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:

Xuân Tùng Nguyễn, INFN / University of Padua

Title: Neural connectivity: a parallel in vitro and in silico analysis

Abstract:

In this talk, we will analyze neuronal recordings obtained by multi-electrode arrays (MEAs) on in vitro neuronal cultures from prof. Vassanelli’s laboratory, trying to estimate synaptic connectivity between the neurons. To this aim, we will adopt several time series analysis techniques, including transfer entropy and coincidence analysis. We will develop a silico dynamical model of the biological neuronal network, building on the standard Izhikevich model and adding some realistic constraints on the network topology. The model will be used to simulate neuronal time series and validate the connectivity inference methods. The end goal of the project is being able to reliably estimate how synaptic connectivity spontaneously varies in time.

How to join online

You can join online via Zoom, using the following link:
https://uni-kl-de.zoom.us/j/63123116305?pwd=Yko3WU9ZblpGR3lGUkVTV1kzMCtUUT09

Dr. Max Sagebaum, RPTU Kaiserslautern-Landau

Title: Introduction to cppfront – a new syntax for C++ – and compile time regular expression generation

Abstract:

cppfront is a compiler for an experimental alternate syntax for C++ (aka Cpp2) initiated and driven by Herb Sutter. The project wants to explore ideas for a new syntax of C++ that makes it simpler, safer, and faster to write C++. cppfront does not want to create a C++ sucessor. It wants to evolve C++. In this talk, we will present the current status of the project and introduce the new syntax. Afterwards, the implementation of a compile-time regular expression generator is presented.

How to join online

You can join online via Zoom, using the following link:
https://uni-kl-de.zoom.us/j/63123116305?pwd=Yko3WU9ZblpGR3lGUkVTV1kzMCtUUT09

Ivan Barranco Gomez, School of Civil, Aerospace and Design Engineering; University of Bristol

Title: Optimising Ion Thruster Performance: Advances in Particle-in-Cell and Monte Carlo Simulations

Abstract:

Motivated by the need to enhance physical modelling and optimization techniques in the design of ion thrusters, this seminar delves into the challenges and advancements in simulating ion thrusters using Particle-in-Cell (PIC) codes coupled with Monte Carlo methods. In this talk, we investigate the physical modelling of ion thrusters, with a particular focus on plasma physics. We address the inherent complexities in PIC codes, which are widely used for simulating plasma behaviour in ion thrusters. These codes combine kinetic and fluid descriptions of plasma particles, relying heavily on Monte Carlo methods for handling collisions and other stochastic processes. Finally, the seminar outlines future research directions, emphasizing the need for continued advancements in simulation techniques and optimization algorithms to further refine the modelling and performance of ion thrusters in aerospace applications.

How to join online

You can join online via Zoom, using the following link:
https://uni-kl-de.zoom.us/j/63123116305?pwd=Yko3WU9ZblpGR3lGUkVTV1kzMCtUUT09

Max Aehle, RPTU Kaiserslautern-Landau

Title: Algorithmic Derivatives of Electromagnetic Shower Simulations

Abstract:

Applying algorithmic differentiation to particle simulations like Geant4 would allow us to evaluate derivatives of simulation outputs with respect to inputs, e.g. of the mean energy depositions in calorimeter layers with respect to geometry parameters. Such derivatives could become instrumental for a number of application like detector optimization or parameter fitting in HEP. However, besides the technical challenge of applying AD to over one million lines of code, there is a number of mathematical/statistical challenges: Does the high density of discontinuities, induced for example by ‘if’ and ‘while’ statements in the code, cause problems in the derivative computation? Are we allowed to treat random numbers like constants with respect to AD? How large is the error when the mean pathwise derivative is evaluated as a proxy for the actual derivative of expectancies computed by Monte-Carlo algorithms?

To approach these questions, we have applied the operator-overloading AD tool CoDiPack to the compact G4HepEm/HepEmShow package for the simulation of electromagnetic showers in a simple sampling calorimeter. After disabling multiple scattering in the simulation, our pathwise gradient estimator approximates derivatives of energy depositions with only a small bias, which proved unproblematic in a simple optimization study. In this talk, we will report on our methodology and give an overview on our findings.

