HGS MathComp Upcoming Events

The major objective of the “Good Scientific Practice” workshops is to know and understand the basic rules and values of the responsible conduct of research in all its stages, according to local, national and international regulations and guidelines. The participants will explore the differences and grey areas between good scientific practice, questionable research practice and scientific misconduct. They will learn how misconduct can be recognized and prevented, how it should be addressed and dealt with in case it occurs, and what damage it can cause if handled improperly. The participants will learn to develop appropriate solutions for difficult situations in the process of science. They are encouraged to speak with colleagues about mistakes and problems.

Please note: Your tutor expects you to participate actively in the interactive workshop. You should be willing to turn on your camera.

Contents:

- Definitions of good scientific practice and scientific misconduct
- Degrees and extent of scientific misconduct
- Examples for responsible and irresponsible conduct of research
- Data management
- Authorship and the process of publication
- Mentoring and supervision
- Conflict management: how to deal with scientific misconduct
- Reactions to scientific misconduct
- Local, national and international guidelines and regulations

Part 1: Time Series Models
1. Notions of stationarity.
2. Vector Autoregressive models and estimation.
3. Spectral density matrices and estimation.
4. Overview of sampling properties of the estimators.

Part 2: Covariance and Matrices
1. Projections, linear regression and partial correlation.
2. Connection between partial correlation and the inverse covariance matrix.
3. The Schur complement for finite and infinite dimensional matrices.
4. Some properties of the inverse of infinite dimensional matrices.

Part 3: Gaussian Graphical models in time series
1. Gaussian graphical models for VAR models.
2. Gaussian graphical models for general time series and its connection to infinite dimensional matrices.
3. Gaussian graphical models for stationary time series and spectral analysis.
4. Estimation of graphical models for both low dimension and high dimensional time series.

Part 4: Graphical models for non-Gaussian time series
1. The exponential family of distributions.
2. The connection between the exponential family and graphical models.
3. Conditionally specified distributions for graphical modelling of multivariate time series.

Learn smart slidewriting with your own examples:

In the theoretical part of the training, you will learn techniques for structuring complex information logically and for designing easy-to-understand slides and stories. In the practical part, you will then apply these techniques to your very own research findings and produce your own presentation. By the end of the training, you will have a coherent storyline of your own academic presentation as well as drafts of most of your slides.

Important:

This training focuses on the logic of good presentations; it does not cover the use of presentation software like PowerPoint.

Receive individual feedback:

Prior to the training, you can send an already existing presentation to the trainer at markus.burger@slidewriting.com and receive individual feedback.

Prerequisites:

Basic Python knowledge and knowledge about data processing, ML models and training of models is required.

Summary:

The AI revolution is moving even more rapidly than the digital revolution and leads to the emergence of completely new tools and technologies that affect the scientific process. In this course, we will learn about data-based research software, tools and communities that are relevant in creating and sharing such software, and about best practices in training machine-learning models. Research software that is based on ML models requires an additional layer of best practices in the implementation, including testing of non-deterministic processes. Security aspects as well as bad examples are discussed to highlight the importance of adhering to a best practices code of conduct.

Learning Objectives:

After the course participants will

- Understand and follow best practices in the underlying dataset
- Understand and follow best practices in training ML models
- Write better data-based research software, including appropriate tests
- Avoid negative impact from legal and security issues

From blogging to news and opinion articles, writing remains one of the best ways for scientists to reach the public. While there are many styles and formats for popular science writing, certain elements of good writing are required. In this workshop, we discuss the various formats of writing as well as why and when each is used. We will also discuss how to tell a good story and common bad habits scientists have when writing for the public. Participants will practice writing short pieces based on their own work or interests that will be shared and discussed during the workshop.

All participants need a microphone and ideally a webcam.

Hardware integrated into the laptop is sufficient.

Topics (selection)

- Formats (blogs, news, opinion, etc.)
- Writing styles
- Elements of a good story
- Story structure

Common mistakes (passive voice, jargon, etc.)

Prerequisites

No prior knowledge by the participants is necessary to participate in this course and it is intended for all scientific audiences. Participants are required to bring their own laptops to work on during the course. Network access (e.g. through Eduroam) is recommended.

Summary

Jupyter notebooks are a great tool for exploring and interacting with data using the Python programming language and its rich ecosystem of libraries. In this course we will cover basic usage of the Pandas library to download a dataset, explore its contents, clean up missing or invalid data, filter the data according to different criteria, and plot visualizations of the data.

Learning Objectives

After the course participants will be able to use a Jupyter notebook to

- Explore a dataset
- Filter the dataset
- Clean up missing / invalid entries
- Generate plots and visualizations of the data

Prerequisites

Participants should have a basic understanding of the Unix Shell e.g. be able to execute commands and edit files. The target audience are scientists that develop software (no restrictions on software scale or programming language) as a side product of their research and host this software on GitHub.

Summary

Continuous Integration (CI) is one of the cornerstones of agile development processes: Before changes are included into the mainline, a number of tests is run automatically to ensure the quality of the software. In this course, we explain how this process is implemented on GitHub.com (GitHub Actions). After a general introduction, participants will work on setting up Github Actions for their own projects. The SSC is available for follow-up consultation work on the CI workflows developed during this course.

Learning Objectives

After this course the participants

- Understand the basics of running CI workflows on GitHub.com
- Know the most important configuration options in GitHub Actions
- Will have CI running for their own research software project!

Prerequisites

Participants should have a basic understanding of the Unix Shell e.g. be able to execute commands and edit files. Participants are required to bring their own laptops to work on during the course. Root privileges and network access (e.g. through Eduroam) on this computer are required. Instructions on how to install Docker on the participants computer will be sent out before the course.

Summary

Container technologies (e.g. Docker containers) have emerged as a fundamental tool of the cloud computing era. In scientific applications, containerization is used to encapsulate the complex execution environment of research software with a number of goals in mind: Setting up user landscapes for Continuous Integration testing, ensuring reproducibility of execution environments and packaging code to run on an HPC system. The workshop involves live coding sessions where participants exercise the learned commands on their own computers.

Learning Objectives

After the course participants will:

- Understand the basic terminology of containerization
- Know where to find and reuse ready-to-use containers
- Know how to create containers for their daily work on their own
- Have built and run a parallel application within a Singularity container

To study the nonlinear properties of complex natural phenomena, the evolution of the quantity of interest can be often represented by systems of coupled nonlinear stochastic differential equations (SDEs). These SDEs typically contain several parameters which have to be chosen carefully to match the experimental data and to validate the effectiveness of the model. In the present paper the calibration of these parameters is described by nonlinear SDE-constrained optimization problems. In the optimize-before-discretize setting a rigorous analysis is carried out to ensure the existence of optimal solutions and to derive necessary first-order optimality conditions. For the numerical solution a Monte-Carlo method is applied using parallelization strategies to compensate for the high computational time. In the numerical examples an Ornstein-Uhlenbeck and a stochastic Prandtl-Tomlinson bath model are considered.

This is a joint work with Jan Bartsch and Robert Denk. The associated paper will appear in the Journal Applied Mathematics & Optimization.

14:00 Denis Zavadski (Supervisor C. Rother)
"Controllable Image Synthesis with Text-to-Image Diffusion Models"

14:20 Damjan Kalsan (Supervisor C. Rother)
"Controlled Generation of Labelled Synthetic Data"

14:40 Hannah Rickmann (Supervisor R. Herzog)
"Gaussian Process Regression Models for Predicting Fatigue Strength from Process and Material Data in Machining"

15:00 Short Coffee Break

15:20 Purusharth Saxena (Supervisor B. Velten)
"Optimal Transport and Causal Inference in Gene Regulatory Networks"

15:40 Wangjun Hu (Supervisor B. Velten)
"Generative latent variable models for spatial transcriptomics data"

16:00 Christian Alber (Supervisor R. Scheichl)
"Generalized Finite Elements and Localized Model Order Reduction"

Join us to take your scientific writing to the next level using AI assistants! Learn how to harness AI to objectify literature review, brainstorming, data visualization, intuitive graphical abstracts, crafting compelling scientific narratives and easy-to-read text. Our interactive sessions will equip you with the skills to effectively communicate your data's story.

Prerequisites:

No prior knowledge by the participants is necessary to participate in this course and it is intended for all scientific audiences. Participants are required to bring their own laptops to work on during the course. Network access (e.g. through Eduroam) is recommended.

