Prof. Dr. Nicolas Gauger, Chair for Scientific Computing (SciComp), TU Kaiserslautern

SciComp Seminar Series

Please contact Prof. Gauger, if you want to register for an online talk in our SciComp Seminar Series or just to register for the seminar.

A list of the already scheduled talks can be found –> here:

Prof. Dr. Bernat Font, Data-Informed Computational Fluid Dynamics (DI-CFD) research group, TU Delft, The Netherlands

Title: Data-informed CFD: From RL for active flow control to optimal numerical methods using AD

Abstract:

The advancement of scientific machine learning (SciML) has reshaped how we control, model, and compute turbulent flows. Together with the widespread use of graphics processing units, we are now in a unique position to perform scale-resolving simulations combining data-driven techniques with classic methods, yielding the data-informed CFD paradigm. In this talk, we will review established numerical methods and physical models which can be enhanced through data, including linear solvers, turbulence models, and active flow control techniques. We will also discuss the solver-in-the-loop optimization approach using automatic differentiation, thus avoiding well-known a-priori/a-posteriori errors. Last, we will conclude with basic guidelines in the use of SciML for CFD.

How to join online

You can join online via Zoom, using the following link:
https://uni-kl-de.zoom-x.de/j/69269239534?pwd=Z9UOzMpkhMjrxVhll3d49sNHFe9Fd1.1

Maoni Ngowa Msinda, Chair for Scientific Computing (SciComp), RPTU University Kaiserslautern-Landau

Title: Modeling Glioblastoma Plasticity Using Partial Differential Equations and Machine Learning

Abstract:

The integration of mathematical modeling and machine learning is opening new directions in computational oncology. In this work, we present a partial integro-differential equation framework for modeling glioblastoma plasticity, incorporating anisotropic diffusion, convection, nonlocal interactions, and nonlinear reaction dynamics. Numerical simulations are performed using finite-difference discretizations together with fourth-order Runge–Kutta time integration, while parameter estimation is formulated as a PDE-constrained optimization problem. To efficiently compute gradients for large-scale inverse problems, we employ reverse-mode automatic differentiation using the CoDiPack library combined with a two-level uniform checkpointing strategy. Gradient-based optimization methods, including gradient descent and L-BFGS-B, are compared for recovering model parameters from observed tumor data. Finally, we discuss future directions involving spatially varying parameters, clinical data assimilation, and machine learning approaches for accelerated inverse modeling.

How to join online

You can join online via Zoom, using the following link:
https://uni-kl-de.zoom-x.de/j/69269239534?pwd=Z9UOzMpkhMjrxVhll3d49sNHFe9Fd1.1

M.Sc. Faezeh Nassajian Mojarrad, Max Planck Institute for Software Systems, Kaiserslautern

Title: The Two-Step Stochastic Quadratic Method: A New Approximation Scheme for Solving SDEs

Abstract:

Solving Itô stochastic differential equations (SDEs) accurately is a persistent challenge in applied mathematics and quantitative modeling in finance, particularly in finance. In this seminar, we first discuss the applications of SDEs in finance before introducing a novel numerical framework: the Two-Step Stochastic Quadratic Method. Unlike standard step-by-step approaches, our method advances the solution by iteratively computing the unknown function across pairs of future time points, utilizing quadratic polynomial approximations over subdivided time intervals. We will explore the theoretical foundations of this new scheme, detailing the rigorous analysis of its convergence, consistency, and stability. Finally, through applied numerical experiments, we will demonstrate that this approach delivers superior accuracy when benchmarked against classic baseline techniques such as the Euler-Maruyama and Milstein schemes.

How to join online

You can join online via Zoom, using the following link:
https://uni-kl-de.zoom-x.de/j/69269239534?pwd=Z9UOzMpkhMjrxVhll3d49sNHFe9Fd1.1

Ilja Kratschmer, RPTU Kaiserslautern-Landau, Kaiserslautern

Title: Curriculum Learning: Task Sequencing in a Grid-Survival Game

Abstract:

Curriculum learning aims to improve reinforcement learning by controlling the order in which an agent encounters training tasks. We first introduce Deep Q-Networks (DQNs) and how neural networks approximate action-value functions in sequential decision problems. We then present the principles of curriculum learning for reinforcement learning, focusing on generation, selection and sequencing of tasks. These concepts are illustrated using a grid-survival game in which a DQN agent must avoid enemies and obstacles in order to survive as long as possible. We describe the environment, the training setup and several possible sequencing strategies and evaluate them against a no-curriculum baseline trained directly on the target task with the same sample budget. Finally, we discuss whether and when curriculum learning improves on this baseline.

