shown, Adversarial Examples can systematically be generated by ILPs, leading to Adversarial Examples, that are provably optimal in respect to a given criterion, e.g. the distance to some given input data. Thus, these methods may be used to analysis the vulnerability of a DNN. However, the structure of these examples highly depends on the parameters of the network, which are often unknown to an attacker. For a broader view, a lager class of DNNs should be considered. Thus, a mixed-integer programming model for generating Adversarial Examples, that are robust with respect to small changes in the weights and biases of a DNN will be given. For relaxations of the model, we will illustrate the impact of robustification on Adversarial Examples. Especially, we present experimental results on the influence of training data on the distance of Adversarial Examples and on the transferability of our examples.

26.06.2024, 12:15h
Bismark Singh
Operational Research Group, University of Southampton, UK
Balancing accessibility and fairness: Optimally closing recycling centers

Abstract: Typically, within facility location problems, fairness is defined in terms of accessibility of users. However, for facilities perceived as undesirable by communities hosting them - such as recycling centers - fairness between the usage of facilities becomes especially important. We develop a series of optimization models for the allocation of populations of users to such recycling centers such that access for users is balanced with a fair utilization of facilities. The optimality conditions of the underlying nonconvex quadratic models state the precise balance between accessibility and fairness. We define new classes of fairness and a metric to quantify the extent to which fairness is achieved in both optimal and suboptimal allocations.
Within the state of Bavaria in Germany, such centers have closed in the last few decades. Using mobility survey data we show how selective closures of these centers can still lead to high levels of recycling access. Our analysis ensures that even when 20% of facilities are closed smartly, the median travel distance by residents to their assigned recycling center increases by only 450 m. Additionally, we find Bavaria suffers from disparity in recycling patterns in rural and urban regions, both in terms of motivation to recycle and the locations of the facilities. We promote a policy that favors retention of recycling centers in rural regions by reserving 75% of open facilities to be in rural areas, while selectively closing facilities in urban regions, to remove these regional differences.
This work is based on two recently published articles in the INFORMS Journal on Computing and Networks.

3.07.2024, 12:15h
Alfred Wassermann
Lehrstuhl Mathematik und ihre Didaktik, Universität Bayreuth
Solve Ax=b with 0/1 variables

Abstract: Finding 0/1 solutions for a system of linear equations Ax=b, where A is an integer matrix and b is an integer vector, is a well-known NP-complete problem with many applications, e.g. in combinatorics, coding theory, and cryptography. In fact, it is one version of integer linear programming. There are many practical algorithms available to solve this problem, for example, integer linear programming solvers like CPLEX and Gurobi, backtracking algorithms like "dancing links", or SAT solvers.
In this talk, the author's software "solvediophant" is presented as an alternative to solve such kinds of problems.
For instances where the vector b consists of large integers, it seems to be faster than other approaches. Moreover, in contrast to other backtracking approaches, the matrix A might also have integer entries of mixed signs. So far, many new combinatorial objects and new record-breaking error-correcting codes have been constructed with "solvediophant".
The algorithm is based on lattice basis reduction together with an exhaustive enumeration of points in a high dimensional lattice. Additionally, solvediophant can be used with a non-standard backtracking approach called "least discrepancy search" which seems to be especially well suited for reduced lattice bases.
solvediophant solves instances up to 1500 variables and can be used not only for 0/1 variables but also for integer variables in a finite interval.

Monday, 8.07.2024, 12:15h, H33 (Inf)
Talk in the lecture series of the CRC 1357 Microplastics
Gholamreza Shiravani
Niedersächsischer Landesbetrieb für Wasserwirtschaft, Küsten- und Naturschutz
3D-Modeling of microplastic transport in tidal rivers by including the bio-geological effects

Abstract: Microplastic-transport mechanisms are using a developed 3D-numerical model for a tidal river from microplastic-sources (point form and diffusive sources) to accepting seas presented. The role of the bio-geological parameters on the fate of microplastics and their corresponding parametrization are explained. Moreover, the model capabilities and uncertainties are through the application of the model for microplastics-transport in the Weser estuary discussed. Finally, the model performance is through the comparison of the model results with microplastic-measurements for the Weser river evaluated. (Participation via zoom possible, see the announcement on the CRC Microplastic seminar site.)

17.07.2024, 12:15h
Ronny Bergmann
Department of Mathematical Sciences, NTNU Trondheim, Norway
The Riemannian Difference of Convex Algorithm in Manopt.jl

Abstract: In many applications nonlinear data is measured, for example when considering unit vectors, rotations, or (bases of) subspaces of a vector space. Modelling this on a Riemannian manifold allows to both reduce the dimension of the data stored as well as focusing on geometric properties of the measurement space compared to constraining a total space the data is represented in. In optimisation this yields unconstrained optimization algorithms, where we have to take the geometry of the optimization domain into consideration. In this talk we consider the task of minimizing the difference of two convex functions defined on a manifold and present the Difference of Convex Algorithm. To make algorithms in general more accessible, we then present the two Julia packages Manifolds.jl and Manopt.jl, that allow to define and use Riemannian manifolds and optimization algorithms employing numerical differential geometry, respectively.