Summary:

The Unix shell is a powerful tool that allows people to do complex things with just a few keystrokes. More importantly, it helps them combine existing programs in new ways and automate repetitive tasks so they aren’t typing the same things over and over again. Use of the shell is fundamental to using a wide range of other powerful tools and computing resources. The course will include hands-on live coding sessions where participants exercise the learned commands on their own computers.

Version control is the lab notebook of the digital world: it is used to keep track of what was done and to collaborate with other people. Its use is the state of the art in software development projects of all scales. However, it is not limited to software: books, papers, small data sets, and anything that changes over time or needs to be shared can and should be stored in a version control system. The course will include hands-on live coding sessions where participants exercise the learned commands on their own computers.

Learning Objectives:

- Have a fundamental understanding of how and why to use the Unix Shell
- Be comfortable with handling files and directories using the command line
- Have experience with advanced usage of the shell e.g. loops, pipes, redirects etc.
- Know how to write their workflows as reusable shell scripts
- Understand the benefits of using version control
- Understand basic git terminology
- Have a good working knowledge of common tasks in Git
- See how Git repositories can help them to move towards practicing Open Science

In this talk, I will introduce Cut Finite Element Methods (CutFEM) for interface problems and present recent developments, such as mass conservation, within this class of unfitted finite element techniques. Finite Element Methods (FEM) are widely used to approximate solutions to partial differential equations (PDEs) in complex geometries, but they typically require conforming computational meshes. The goal of CutFEM is to allow interfaces or boundaries to intersect the computational mesh arbitrarily while preserving desirable properties  of standard FEM. I will explain how CutFEM maintains accuracy and avoids ill-conditioning of linear systems, even with unfitted meshes. I will begin with stationary problems, and then present our approach for discretizing time-dependent PDEs in evolving domains, and demonstrate its application to simulations of surfactant dynamics in incompressible two-phase flow systems.

Prerequisites:

Basic Python knowledge is required. Participants need a laptop/PC with Visual Studio Code installed and a working Python environment. Participants need to have access to GitHub copilot (either through free trial, individual license such as GitHub student (free), or other form of license).

Summary:

Generative AI is emerging as a major creative force that supports humans in content creation. Specifically trained models can support software developers with their software projects and lead to time savings and a shift in what aspects of generating software are more important on a day-to-day basis. In this course, we will learn about GitHub Copilot, the differences to ChatGPT, and how to set up and use GitHub Copilot in coding projects. Best practices in using Copilot, as well as recommendations how to use it efficiently and safely will be introduced.

Learning Objectives:

After the course participants will

- Know what GitHub Copilot is and what is happening under the hood
- Use Copilot in an IDE
- Improve Copilot's suggestions through providing appropriate context
- Know when and when not to use Copilot
- Know about privacy and security when using GitHub copilot

Prerequisites:

Basic Python knowledge is required.

Summary:

Python has rapidly advanced to the most popular programming language in science and research. From data analysis to simulation and preparation of publications, all can be done in Python with appropriate libraries and implementing own modules. We will discuss Python Enhancement Proposals (PEP) and how these can help you write cleaner code. Common pitfalls in Python will be explained with examples. We will demonstrate typical “bad programming” and how to code the examples in a more pythonic way.

Learning Objectives:

After the course participants will

- Understand the basic PEP recommendations
- Use a linter and code formatter to ensure following of the guidelines
- Write better=more readable code
- Avoid bugs through best practices for example in passing keyword arguments

14:30 Lina Deschamps (Supervisor P. Albers)
“On a hyperkähler view of the magnetic geodesic flow on CP^n“

14:50 Hans Reimann (Supervisor C. Strauch)
“Developing and Incorporating Strategies for Data-Driven Control and Data Assimilation”

15:10 Margot Chazotte (Supervisor D. Schapiro)
“Spatial imaging in the gastrointestinal tract”

15:30 Jean Radig (Supervisor C. Herrmann)
"AI for patient-specific cancer cure"

15:50 Daniel Galperin (Supervisor U. Köthe)
"Invertierbare Neuronale Netze für ein verbessertes Verständnis von Infektionskrankheiten"

Prerequisites:

Basic Python knowledge and a laptop is required. Experience with writing tests is not required.

Summary:

An automated test suite makes it much easier to maintain, extend and debug your Python code. In this course we will learn how to write automated tests in Python using the pytest library. After introducing the key concepts, the majority of the course will be hands-on, writing and running tests.
Learning Objectives

After the course participants should be able to:

- Install and run pytest
- Write simple tests
- Use temporary files in tests
- Use fixtures to manage resources
- Parametrize tests
- Add an automated test suite to their existing python projects

Everyone's talking about digitalization. But before you can talk about innovation, you have to adopt these technologies. In order to understand which technologies will have a real impact on a company's product portfolio, technologies must be analyzed and understood. Only then can they be scaled into business processes. At Merck, the interface between business and technology is critical. A few examples illustrate how innovation can be achieved through digital technologies such as AI and computational science.

Dr. Philipp Harbach is a dedicated digital innovator, currently serving as the Global Head of Group Digital Innovation at Merck KGaA, Darmstadt, Germany. Alongside his diverse team of professionals, he focuses on driving digital innovation using modern data science and computational modeling methods. Prior to his current role, Dr. Harbach served as the Head of Digital Chemistry, where he explored the development of cutting-edge digital technologies and products for chemistry-related use cases.
He first joined Merck in 2014 as a Laboratory Head, where he established a foundation for computational chemistry within the Electronics business, specifically in the field of organic light-emitting diodes (OLEDs).
Dr. Harbach earned his PhD in theoretical chemistry from the Interdisciplinary Center for Scientific Computing (IWR) at Heidelberg University in 2013. His research focused on developing and applying advanced theoretical methods for
molecular systems in photochemistry, which resulted in several international publications and citations.

Mit dem Fokus auf den Bereich Datenanalyse vermittelt dieser Kurs die Grundlagen in der Programmiersprache R. Von der Installation, den wesentlichen Objekttypen und Programmierbasics über einfache Analysen und Visualisierungen erhalten die Teilnehmer:innen in diesem Training alle Voraussetzungen, um erste Anwendungen in R zu schreiben.
Die Teilnehmer:innen sollen bereit sein, während des Kurses ihre Kamera einzuschalten und ihren Bildschirm zu teilen.

This hands-on workshop provides an overview of the German research funding landscape and explains how to exploit funding opportunities at the postdoctoral stage for further career advancement. Selected EU programmes within the framework of Horizon Europe are also covered. Practice scenarios for different research settings help to understand how to select and approach individual funding programmes to establish yourself as a principal investigator and build your research capacity.

Moderation:

Prof. Dr. Barbara Paech, Software Engineering, Universität Heidelberg

SprecherInnen:

Quereinstieg
o Kerstin Lambert, Digitalisierungsmanagerin, EPPLE GmbH
o Eda Özsahin, IT-Projektkeiterin
o Claire Zschiesche, Ausbildungsleitung Informatik, SAP SE (Panel)
o Sibylle Hofmeyer, PR-Beraterin, Hofmeyer PR

MINT-Studium, IT-Firma
o Franziska Bierer, Vorstand, andrena objects ag (Panel)
o Dominique Cheray, Projektleiterin und Softwareentwicklerin HMS Analytical Software GmbH

Lehramt
o Lisa Galvagno, Studienrätin und Studierende im Erweiterungsfachmaster Informatik (Geographie, Politik, Wirtschaft, Informatik) (Panel)
o Prof. Dr. Claudia Hildebrandt, CZS Stiftungsprofessur für Informatik und ihre Didaktik, Pädagogische Hochschule Heidelberg, Heidelberg School of Education

Firmengründung
o Ines Korkowski, Gründerin und Geschäftsführerin KRAFTSTOFF Fashion GmbH (Panel)
o Dr. Monika Pobiruchin, Hochschule Heilbronn und Colorful Bytes GmbH

Wissenschaft
o Dr. Claudia Nass, Expert Cross-Disciplinary Requirements Engineering, Fraunhofer Institut (Panel)
o Prof. Dr. Barbara Paech und Prof. Dr. Claudia Hildebrandt

Course description
Producing certified, reliable code is crucial in software engineering. Taking it a step further, modern programming languages with strong type-checking capabilities can even be used to state theorems about abstract mathematical objects and verify the correctness of proofs.

In this workshop, we will explore the world of proof assistants and learn how to explain mathematics to a computer using Lean. The goal is to help participants develop new coding skills while becoming acquainted with formal methods on a larger scale.

Contents of the workshop
The course will consist of four 2-hour sessions, spread over two days. Each session will combine a lecture with hands-on computer practice. Please bring your laptop.