How to join online:

You can join online via Zoom, using the following link:
https://uni-kl-de.zoom-x.de/j/69269239534?pwd=Z9UOzMpkhMjrxVhll3d49sNHFe9Fd1.1

Naukshatro Bose, RPTU Kaiserslautern-Landau, Kaiserslautern

Title: Field Inversion and Physics-Informed Data Assimilation using JAX-FEM

Abstract:

Data Assimilation for PDEs has been an active field with particularly interesting applications in enhancement of CFD Turbulence models. This has multiple facets, however, Field Inversion allows data assimilation for latent fields, which is very useful for both, generalization to unseen flows and study of latent dynamics in a PDE system where not all parameters are known. This talk discusses the approaches to field inversion and data assimilation on the Poisson Equation using the jax-fem library of differentiable FE solvers.

How to join online:

You can join online via Zoom, using the following link:
https://uni-kl-de.zoom-x.de/j/69269239534?pwd=Z9UOzMpkhMjrxVhll3d49sNHFe9Fd1.1

Prof. Dr. Bernd Ulmann, FOM University of Applied Sciences in Frankfurt/Main

Title: Analog and Hybrid Computing in the 21st Century

Abstract:

Computation by electronic analogues is gaining more and more interest in many areas of research as well as industrial applications. Being one central pillar of unconventional computing, this particular paradigm offers unmatched advantages with respect to energy efficiency as well as computational power for certain problem domains not limited to the treatment of systems of ODEs and PDEs but also areas such as AI. This talk gives an introduction into modern analog and hybrid computing with examples and an outlook to future application areas.

More about the speaker:

Bernd Ulmann studied mathematics with a minor in philosophy at Johannes Gutenberg University in Mainz. After several years in industry (process automation in oil refineries) he wrote his PhD at Hamburg University about analog computers, his life long favourite topic. He is professor at the FOM university of applied sciences in Frankfurt/Main and is a co-founder of anabrid GmbH where modern, highly integrated analog computers are developed. In his free time he collects, restores and builds analog computers (what else 🙂 ).

How to join online

You can join online via Zoom, using the following link:
https://uni-kl-de.zoom-x.de/j/69269239534?pwd=Z9UOzMpkhMjrxVhll3d49sNHFe9Fd1.1

Dr. Bora Kalpakli, LIKUA, Heidelberg, Germany

Title: NeuralFlow: Learning CFD Solution Operators Through Solver Backpropagation

Abstract:

Computational Fluid Dynamics (CFD) traditionally obtains flow solutions by repeatedly
applying numerical operations that reduce the imbalance in the discretized flow field until
a converged state is reached. Although modern CFD algorithms are accurate and robust,
their computational cost remains a limitation for applications requiring repeated simulations,
large parameter studies, optimization, or near-real-time prediction. NeuralFlow investigates
a different approach: a neural operator is trained not to reproduce a database of previously
computed CFD solutions, but to learn how the CFD solver itself drives a flow field toward
convergence.
NeuralFlow is based on a Graph Neural Operator (GNO) defined directly on an unstructured
finite-volume mesh. Mesh cells are represented as graph nodes and their physical
connections as edges. Starting from an initial flow field, the GNO predicts corrections to the
primitive variables and is applied recurrently over multiple rollout steps. This local graph
formulation allows the same learned operator to act on different parts of an unstructured
mesh while progressively transforming the current state toward a converged flow solution.
A central feature of NeuralFlow is that no reference CFD solution data are required for
training. It is also fundamentally different from a Physics-Informed Neural Network (PINN):
the governing differential equations are not imposed directly on the neural network as a
collocation-based loss. Instead, the predicted flow state is evaluated by the existing CMPS
CFD solver. CMPS performs its normal finite-volume operations, including numerical flux
evaluation, boundary-condition treatment, residual assembly, and Jacobian evaluation. Exact
solver sensitivities are obtained by automatic differentiation and are used to construct the
gradient with respect to the GNO output. This solver-derived gradient is then backpropagated
through the neural network, allowing the GNO weights to learn directly from the behaviour
of the numerical CFD algorithm.
The seminar will present the NeuralFlow architecture, the graph representation of finitevolume
meshes, recurrent rollout training, and the coupling between forward-mode automatic
differentiation inside the CFD solver and reverse-mode backpropagation inside the neural
network. Initial results show that the learned operator can progressively reduce CFD
residuals and reproduce laminar-flow behaviour without supervised solution data. The longterm
objective is a generalizable learned CFD solver that retains the numerical and physical
structure of conventional CFD while transferring part of the iterative solution process to a
trainable graph neural operator.

How to join online

You can join online via Zoom, using the following link:
https://uni-kl-de.zoom-x.de/j/69269239534?pwd=Z9UOzMpkhMjrxVhll3d49sNHFe9Fd1.1