How to join online

You can join online via Zoom, using the following link:
https://uni-kl-de.zoom.us/j/63123116305?pwd=Yko3WU9ZblpGR3lGUkVTV1kzMCtUUT09

Daniel Maue, RPTU Kaiserslautern-Landau

Title: Depth-Aware Traffic Sign Recognition – A Multi-Stage 3D Localization & Classification Pipeline for Traffic Signs

Abstract:

Recognizing objects within images is a key task in computer vision. There are many traditional algorithmic approaches for locating objects in images and with the rise of research of artificial intelligence also deep learning recognition models were proposed. State-of-the-art methods already achieve matching or better performance than humans. The goal of this thesis is to give the common recognition task a third dimension, by not only locating objects in the 2D image but also estimating the depth of this object within the captured scene. Especially traffic sign recognition would benefit from the additional depth information, since capturing the distance and thus the order of oncoming traffic signs could greatly improve safety in vehicles since the depth can have important semantic meanings to the current and future driver assistance system. Therefore this bachelor thesis focuses on proposing a multi-stage 3D localization pipeline for traffic signs, by thoroughly analyzing and comparing state-of-the-art methods in object recognition and depth estimation. With that, these two topics are first introduced by covering related work and the needed theoretical background. Additionally, the considered approaches for object recognition and depth estimation are explained and finally evaluated in experiments, to find the most suitable methods for a 3D localization pipeline.

Johannes Albert Stegmann, RPTU Kaiserslautern-Landau

Title: Improving Evaluation Performance for Large Derivative Computations Through Cache Optimisation Techniques in CoDiPack

Abstract:

This presentation will discuss optimisation techniques to improve CPU caching between statements of large derivative calculations in the C++ CoDiPack library. These enhancements focus on improving cache friendliness through the reordering of statements and reallocating identifiers given to the variables in the statements. Modifying identifiers aims to improve spatial locality caching, while reordering statements takes advantage of temporal caching.

Initial tests show that identifier selection significantly impacts evaluation performance. With a basic reallocation of identifiers, performance improvements of greater than 25% are possible. Investing a few minutes in upfront processing leads to substantial time savings in large derivative calculations. Additionally, how users write their programs using CoDiPack affects the performance of functionally identical tapes. The goal of these techniques is to reduce the dependency on the structure of the user’s program. The reordering of a tape is further used to remove redundant or unused statements.

To support these techniques, we have added functionality to CoDiPack to enable tape storage. This includes providing text and graphical methods for easier debugging and a binary file output for efficient storage outside the current program’s scope. The stored tape can later be reloaded into CoDiPack to reproduce a valid tape.

This presentation discusses the storage and cache enhancement methods, ideas for further improvement, and the preliminary results of the implemented solutions.

How to join online

You can join online via Zoom, using the following link:
https://uni-kl-de.zoom.us/j/63123116305?pwd=Yko3WU9ZblpGR3lGUkVTV1kzMCtUUT09

Prof. Sarah Neuwirth, Johannes Gutenberg University of Mainz

Title: Transforming HPC with Holistic Performance Engineering and the Pursuit of Reproducibility

Abstract:

HPC applications are evolving not only to include traditional modeling and simulation workloads but also so-called scale-out workloads, including artificial intelligence, big data analytics, deep learning, and complex workflows. Given the ever-increasing complexity of supercomputers and the emergence of exascale computing, these trends are leading to a widening gap between expected and observed peak performance. Therefore, holistic performance engineering is critical to bridge this gap through reproducible benchmarking, modeling, optimization, and analysis of large-scale HPC workloads.

This seminar will delve into innovative approaches to scale HPC performance engineering horizons and unlock the full potential of computational capabilities. Strategies for optimizing hardware and software utilization, identifying and mitigating performance bottlenecks, and implementing cutting-edge techniques to maximize scalability, efficiency, and productivity will be explored. Moreover, this talk will question to what extent reproducibility, replicability, comparability, and portability should influence the performance analysis cycle of workloads and supercomputing systems.

How to join online

You can join online via Zoom, using the following link:
https://uni-kl-de.zoom.us/j/63123116305?pwd=Yko3WU9ZblpGR3lGUkVTV1kzMCtUUT09

Lena Dittrich, RPTU Kaiserslautern-Landau

Title: An overview over publications simulating Ride-pooling services using the simulation framework MATSim

Abstract:

Ride-pooling Services, also often referred to as Ride-Hailing or Ride-Sharing Services, are becoming more and more common. Humanity is facing problems like climate change and increased population density, and we are increasingly aware of the necessity to change and adjust our way of life accordingly. To combat this, we are attempting to rebuild and adapt our infrastructure to our new challenges. Public transportation systems are an important part of that transition and, therefore, a useful and much needed area of research. An efficient system will attract more passengers who will substitute their private vehicle by public transportation, thereby saving not only carbon emissions but also space and resources that can be used more efficiently elsewhere.
This talk will give an overview over some research that has been done about Ridepooling services. We will look at publications in which different scenarios were simulated using the agent-based simulation framework MATSim to asses the necessary fleet size of a ride-pooling service provider.