- Lecture 1: Basic Lean syntax
- Lecture 2: Tactics
- Lecture 3: Dependent types
- Lecture 4: Group project

Participants will be asked to work in pairs. The instructor will provide practice files for each session and no prior installation of Lean is required.

You will learn to visually communicate your complex research ideas and results so your messages are effortlessly understood by any specific audience (scientists or non scientists). We will not focus on aesthetics but on how understanding human visual perception can inform your design decision for better comprehension of your scientific images, posters, and slides. You will also design a graphical abstract of your research, discuss it with peer scientists in a group exercise, and get actionable advice and feedback on your own materials. It is an immersive workshop, comprehensive, structured, memorable, easy to follow, useful and fun. More at https://www.seyens.com

Generative machine learning is offering new ways of tackling complex scientific problems. I will describe recent efforts in the context of multiscale molecular simulations of soft-matter systems: a combination of free-energy calculations, representation learning, and Bayesian optimization to design small molecules; a backmapping strategy that learns the atomistic distribution of atoms conditional on the coarse-grained degrees of freedom; and finally a denoising diffusion model to efficiently compute free energies.

In the academic world, clear, concise, and well-written texts play an important role in convincing journal editors and conference organisers to accept a paper for review and publication, or to invite a researcher to present at a conference. This workshop provides PhD students with strategies to write short texts efficiently and effectively. It enables participants to organise ideas and structure texts effectively, to present their own and other researchers‘ findings and opinions appropriately, and to use correct terminology and vocabulary.

Academic posters are a useful communication method to present your PhD project, engage people in fruitful discussions and gain new input for your research. To suit that purpose, a poster needs to be carefully designed, both from the contents' side and the visual appearance. Starting with the content, we will extract the key messages from your projects. All the way from the research question at hand, to the results and finally to the discussion and open questions. Keeping the target audience in mind, we will determine what should be on the poster and how we can communicate it efficiently. With the aim in mind to reduce text to an absolute minimum, we will focus next on the visuals that should go on the poster. These can include photographs, maps, schematic drawings, diagrams or other data visualizations. Having determined the contents, we will address the graphic design both of the individual visuals as well as of the poster as a whole. You will learn how to use a grid and white space, choose colors, fonts and font sizes etc. to produce an aesthetically pleasing and effective poster.

During the workshop each participant will design his / her poster about his / her own PhD project. You need to bring text and figures (diagrams, charts, schematics, photos etc.) describing your project. Poster drafts are welcome as well.

14:30 Johann-Friedrich Feiden (Supervisor C. Rother)
“Improving Monocular Depth Estimation and Robustness Testing”

14:50 Sebastian Pauly (Supervisor A Dreuw)
“Computational Investigations and Design of Molecular Solar Thermal Systems“

15:10 Xingjian Qi (Supervisor R. Strzodka)
“Parallel computing in low precision”

15:30 Théo André (Supervisor A. Marciniak-Czochra)
"The role of a second diffusive component in reaction-diffusion-ODE systems"

15:50 Zehua Zhang (Supervisor B. Velten)
“Modeling Drosophila Muscle Development from Multi-omics Data: A Gene Regulatory Perspective”

14:30 Sebastian Stricker (Supervisor B. Savchynskyy)
“Unsupervised learning of multi-graph matching algorithms with applications in bioimaging”

14:50 Wilfredo Colmenares (Supervisor R. Strzodka)
“Parallel Krylov Subspace Methods“

15:10 Francisca Vieira (Supervisor B.Velten)
“Modelling perturbation responses along a time course”

15:30 Lukas Hatscher (Supervisor Denis Schapiro)
"Spatial Omics for improving stratification of lung cancer and multiple myeloma
Patients"

15:50 Krešimir Beštak (Supervisor D. Schapiro)
“Computational methods for spatial omic analysis of myocardial tissue in the context of cardiovascular diseases”

Collecting, processing and analyzing data are central activities for virtually every researcher. Topics like data sharing and data publication are becoming increasingly important. Nevertheless, many research projects lack a structured and well-organized data management. This course is meant to give a general, discipline-independent introduction into various topics central to an efficient management of research data with a special focus on questions related to data archiving and data sharing. Both are central aspects of good scientific practice. Archiving and long-term preservation of research data are prerequisites for the scrutiny of scientific results based on the analysis of this data. Data sharing on the other hand increases transparency of research results and enables possible re-usage of data for new research questions, in combination with additional data sets and in interdisciplinary contexts.

Abstract: The overall aim of this course is to study matrix computations that lie at the core of a wide variety of applications in machine learning and data science. We will give an introduction to the mathematical theory of computational linear algebra (with derivations of the methods and some elementary proofs). This will broadly include methods for solving linear systems of equations, least-squares problems, eigenvalue problems, and other matrix decompositions. We will learn to implement the computational methods efficiently, as well as how to thoroughly test their implementations for accuracy and performance. More importantly, we will learn how to apply these methods to solve realistic machine learning application problems, including collaborative filtering (recommendation systems), image deblurring, kernel methods, and beyond.

It’s far more than “just” a presentation: defending your thesis is the last (and one of the biggest) challenges in your academic qualification – and it doesn’t only require well-prepared content. This workshop addresses aspects of the defense related to communication and presentation. We will work on participants’ body language as well as on strategies for convincing argumentative and rhetorical positioning during the disputation. A significant portion of the workshop will be devoted to simulated defenses and evaluations of the performances.

Galaxy formation and evolution is a multi-scale, multi dimensional and multi-physics problem. While using modern high-resolution telescopes can offer us high-quality insights in galaxies, deciphering the physics of galaxy formation requires cosmological simulations which track the evolution over the history of the Universe. In this talk, I will describe the various ways to compare cosmological simulations with observations of galaxies. I will present a rich dataset of observed galaxies and present novel ways to constrain galaxy formation models. I will show that modern cosmological simulations reproduce various properties of observed galaxies however, fall short in matching the dynamical properties.

"Wow, that presenter is so good in front of an audience. If only that were easier for me!" Being a good speaker is often just a question of developing a set of skills and techniques. The use of voice and body language, an effective presentation structure and the dynamic use of language require awareness and practice. The workshop helps to identify and explore these requirements, from self-reflection to self-assurance and long-term excellence.

Computational-based mass spectrometry (MS) is rapidly evolving, challenging us to process and interpret complex multi-dimensional data. There are many computational MS projects to aid in this process, but mastering or even starting to use them can be a challenge. Therefore, we want to provide a platform in Prague to meet, discuss, and learn how to move MS-based omics forward. Additionally, we want to offer a hands-on training workshop on the use of generative AI tools for scientific writing presentations, where we can learn their use, opportunities, and ethical usage.

The recent development of AI-based software to create texts, images or even videos has an increasing impact on the work of journalists. This talk will outline benefits as well as concerns and risks of this emerging technology for journalism.

14:30 Janik Fechtelpeter (Supervisor G. Koppe)
“Control of dynamical systems and recurrent neural networks with applications to clinical and behavioral psychology“

14:50 Luise Nottmeyer (Supervisor J. Rocklöv)
“Indicator Development for Environmental Antimicrobial Resistance Monitoring“

15:10 Yichao Liu (Supervisor J. Rocklöv)
“Using machine learning for citizen science surveillance of ticks and prediction of tick- borne diseases with climate change”

15:30 Denis Kiesewetter (Supervisor F. Gräter)
"Incorporating Redox Chemistry into Classical Molecular Dynamics Simulations by Machine Learning"

15:50 Letao Huang (Supervisor A. Dreuw)
“The Ground and Excited State Absorption of Triphenylamine (TPA) and Dimethylmethylene-Bridged Triphenylamine (DTPA): A Quantum-Chemical Study”

Prerequisites:

Basic Python knowledge is required. Participants are recommended to bring a laptop.

Summary:

A good test suite makes extending, maintaining and debugging a codebase both easier and faster. In this course we will look at the different kinds of tests, and understand how to write good tests. We will also cover different testing strategies, such as test-driven-design when writing new code, or acceptance testing when working with legacy code that doesn’t have a good test suite. Code samples will use the Python testing framework pytest but the concepts also apply to other frameworks and languages.

Learning Objectives:

After the course participants will

- Understand the different kinds of tests
- Understand different testing strategies
- Write better tests of their code
- Deal better with legacy code that is missing tests

In times of globally growing pressure on universities the principle of academic freedom is fiercely debated within and outside academia. What does it mean? What is covered? What is not? Is it an institutional or a personal right? Where are its limits? What are the threats?

Surprisingly few academics have well-informed knowledge what academic freedom is and what it is not. Debates and actions are often driven by emotions and subjective feelings about a “proper university“ instead of proper knowledge.