Winter Semester 2023/2024

29.11.2023, 12:15h
Mathias Oster
Institute for Geometry and Practical Mathematics, RWTH Aachen University
Empirical Tensor Train Approximation in Optimal Control
Abstract: We display two approaches to solve finite horizon optimal control problems. First we solve the Bellman equation numerically by employing the Policy Iteration algorithm. Second, we introduce a semiglobal optimal control problem and use open loop methods on a feedback level. To overcome computational infeasability we use tensor trains and multi-polynomials, together with high-dimensional quadrature, e.g. Monte-Carlo. By controlling a destabilized version of viscous Burgers and a diffusion equation with unstable reaction term numerical evidence is given.

13.12.2023, 12:15h
Johannes Margraf
Artificial Intelligence in Physico-Chemical Material Analysis, Universität Bayreuth
Designing Molecules and Materials with Machine Learning

10.1.2024, 12:15h
Rainer Hegselmann
Frankfurt School of Finance & Management
Two-armed bandits versus Carnapian truth seekers and epistemic free riders with bounded confidence

17.1.2024, 12:15h
Agnes Koschmider
Wirtschaftsinformatik und Process Analytics, Universität Bayreuth
How to efficiently pre-process unstructured data for process mining?
Abstract: Process mining is a promising approach to find additional patterns in data and in that way to give new insights into the data. The challenge of process mining on unstructured data is to efficiently pre-process the data in a way that process mining can give additional insights. If the data is not clustered appropriately, the result might be distorted (i.e., there is a correlation between clustering and the discovered process model). This talk presents approaches for change point detection and encodings allowing to divide the pre-processed data representative for process mining.

31.1.2024, 12:15h
Mario Sperl
Angewandte Mathematik, Universität Bayreuth
Curse-of-dimensionality-free approximations of optimal value functions with neural networks

7.2.2024, 12:15h
Dominik Kamp
Wirtschaftsmathematik, Universität Bayreuth
Nachfragedynamische Erweiterungen für das Stochastic Guaranteed Service Model auf realistischen Lagernetzen

Summer Semester 2023

10.5.2023, 12:15h
Janosch Hennig
Chair of Biochemistry IV - Biophysical Chemistry, Universität Bayreuth
AI-driven revolutions in structural biology: a new dawn for biomolecular NMR spectroscopy

17.5.2023, 12:15h
Janin Henkel-Oberländer
Chair of Nutritional Biochemistry, Universität Bayreuth
Challenges in histological tissue analysis

31.5.2023, 12:15h
Rainer Hegselmann
Frankfurt School of Finance & Management
Bounded Confidence Revisited

14.6.2023, 12:15h
Karl Worthmann
TU Ilmenau
Data-based prediction of dynamical (control) systems

28.6.2023, 12:15h
Athanasios Antoulas
Rice University, Houston, USA
Interpolatory methods for model reduction and the Loewner framework

5.7.2023, 12:15h
Ruben Mayer
Lehrstuhl für Data Systems, Universität Bayreuth
Recent Advances in Graph Partitioning for Increasing the Performance of Large-Scale Distributed Graph Processing
Abstract: Graph-structured data is found in various domains such as social networks, websites, and recommendation networks. To analyze large graphs and gain high-level insights, distributed graph processing frameworks such as Spark/GraphX and Giraph have been established. For distributed processing, the graph needs to be split into multiple partitions, while the cut size and balancing of the partitions need to be optimized. This problem is known as graph partitioning.
In this talk, I will summarize recent advances of graph partitioning and introduce important new concepts that have been developed in my group. First, two novel techniques that reduce the memory footprint of graph partitioning while maintaining a high partitioning quality: Hybrid Edge Partitioning and Two-Phase Streaming. Second, EASE, a framework for optimizing the choice of partitioning technique for a given graph and processing algorithm. EASE is based on machine learning and achieves better performance than a manual partitioner selection based on heuristics. Finally, I will provide an outlook on open problems.

12.7.2023, 12:15h
Thomas Bocklitz
AG Künstliche Intelligenz in der Spektroskopie und Mikroskopie, Universität Bayreuth
AI for spectroscopy and microscopy: inverse modelling and data modelling tasks

19.7.2023, 12:15h
Michael Wilczek
Lehrstuhl Theoretische Physik I, Universität Bayreuth
Insights into turbulence from fully resolved simulations
Abstract: Fluid turbulence plays an important role in nature and engineering processes. Despite its importance, many aspects still remain to be understood. From a physics perspective, one challenge is to derive theories of turbulence which allow us to understand and predict nontrivial statistical features of turbulence such as the frequent occurrence of extreme events. Fully resolved turbulence simulations provide a useful framework to investigate the spatio-temporal properties of turbulence. In this presentation, I will discuss some recent works which demonstrate how theoretical modeling and simulations can be combined to better understand fundamental aspects of turbulence.