How to join online

You can join online via Zoom, using the following link:
https://uni-kl-de.zoom.us/j/63123116305?pwd=Yko3WU9ZblpGR3lGUkVTV1kzMCtUUT09

Prof. Hans Hagen, Geometric Modeling Group, RPTU

There will be two short presentations:

Title: Differential Geometry as a ” tool box”

Abstract:

Geometric Modelling Algorithms need a strong mathematical basis.
One is Differential Geometry. We present in this talk some basic concepts
in an “intuitive way”. Geometry has a lot to do with intuition and imagination.
Same “tools” work in geometric modelling, some do not.
After this introductionary talk we will and should have a constructive and critical discussion.

Title: Killing Vector Fields — a concept for simulation and scientific visualization?

Abstract:

Killing Vector Fields ( named after Wolfgang Killing) are infinitesimal generators of isometries.
The flow generated by these fields are continuous isometries of the surface.
There is no distortion moving long these directions.
We present in this talk the basic ideas and the PDE system, which has to be solved.

How to join online

You can join online via Zoom, using the following link:
https://uni-kl-de.zoom.us/j/63123116305?pwd=Yko3WU9ZblpGR3lGUkVTV1kzMCtUUT09

Diksha Gupta, RPTU Kaiserslautern-Landau

Title: RNN-based Mutation Rate Prediction for SARS-CoV-2

Abstract:

The COVID-19 pandemic has thrusted the world into an unprecedented crisis, prompting
widespread efforts to understand the dynamics of viral transmission, adaptation,
and evolution. One of the key challenges in combating the pandemic is the emergence of
novel variants of the SARS-CoV-2 virus, which can potentially affect the efficacy of vaccines,
diagnostics, and therapeutics. Predictive modeling of COVID-19 mutation rates
has therefore become a critical area of research, offering insights into the evolutionary
trajectory of the virus and informing public health strategies.
This thesis presents a comprehensive analysis of the predictive performance of Long
Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) architectures in modeling
COVID-19 mutation rates. The primary objective of the study is to assess the efficacy
of these recurrent neural network models in capturing the complex patterns underlying
mutation dynamics and to evaluate their suitability for predictive tasks in bioinformatics.
The methodology involves collecting and preprocessing a diverse array of genomic
data, including viral genome sequences and associated metadata. Feature engineering
techniques are employed to represent genomic sequences in a format amenable to deep
learning models. LSTM and GRU architectures are then trained on the preprocessed
data to predict mutation rates associated with various genomic regions and mutation
types.
Through a series of experiments conducted on different mutation types and datasets,
the study systematically evaluates the performance of LSTM and GRU models. Performance
metrics such as root mean squared error, and mean absolute error are computed
to assess the models’ predictive capabilities. Additionally, visualization techniques are
employed to analyze the learned representations and gain insights into the underlying
mutation patterns. The findings of the study reveal the potential of LSTM and GRU
models in predicting COVID-19 mutation rates with high accuracy and precision.
Overall, this thesis contributes to the growing body of knowledge on computational
approaches to understanding viral evolution and offers valuable insights into the application
of deep learning models in COVID-19 research. By leveraging machine learning
techniques, researchers can gain a deeper understanding of the mechanisms driving viral
evolution and develop more effective strategies for combating the COVID-19 pandemic.

Prof. Hans Hagen, Geometric Modeling Group, RPTU

Title: Non-Standard Metrics for Machine Learning

Abstract:

Euclidean Metrics and L2-Norms are the dominating standards
— . BUT are they good enough in all applications ?—-
I do not think so. Finslertyp metrics , which deal with point locations and directions show a lot of promise of being influential in important research areas.
The Lagrange Multiplier technique is a very powerfull tool assuming the “right energy functionals”.
What about energy functionals depending on non – standard metrics for GANs ( Generative Adversal Networks) ??
This would open the door for using Lagrange Multipliers.
To start a constructive and critical discussion is the goal of this talk .

How to join online

You can join online via Zoom, using the following link:
https://uni-kl-de.zoom.us/j/63123116305?pwd=Yko3WU9ZblpGR3lGUkVTV1kzMCtUUT09