In this course we will:

I. discuss the principle of academic freedom – its legal status and its formal applicability

II. learn to discern academic freedom from other civil liberties (e.g. open speech, freedom of opinion)

III. understand the complexity of taking action in the face of violations to academic freedom

IV. sharpen our awareness for threats to academic freedom in our own societies.

The course and discussions will be in English.

We expect engagement and interest – no marks will be given!

For any queries please contact thomas.meier@zaw.uni-heidelberg.de

The consideration of uncertainty in optimization problems is an important aspect in many practical applications. In this talk we consider optimization problems with partial differential equation (PDE) constraints, where uncertainty occurs in parameters (for example material parameters, usage scenarios) as well as in the optimization variables (e.g. manufacturing tolerances). There exist several approaches to handle uncertainty in optimization, in particular stochastic optimization, probabilistic constraints and (distributionally) robust optimization.

We will focus on robust and distributionally robust optimization approaches for problems with expensive PDE-constraints. In robust optimization, relevant realizations of the uncertain variables are described by an uncertainty set and it is required that the optimization variables satisfy critical constraints for all realizations of the uncertain variables in the uncertainty set. In distributionally robust optimization one considers an ambiguity set of relevant distributions for uncertain parameters and requires that the constraints are satisfied in expectation for all distributions in the ambiguity set.

The resulting robust formulation is a semi-infinite optimization problem and can be written in a min-max form. We present various techniques how to approximate and reformulate the robust problem by using optimality conditions and duality theory in order to obtain a tractable problem. Depending on the approach, the reformulation contains conic constraints, complementarity conditions and conjugate functions. We will sketch suitable solution methods.

Moreover, we will illustrate, how adjoint techniques as well as model order reduction and error estimation can be used to solve these problems efficiently in the PDE constraint case.

We present several application examples, in particular the design optimization of electric machines under uncertainty.

For more see the abstract_file:

Machine learning can be performed in all major programming languages, but Python has established itself as a standard. In this course, you will learn the fundamentals of machine learning (data collection, cleaning, modeling and prediction) through hands-on exercises on real-world data sets. You will understand the differences between supervised and unsupervised learning and get to know common models and algorithms. Furthermore, you will be able to evaluate the performance of the trained models.

Nowadays researchers can obtain quantitative measurements of the abundances and activities of different types of molecules across experimental conditions and spatio-temporal resolutions. In this talk, I will discuss how latent variable models can help us to extract major underlying patterns from the noisy, heterogeneous and high-dimensional data and how to accommodate the statistical properties of molecular data in such models. Case studies will demonstrate how this can help us to highlight core biological processes along development, in health and disease and pinpoint underlying molecular drivers.

Mastering the craft of writing is a core aspect of developing as doctoral students. This workshop provides a comprehensive view (and review) of helpful norms, rules and skills to help doctoral students improve their writing while gaining confidence in their capacity to exercise a foundation skill. As ‘expert writers,’ participants will be encouraged not just to write but also to align with readers’ expectations, and to avoid pretentious prose. Selected issues of grammar, vocabulary, and text structure are discussed in service of these topics.

Objectives:

- Identify and fix problems in your own long-form texts with respect to paragraph writing, coherence and unity;
- Make your writing more impactful and valuable;
- Address problems relating to grammar and (academic) vocabulary;
- Effectively proofread your own texts and incorporate feedback from your peers, teacher and doctoral advisor into your own text.

From blogging to news and opinion articles, writing remains one of the best ways for scientists to reach the public. While there are many styles and formats for popular science writing, certain elements of good writing are required. In this workshop, we discuss the various formats of writing as well as why and when each is used. We will also discuss how to tell a good story and common bad habits scientists have when writing for the public. Participants will practice writing short pieces based on their own work or interests that will be shared and discussed during the workshop.

Prerequisites:

No prior knowledge by the participants is necessary to participate in this course and it is intended for all scientific audiences. Participants are required to bring their own laptops to work on during the course. Network access (e.g. through Eduroam) is recommended.

Summary:

Jupyter notebooks are a great tool for exploring and interacting with data using the Python programming language and its rich ecosystem of libraries. In this course we will cover basic usage of the Pandas library to download a dataset, explore its contents, clean up missing or invalid data, filter the data according to different criteria, and plot visualizations of the data.

Learning Objectives:

After the course participants will

- Explore a dataset
- Filter the dataset
- Clean up missing / invalid entries
- Generate plots and visualizations of the data

Modelle bilden die Grundlage für quantitative Entscheidungen: Analyse, Verbesserung und Optimierung basieren auf mathematischen Beschreibungen unserer Umwelt. Der Zusammenhang zwischen Optimierung und Nachhaltigkeit zeigt sich deutlich in der effizienten Nutzung von Ressourcen und der Reduktion von Umweltbelastungen, wodurch langfristige und nachhaltige Lösungen entwickelt werden können. Beim Modellierungstag stellen Redner:innen der Universität und ihrer Partnerinstitute und -unternehmen Forschung mit Bezug zur Energieversorgung, Logistik und Hitzestressvermeidung vor.

Auf dieser Netzwerkveranstaltung der HGS MathComp Graduate School treffen Expert:innen und Wissenschaftler:innen zusammen und beleuchten gemeinsam Trends in der modellgestützten Forschung und Entwicklung aus unterschiedlichen Blickwinkeln. In diesem Erfahrungsaustausch liegt das Potenzial für zukünftige Zusammenarbeit zwischen Universität und Wirtschaft für den direkten Transfer von Forschungsergebnissen zur Prozessoptimierung und zum intelligenten Ressourceneinsatz.

Understanding and predicting maladaptive human behavior and cognition, which are pertinent to many psychiatric disorders, requires understanding their underlying dynamics and dynamical mechanisms. Modifying these behaviors requires control over their dynamics.

My group focuses on inferring the dynamics underlying human behavioral and neural time series, utilizing both process-driven and data-driven modeling approaches. Based on these methods, we develop mental health applications where dynamics are used to predict and control future system states, or to study their generative mechanisms. Our applications include: 1) Mobile apps that assess psychological ratings as proxies for mental health states, forecast changes over time, and use these forecasts to tailor mobile mental health exercises. 2) Web-based platforms in which we specifically tailor cognitive experimental paradigms to reliably assess complex decision-making. And 3) we develop and apply models for the reliable detection of neural dynamical systems features predictive of psychiatric dysfunction.

In this colloquium, I want to introduce to you the new RG and its members, along with current and future research directions.

Prerequisites:

Basic Python knowledge and knowledge about data processing, ML models and training of models is required.

Summary:

The AI revolution is moving even more rapidly than the digital revolution and leads to the emergence of completely new tools and technologies that affect the scientific process. In this course, we will learn about data-based research software, tools and communities that are relevant in creating and sharing such software, and about best practices in training machine-learning models. Research software that is based on ML models requires an additional layer of best practices in the implementation, including testing of non-deterministic processes. Security aspects as well as bad examples are discussed to highlight the importance of adhering to a best practices code of conduct.

Learning Objectives:

After the course participants will

- Understand and follow best practices in the underlying dataset
- Understand and follow best practices in training ML models
- Write better data-based research software, including appropriate tests
- Avoid negative impact from legal and security issues

Prerequisites:

Basic Python knowledge is required. Participants need a laptop/PC with Visual Studio Code installed and a working Python environment. Participants need to have access to GitHub copilot (either through free trial, individual license such as GitHub student (free), or other form of license).

Summary:

Generative AI is emerging as a major creative force that supports humans in content creation. Specifically trained models can support software developers with their software projects and lead to time savings and a shift in what aspects of generating software are more important on a day-to-day basis. In this course, we will learn about GitHub Copilot, the differences to ChatGPT, and how to set up and use GitHub Copilot in coding projects. Best practices in using Copilot, as well as recommendations how to use it efficiently and safely will be introduced.

Learning Objectives:

After the course participants will

- Know what GitHub Copilot is and what is happening under the hood
- Use Copilot in an IDE
- Improve Copilot's suggestions through providing appropriate context
- Know when and when not to use Copilot
- Know about privacy and security when using GitHub copilot

It’s far more than “just” a presentation: defending your thesis is the last (and one of the biggest) challenges in your academic qualification – and it doesn’t only require well-prepared content. This workshop addresses aspects of the defense related to communication and presentation. We will work on participants’ body language as well as on strategies for convincing argumentative and rhetorical positioning during the disputation. A significant portion of the workshop will be devoted to simulated defenses and evaluations of the performances.

Prerequisites:

Basic Python knowledge is required.

Summary:

Python has rapidly advanced to the most popular programming language in science and research. From data analysis to simulation and preparation of publications, all can be done in Python with appropriate libraries and implementing own modules. We will discuss Python Enhancement Proposals (PEP) and how these can help you write cleaner code. Common pitfalls in Python will be explained with examples. We will demonstrate typical “bad programming” and how to code the examples in a more pythonic way.

Learning Objectives:

After the course participants will

- Understand the basic PEP recommendations
- Use a linter and code formatter to ensure following of the guidelines
- Write better=more readable code
- Avoid bugs through best practices for example in passing keyword arguments

Syllabus:

- Motivation: coherent structures in PDE
- Stability theory for non-linear ODEs
- Asymptotic stability in dissipative systems
- Essential and point spectrum of a linear operator
- Semigroups and exponential stability
- Non-linear estimates and orbital stability

Prerequisites:

Prior knowledge about version control with Git is required. This can e.g. be acquired by attending the SSC’s introductory course “Version Control with Git” and “Intermediate Topics in Version Control with Git”. This course is intended for researchers that develop their own research software. Participants are required to bring their own laptops to work on during the course. Network access (e.g. through Eduroam) is recommended.

Summary:

Applying version control can be a game changer for a collaborative research software project. However, even in projects that already successfully use Git, there is often room for improvement of the employed Git workflows. This workshop explores some advanced topics that increase the participants’ repertoire of Git workflows: Bisection, Worktrees, Submodules, Large File Storage and Pre-commit Hooks. Additionally, we showcase invasive methods of modifying a Git repository via Rebase.

Learning Objectives:

After the course participants will

- Know about advanced Git features like submodules, large file storage
- Be able to bisect regression in a Git repository
- Understand the advantages of worktrees
- Be able to apply pre-commit hooks for their Git repositories
- Have used Rebase to change a repository’s Git history

Prerequisites:

Prior knowledge about version control with Git is required. This can e.g. be acquired by attending the SSC’s introductory course “Version Control with Git”. This course is intended for all scientific audiences. Participants are required to bring their own laptops to work on during the course. Network access (e.g. through Eduroam) is recommended.

Summary:

Applying version control can be a game changer for a collaborative research software project. However, even in projects that already successfully use Git, there is often room for improvement of the employed Git workflows. This workshop tries to delve into collaborative workflows using Git branches and discuss their strengths. Special emphasis is given to the resolution of merge conflicts that may be required in the process of dealing with diverging branches.

Learning Objectives:

After the course participants will

- Understand git branches and typical workflows using them
- Have seen collaborative workflows on GitHub (Pull Requests, Forks etc.)
- Gained first experience in resolving merge conflicts
- Know how to use rebase to clean up a branch’s history

Prerequisites:

No prior knowledge by the participants is necessary to participate in this course and it is intended for all scientific audiences. Participants are required to bring their own laptops to work on during the course. Network access (e.g. through Eduroam) is recommended.

Summary:

The Unix shell is a powerful tool that allows people to do complex things with just a few keystrokes. More importantly, it helps them combine existing programs in new ways and automate repetitive tasks so they aren’t typing the same things over and over again. Use of the shell is fundamental to using a wide range of other powerful tools and computing resources. The course will include hands-on live coding sessions where participants exercise the learned commands on their own computers.

Version control is the lab notebook of the digital world: it is used to keep track of what was done and to collaborate with other people. Its use is the state of the art in software development projects of all scales. However, it is not limited to software: books, papers, small data sets, and anything that changes over time or needs to be shared can and should be stored in a version control system. The course will include hands-on live coding sessions where participants exercise the learned commands on their own computers.

Learning Objectives:

- Have a fundamental understanding of how and why to use the Unix Shell
- Be comfortable with handling files and directories using the command line
- Have experience with advanced usage of the shell e.g. loops, pipes, redirects etc.
- Know how to write their workflows as reusable shell scripts
- Understand the benefits of using version control
- Understand basic git terminology
- Have a good working knowledge of common tasks in Git
- See how Git repositories can help them to move towards practicing Open Science

Computational modeling of the cardiovascular system can help to understand the relevant mechanisms and help to tailor treatments for heart diseases. We introduce a fully coupled PDE system for electrophysiology and cardiac mechanics combined with ODEs of ionic concentrations, the blood pressure in the cardiac chambers and a model for the circulatory system.

The model is realized in the parallel software system M++, and we show that the full model for a heart beat is scaling very well on modern HPC systems.

10:30 - Robin Fallegger (Supervisor J. Saez-Rodriguez)
“Meta-analysis of chronic kidney disease single cell atlases“

10:50 - Chiara Schiller (Supervisor D. Schapiro)
“Profiling intrahepatic cholangiocarcinoma samples to leverage spatial information into clinical Applicability“

11:10 - Philipp Schäfer (Supervisor J. Saez-Rodriguez)
“Dissecting immune system dysfunction with spatial omics data”

11:30 - 10-minute break

11:40 - Bárbara Zita Peters Couto (Supervisor J. Saez-Rodriguez)
“On Targeted Therapy Discovery to Overcome Chemoresistance in Ovarian Cancer“

12:00 - Leonie Küchenhoff (Supervisor J. Saez-Rodriguez)
“Fibrotic disease signatures across organs”

12:20 - Miguel Ángel Hernández Hernández (Supervisor J. Saez-Rodriguez)
“Multi-omic integration and analysis of IgA nephropathy disease mechanisms”

12:40 - 80-minute break / 12:45 - 13:30 HGS MathComp Mixer

14:00 - Marc Ickler (Supervisor F. Hamprecht)
“Machine Learning for Molecular Physics”

14:20 - Carolin Lindow (Supervisor A. Marciniak-Czochra)
“Mathematical modelling and model-based data analysis of structured stem cell systems”

14:40 - 10-minute break

14:50 - Albert Li (Supervisor C. Herrmann)
“Development of deep learning models to predict context-specific response to interferon treatment”

15:10 - Fariha Mahzabin Annesha (Supervisor J. Wittbrodt)
“Correlative Super-Resolution light and X-ray Microscopy to probe genome organisation in Cells”

Lab presentations:
Tobias Buck, Holger Fröning, Denis Schapiro

Science talks:
• William Oliver (Buck lab): Machine Learning for cosmological simulations
• Hendrik Borras (Fröning lab): Probabilistic Photonic Computing with Chaotic Light
• Miguel A. Ibarra Arellano (Schapiro lab): Finding broken cells with AI and computer vision

14:30 - Charles Hatfield (Supervisor J. Rockloev)
“Modeling of urban microclimate and mobility effects on mosquito abundance and vectorial competence”

14:50 - Hannah Weiser (Supervisor B. Höfle)
“It all starts with data – Improving machine learning-based point cloud algorithms for forest applications with synthetic training data generated by virtual laser scanning”

15:10 - Ronald Tabernig (Supervisor B. Höfle)
„Simulating laser scanning of dynamic virtual 3D scenes for improved 4D point cloud-based topographic change analysis“

15:30 - 10-minute break

15:40 - Niklas Reinhardt (Supervisor J. Zech)
“Statistical Operator Learning”

16:00 - Marco Hübner (Supervisor L Maier-Hein)
“Spectral Image Synthesis for Training Intelligent Surgical Systems”

16:20 - Julian Niederer (Supervisor E. Kostina)
"Numerical Methods for Nonlinear Model Predictive Control for Control Problems with Blocks of Vanishing Constraints”

Syllabus:

- Introduction: mathematical models and inverse problems.
- Basic topics in probability theory and functional analysis.
- Inverse problems, Bayesian inference and uncertainty quantification.
- Existence and consistency (stability) of posterior distributions.
- Exploring the posterior and sampling methods.
- Stability estimates for linear PDEs and physically informed models.

- Day 1: 2 x 90 minutes sessions
- Day 2 to 4: 1 90 minutes session + 1 90 minutes practice session

This course introduces the general-purpose programming language Python, which is used by web developers, data scientists and machine learning experts. Understanding the basics of Python will allow you to grasp the concepts of tools you might encounter and quickly apply them to your own research.

Within the scope of this course, you will master the Python philosophy, syntax, and writing your own scripts and modules. In addition, you will use your newly acquired skills to perform hands-on exercises and learn to conduct reproducible in-silico research. After completing this course, you will be ready to start your own Python journey and delve deeper into the world of Data Science.

Target group

This workshop is designed for doctoral candidates with previous presentation experience.

Objectives

"Wow, that presenter is so good in front of an audience. If only that were easier for me!" Being a good speaker is often just a question of developing a set of skills and techniques. The use of voice and body language, an effective presentation structure and the dynamic use of language require awareness and practice. The workshop helps to identify and explore these requirements, from self-reflection to self-assurance and long-term excellence.

Description

This seminar provides participants the opportunity to improve their conference presentation skills. Constructive feedback from the trainer and group members give the speaker a healthy amount of input while practicing new ideas and techniques to enhance the quality of their speech and overall impact of the talk.

Participants will be required to prepare a 3 to 5 minute overview of their work; the use of slides is optional. This will provide a basis for applying the practical aims of the workshop.

Throughout the two-day workshop, participants will be guided through interactive exercises to improve non-verbal communication, improve the ability to listen and react generously, and to integrate focusing techniques, which empower the speaker. Attention will also be given to structural and language aspects to improve clarity and flow of the talk.

Contents in Brief

- Effectively introducing yourself
- Engaging the audience in your talk
- Affirming the strengths and individual style of the speaker
- Improving body language and vocal quality
- Structuring your talk
- Constructive tactics for dealing with nervousness
- Dealing with challenging questions (Q&A sessions)
- Networking at conferences

Methods

- Voice and body techniques
- Partner work
- Language practice and analysis
- Interactive activities with online tools
- Videotaping and feedback sessions

Observations of the cosmos on the largest visible scales have revealed it to be amazingly simple. Likewise, the Large Hadron Collider has found no new particles on the tiniest scales probed. In contrast, the most popular theoretical paradigms including string theory and inflation, have grown increasingly complex and contrived. We have therefore launched a search for more economical and predictive explanations. In the process, we stumbled upon 1) the simplest-yet explanation for the cosmic dark matter, soon to be tested by galaxy surveys, 2) a thermodynamic explanation for the large scale geometry of the cosmos, based on an elegant new calculation of gravitational entropy, 3) an explanation of the big bang singularity as a kind of “mirror” which enforces CPT symmetry (the deepest known symmetry of nature, linking particles to antiparticles), and 4) a new explanation of the primordial density variations which led to the formation of galaxies, explaining them quantitatively in terms of known physics. I’ll outline the new theory’s main challenges and opportunities, ranging from understanding the Higgs boson to consistently coupling the quantum fields of the Standard Model to gravity, as well as future observational tests.

On the internet, video formats have long conquered the world. YouTube has long since overtaken television, especially among the younger generation, and many people move naturally in the world of short videos. In science, however, a large part of the community still shies away from using this medium.

In this workshop we will discuss the basics of (short) video production from the idea to the final upload. More importantly, we will shoot our first videos and gain our own experience with the medium of science video. In particular, we will also address the type of equipment needed and options for each budget.

Topics (selection)

- Equipment for video production
- Basics of video storytelling
- Video production from start to finish
- Best practice in recording
- Introduction to video editing software
- Upload to popular platforms

The possibility to record video formats should be given. A smartphone is sufficient for this purpose. Essential video equipment (light, audio and video hardware) will be provided.

In the academic world, clear, concise, and well-written texts play an important role in convincing journal editors and conference organisers to accept a paper for review and publication, or to invite a researcher to present at a conference. This workshop supports PhD students and novice researchers who need to produce scientific papers in English. Providing participants with strategies to write short texts efficiently and effectively, it also focuses on standards of professional papers aimed at Anglo-Saxon or international research communities. It enables participants to organise ideas and structure texts effectively, to present their own and other researchers‘ findings and opinions appropriately, and to use correct terminology and vocabulary.

During the workshop, participants work with their own texts as well as with examples from their own disciplines that they bring along and consider to be particularly well written. They discuss features of good scientific papers and are equipped to use adequate language in different genres and for different audiences. In addition, they receive peer feedback on their own drafts. All exercises empower them to produce clearer, and more correct, concise, and reader-oriented papers.

The two-day workshop covers the following topics:

- taking inventory: participants‘ strengths and challenges in writing scientific papers in English
- a brief introduction to research and writing processes
- using text analysis to become a better writer
- reporting findings, ideas, and opinions professionally and adequately
- making yourself understood: principles of clear and concise writing
- structuring ideas, organising texts: transitions, connectives, & co.
- working effectively with co-authors and constructive text feedback
- useful online and offline resources

We are currently witnessing the second quantum revolution and with it the advent of quantum technological devices for information processing purposes. Understanding these systems and their capabilities as well as developing robust algorithms for them, requires a fundamental understanding of complex quantum many-body systems as well as ways to characterize their properties efficiently. Within this course, we will help the participants to obtain proficiency in all of these aspects of quantum theory. A tentative list of topics includes:

- Entanglement/Non-Locality
Lecture 1: States, Measurement, Tensor Product, Maximally Entangled State
Lecture 2: Tensor Product (contd.), Density Matrices, Reduced States, Entropy, Entanglement Entropy.
Lecture 3: Bell-inequalities, Bell’s Theorem, Non-local Games/Protocols.

- Quantum Spin-System
Lecture 1: Spin Hamiltonians, Spin-½ Heisenberg Model, SU(2), Ground States, Ground State Energy
Lecture 2: Free Energy, Thermodynamic Limit, Thermal State
Lecture 3: Symmetry Breaking, Mermin-Wagner
Lecture 4: AKLT, Matrix Product States

For each of these topics, we will provide introductory lectures to get students acquainted with the topics before entering into the in-person phase, where experts and practitioners of these fields will give further insight.

Learning Outcome:

The goal is to provide the students with a blended learning approach to the course content on the mathematics of quantum theory with special emphasis on complex quantum many-body systems and quantum information theory. The main intended learning outcomes include an overview of the latest developments in those fields. Furthermore, the course will help the participants to develop competencies to engage in self-organized cross-university and interdisciplinary collaborations via online group work as well as to give and receive peer-feedback on results. Inviting external experts in the field will also give young researchers in the field the opportunity to grow their professional network considerably.

Podcasts offer researchers diverse opportunities to participate actively and creatively in public debates and present their research. Whether in the form of a personal interview, a background report or an entertainingly moderated show, in podcasts listeners get to know the people behind the science.

What are the current trends on the worldwide podcast scene? How do you turn a good idea into a viable concept for a podcast? What do you need to produce your own podcast? What are the dos and don’ts of planning and implementation? And how do you cut a podcast?

The online podcast seminar introduces you to the world of podcast science communication. You will get tips, learn the tricks of using podcasts for your research topic and discover how to build a podcast step by step by means of practical exercises: from the initial concept through the production phase to subsequent marketing.

In addition to work during the seminar, the online sessions will be followed by two assignments that are a constituent part of the seminar to be done at home. Please allow some extra time for them.

Thematic focus

- (Science) Podcasts: what formats and trends are there?
- Planning and drawing up your own podcast
- Writing for the ear: how do you generate images in the mind?
- Step by step to your first podcast audio clip
- Sound and technology: how do you create a rich sound without a recording studio?
- Marketing: how do I reach listeners with my podcast?

Single-cell RNA-sequencing records the expression values of thousands of genes in individual cells, making the study of cell functionality and development possible at high resolution. To understand the structure of single-cell data on a global level, topological descriptors, such as the presence of loops or voids in the data, are useful. For instance, loops might challenge the assumption of hierarchical cell development or be indicative of missing data in datasets of fully differentiated cells. The established computational tool for detecting the topology of point clouds is persistent homology. However, single-cell data is high-dimensional and notoriously noisy. We show that in high dimensionality traditional persistent homology becomes very sensitive to noise and fails to detect the correct topology. The same holds true for existing refinements of persistent homology. As a remedy, we find that spectral distances, such as diffusion distance and effective resistance, allow persistent homology to detect the correct topology even in the presence of high-dimensional noise. In particular, persistent homology with spectral distances can robustly detect the correct topology of single-cell data.
For more see the abstract_file:

Prerequisites:

Experience or interest in publishing your Python code and a laptop is required.

Summary:

In this course we will learn how to package a Python library, how to publish it on PyPI and on conda-forge, as well as look at more advanced topics like building pre-compiled wheels including c++ extensions using pybind11, and automatically publishing new releases using continuous integration and cibuildwheel.
Learning Objectives

After the course participants should be able to:

- Create a modern pyproject.toml Python package
- Publish this package to PyPI
- Set up continuous integration to automatically publish to PyPI
- Understand the basics of conda-forge publishing
- Create binary wheels including c++ pybind11 extensions

14:30: Anna Lena Schaible (Supervisor T. Buck)
“A numerical investigation on galactic outflows using galaxy simulations and MUSE and ALMA data”

14:50: Miguel Angel Ibarra Arellano (Supervisor D. Schapiro)
“Leveraging spatial-omics through the development of accessible machine learning frameworks with a focus on tissue architecture”

15:10: Josef Martinek (Supervisor R. Scheichl)
“Monte Carlo methods for Particle Transport”

15:30: Charlotte Boys (Supervisor J. Saez-Rodriguez)
“Multi-omics analysis of IgA nephropathy”

From blogging to news and opinion articles, writing remains one of the best ways for scientists to reach the public. While there are many styles and formats for popular science writing, certain elements of good writing are required. In this workshop, we discuss the various formats of writing as well as why and when each is used. We will also discuss how to tell a good story and common bad habits scientists have when writing for the public. Participants will practice writing short pieces based on their own work or interests that will be shared and discussed during the workshop.

All participants need a microphone and ideally a webcam.

Hardware integrated into the laptop is sufficient.

Topics (selection)

- Formats (blogs, news, opinion, etc.)
- Writing styles
- Elements of a good story
- Story structure
- Common mistakes (passive voice, jargon, etc.)

In recent years, machine learning (ML) and artificial intelligence (AI) methods have begun to play a more and more enabling role in the sciences and in industry. In particular, the advent of large and/or complex data corpora has given rise to new technological challenges and possibilities. In his talk, Müller will touch upon the topic of ML applications in the sciences, in particular in chemistry and physics. He will also discuss possibilities for extracting information from machine learning models to further our understanding by explaining nonlinear ML models. Finally, Müller will briefly discuss perspectives and limitations.

About Prof. Müller:

Klaus-Robert Müller has been a professor of computer science at Technische Universität Berlin since 2006; at the same time he is directing rsp. co-directing the Berlin Machine Learning Center and the Berlin Big Data Center and most recently BIFOLD. He studied physics in Karlsruhe from1984 to 1989 and obtained his Ph.D. degree in computer science at Technische Universität Karlsruhe in 1992. After completing a postdoctoral position at GMD FIRST in Berlin, he was a research fellow at the University of Tokyo from 1994 to 1995. In 1995, he founded the Intelligent Data Analysis group at GMD-FIRST (later Fraunhofer FIRST) and directed it until 2008. From 1999 to 2006, he was a professor at the University of Potsdam. From 2012 he has been Distinguished Professor at Korea University in Seoul. In 2020/2021 he spent his sabbatical at Google Brain as a Principal Scientist. Among others, he was awarded the Olympus Prize for Pattern Recognition (1999), the SEL Alcatel Communication Award (2006), the Science Prize of Berlin by the Governing Mayor of Berlin (2014), the Vodafone Innovations Award (2017), Pattern Recognition Best Paper award (2020), Digital Signal Processing Best Paper award (2022). In 2012, he was elected member of the German National Academy of Sciences-Leopoldina, in 2017 of the Berlin Brandenburg Academy of Sciences, in 2021 of the German National Academy of Science and Engineering and also in 2017 external scientific member of the Max Planck Society. From 2019 on he became an ISI Highly Cited researcher in the cross-disciplinary area. His research interests are intelligent data analysis and Machine Learning in the sciences (Neuroscience (specifically Brain-Computer Interfaces, Physics, Chemistry) and in industry.

In this talk, I will venture a glimpse into the future of cutting-edge numerical methods and software for computational fluid dynamics (CFD) for many aerospace applications. While established, off-the-shelf commercial solutions for a broad range of applications exist, the current push towards fundamental changes in the fields of energy generation and transport necessitates computational tools that can keep pace and facilitate discovery through computation.
This talk will give an overview of the ingredients of modern simulation software for CFD, in particular for multi-scale, multi-physics and multi-fidelity problems. Among these are high-order discretization schemes that can adapt locally in space and time to the underlying resolution requirements and thus combine accuracy and robustness. While the potential of leveraging data-driven approaches has become apparent, I advocate for a shift from a purely data-driven to a data-informed approach, which combines both machine learning models (ML) and classical solution schemes consistently. Last but not least, modern discretization methods and ML/CFD hybrid schemes must produce efficient and reliable results. For this, GPU-parallelization and an incorporation of the FAIR principles into the simulation stack must be part of the next generation CFD solvers.

Working as a PhD student you have the challenging task of developing research findings and write you doctoral thesis within three years. This alone is a demanding job. In addition, it is vital to the scientific process that your findings are presented to the scientific community. For most PhD students this is the first big project in their professional life and it could have a crucial impact on their future professional career. PhD students are highly motivated when they start their PhD studies but may underestimate the need for professional management for this three-year project "doctoral thesis".

This seminar demonstrates how to approach the doctoral thesis in a professional way. Project management tools and techniques are used, tailored to the specific situation of PhD students. You will learn how to set a project vision, define clear objectives, gain buy-in from your supervisor and other colleagues in your group, and how to develop a project plan, which is structured and at the same time flexible enough to easily adjust to unexpected findings. You will establish a "controlling cycle" which helps you to recognize risks and problems as early as possible, and you will learn how to manage critical situations and deal with ups and downs. Furthermore, networking with colleagues, supervisors and other people are important topics of this seminar.

Throughout the seminar, you will work on your own doctoral thesis and share your experience with others. This seminar is most beneficial for PhD students who are in the early phases of their doctoral thesis. At the end of the seminar you will have established a strategy on how to approach your own doctoral thesis. During the follow-up REVIEW we will share experience and best practices and deal with open questions from the first module.

This seminar will help you to make the most effective use of your three years and finish your doctoral thesis on time.
You will also learn and practice the basic concepts of project management – a prerequisite in industries and research institutions. Working as a PhD student you have the challenging task of developing research findings and write you doctoral thesis within three years. This alone is a demanding job. In addition, it is vital to the scientific process that your findings are presented to the scientific community. For most PhD students this is the first big project in their professional life and it could have a crucial impact on their future professional career. PhD students are highly motivated when they start their PhD studies but may underestimate the need for professional management for this three-year project "doctoral thesis".

This seminar demonstrates how to approach the doctoral thesis in a professional way. Project management tools and techniques are used, tailored to the specific situation of PhD students. You will learn how to set a project vision, define clear objectives, gain buy-in from your supervisor and other colleagues in your group, and how to develop a project plan, which is structured and at the same time flexible enough to easily adjust to unexpected findings. You will establish a "controlling cycle" which helps you to recognize risks and problems as early as possible, and you will learn how to manage critical situations and deal with ups and downs. Furthermore, networking with colleagues, supervisors and other people are important topics of this seminar.

Throughout the seminar, you will work on your own doctoral thesis and share your experience with others. This seminar is most beneficial for PhD students who are in the early phases of their doctoral thesis. At the end of the seminar you will have established a strategy on how to approach your own doctoral thesis. During the follow-up REVIEW we will share experience and best practices and deal with open questions from the first module.

This seminar will help you to make the most effective use of your three years and finish your doctoral thesis on time.
You will also learn and practice the basic concepts of project management – a prerequisite in industries and research institutions.

In this lecture, we cover concepts and selected techniques in visual data analysis. We investigate approaches for data transformation, visual representation, and interactive exploration, and their role in visual sense-making.

Lab presentations:
Tristan Bereau, Fred Hamprecht, Caroline Heneka

Science talks:
• Tristan Bereau: Machine learning representation for molecular dynamics
• Roman Remme (Hamprecht lab): Machine Learning meets Density Functional Theory 
• Lara Alegre (Heneka lab): Machine Learning for large radio surveys

Navigating the challenges of the global COVID-19 pandemic required strategic decisions tailored to each country's unique circumstances. In less developed nations, the threat of overwhelming hospital capacity was especially severe. Rather than building scenarios based on mathematical models, our team used a dynamic forecasting approach. We developed a series of models that provided 4-week probabilistic forecasts, complete with uncertainty quantification. These forecasts, crucially informed by real-time data, predicted the demand for hospital beds and ventilators, which served as the backbone for decisions and public policies adopted by federal health authorities in Mexico from April 2020 to January 2022. The journey was a challenging one. An incompletely characterized virus and the unpredictable dynamics of societal behavior made crafting a useful model difficult.

In this talk, I will present a retrospective of these models and critically review their objectives, successes, and limitations. Our methodology and modeling decisions will be presented, from the intricacies of data used for model fitting to the balance between model complexity and parameter identifiability. The predictive power and performance of our models will also be reviewed. Beyond the algorithms and forecasts, I'll share our experiences collaborating with health authorities and communicating with the general public. Discover the highs and lows, the challenges faced, and the insights we gained.

This presentation consists of two parts and perspectives.

PD Delia Dumitrescu from social/political sciences will speak about communication research using a multi-country, multi-language data set containing COVID-19 social media disinformation posts. Delia will present the initial difficulties related to the dataset and how she approached the Scientific Software Center for support. This support resulted in AMMICO (AI Media and Misinformation Content Analysis Tool): AMMICO is a publicly available software package written in Python 3, whose purpose is the simultaneous evaluation of the text and graphical content of image files. After describing the software features, Delia will provide an assessment of its performance using the social media disinformation posts. This part will conclude by highlighting the tool’s advantages for communication research and by discussing further development ideas.

In the second part of the talk, Dr Inga Ulusoy from the Scientific Software Center (SSC) will speak about the role of the SSC in this project. The initial consultation evolved into a software development project of the SSC. Inga will describe how the project and communication was structured and how the software was designed to fit the needs of the researchers. In the end, Inga will demonstrate how AMMICO can be utilized in further research projects, and conclude by summarizing the ways the SSC can support research projects through research software development and the implementation of software engineering best practices.

Neuroscience-inspired artificial intelligence aims to develop AI systems that learn and act in ways similar to the human brain. This talk will discuss some recent advances in this field, starting with successful examples of neuroscience-inspired machine learning approaches, such as Vector Symbolic Architectures, Hierarchical Temporal Memory, and FlyHash. We also outline recent studies on the similarity of processing in transformers, the deep learning models at the core of Large Language Models, and the structure of brain activities recorded with fMRI during listening tasks. Finally, we discuss the fundamental differences between backpropagation and likely biological learning mechanisms, and explore alternative learning algorithms such as Geoffrey Hinton's forward-forward algorithm. We conclude by examining the implications of these findings for open problems in AI such as one-shot learning, model robustness, and explainability.

Numerical simulations are widely used as a third pillar besides experimental and theoretical investigations in many sciences such as physics and chemistry as well as engineering science. It requires the development of robust and efficient numerical methods for the resolution of the underlying physical laws that arise often in form of partial differential equations (PDEs). In this talk, I will describe two interdisciplinary stories. While illustrating the key-ideas of the theory and methods, I will also highlight the occasions where not only the mathematical tools have been successfully developed and transferred to the application but also where the interdisciplinary interactions raised new mathematical questions and triggered new answers and theories in mathematics. From the application viewpoint, this talk will touch upon implicit solvation models and Born-Oppenheimer molecular dynamics but the methods rely on various and divers mathematical concepts from Grassmann manifolds to descriptors from machine learning and perturbation theory.

Prerequisites:

No prior knowledge by the participants is necessary to participate in this course and it is intended for all scientific audiences. Participants are required to bring their own laptops to work on during the course. Network access (e.g. through Eduroam) is recommended.

Summary:

The Unix shell is a powerful tool that allows people to do complex things with just a few keystrokes. More importantly, it helps them combine existing programs in new ways and automate repetitive tasks so they aren’t typing the same things over and over again. Use of the shell is fundamental to using a wide range of other powerful tools and computing resources. The course will include hands-on live coding sessions where participants exercise the learned commands on their own computers.

Version control is the lab notebook of the digital world: it is used to keep track of what was done and to collaborate with other people. Its use is the state of the art in software development projects of all scales. However, it is not limited to software: books, papers, small data sets, and anything that changes over time or needs to be shared can and should be stored in a version control system. The course will include hands-on live coding sessions where participants exercise the learned commands on their own computers.
Learning Objectives

After the course participants will:

- Have a fundamental understanding of how and why to use the Unix Shell
- Be comfortable with handling files and directories using the command line
- Have experience with advanced usage of the shell e.g. loops, pipes, redirects etc.
- Know how to write their workflows as reusable shell scripts
- Understand the benefits of using version control
- Understand basic git terminology
- Have a good working knowledge of common tasks in Git
- See how Git repositories can help them to move towards practicing Open Science

Feeling anxious about an upcoming international conference? Is public speaking one of your greatest fears? Unsure how you will sound in front of an audience of experts? The aim of this workshop is to help you plan and deliver an effective academic presentation in English. The workshop will be divided into two parts:

Part 1: the nuts and bolts of an effective academic

Part 2: interactive and practice session with attention to both rhetorical devices and non-verbal communication strategies.

The workshop’s ultimate goal is to help you gain a sense of comfort and confidence while presenting your research in public. Participants are encouraged to bring a sample presentation of their own for practice.

Objectives:

- Identify and review rhetorical and oral presentation strategies
- Develop confidence and poise for the effective delivery of your research results
- Help you develop your own presentation style

Mutual predictability is the key to interaction. Or in simpler terms: "experience makes teamwork". Of course, prediction isn't all that simple. We'll look at "generative AI", but without the hype, and see how we use that to learn the dynamics of complex stochastic systems, use that to predict -- and control. And give some examples, in robotics and beyond, of where this can be used.
For more see the abstract_file:

Lab presentations:
Christoph Dieterich, Sandy Engelhardt, Britta Velten

Science talks:
• Adrian Chan (Dieterich lab): Fingerprinting your RNA - one molecule at a time
• Sven Köhler (Engelhardt lab): 4D Self-Supervised Deep Cardiac Motion Modelling
• Florin Walter (Britta Velten lab): Probabilistic Factor Models for Subcellular Spatial Transcriptomics

Collecting, processing and analyzing data are central activities for virtually every researcher. Topics like data sharing and data publication are becoming increasingly important. Nevertheless, many research projects lack a structured and well-organized data management. This course is meant to give a general, discipline-independent introduction into various topics central to an efficient management of research data with a special focus on questions related to data archiving and data sharing. Both are central aspects of good scientific practice. Archiving and long-term preservation of research data are prerequisites for the scrutiny of scientific results based on the analysis of this data. Data sharing on the other hand increases transparency of research results and enables possible re-usage of data for new research questions, in combination with additional data sets and in interdisciplinary contexts.

14:30 Armand Rousselot (Supervisor U. Köthe)
"Invertible Neural Networks for Model-Based AI"

14:50 Maike Rees (Supervisor L. Maier-Hein)
“Deep learning for functional parameter quantification in medicine”

15:10 Lin Tian (Supervisor A. Dreuw)
“Theoretical Description and Prediction of Molecular Properties of N-Heteropolycycles through quantum chemical methods”

15:30 Julian Heidecke (Supervisor J. Rocklöv)
“Climate Change and infectious diseases – the case of West Nile virus: A modeling and data science approach”

In the academic world, clear, concise, and well-written texts play an important role in convincing journal editors and conference organisers to accept a paper for review and publication, or to invite a researcher to present at a conference. This workshop supports PhD students and novice researchers who need to produce scientific papers in English. Providing participants with strategies to write short texts efficiently and effectively, it also focuses on standards of professional papers aimed at Anglo-Saxon or international research communities. It enables participants to organise ideas and structure texts effectively, to present their own and other researchers‘ findings and opinions appropriately, and to use correct terminology and vocabulary.

During the workshop, participants work with their own texts as well as with examples from their own disciplines that they bring along and consider to be particularly well written. They discuss features of good scientific papers and are equipped to use adequate language in different genres and for different audiences. In addition, they receive peer feedback on their own drafts. All exercises empower them to produce clearer, and more correct, concise, and reader-oriented papers.

It’s far more than “just” a presentation: defending your thesis is the last (and one of the biggest) challenges in your academic qualification – and it doesn’t only require well-prepared content. This workshop addresses aspects of the defense related to communication and presentation. We will work on participants’ body language as well as on strategies for convincing argumentative and rhetorical positioning during the disputation. A significant portion of the workshop will be devoted to simulated defenses and evaluations of the performances.

The understanding of the appearance of extremes in real-life time series (such as weather and climate observations, returns of stock prices, exchange rates, and stock indices, insurance claim data, failures in energy and social networks) requires suitable probabilistic models and their statistical analyses. Over the last 15-20 years such models and statistical tools have been developed under the assumption of serial dependence. They supplement classical extreme value analysis which deals with independent data.

The goals of the course are

- to introduce and discuss the recent developments of extreme value theory in the time series context. The main focus will be on heavy-tail phenomena, where extremes are particularly severe, and clustering effects when extremes appear in clumps, 

- to provide suitable statistical tools for analyzing the aforementioned phenomena,

- to provide relevant knowledge to graduate students about extreme behavior of random systems in contrast to their average behavior, 

- to learn about applications of extreme value theory from top experts in the field.

Prerequisites:

Basic Python knowledge and a laptop is required. Experience with writing tests is not required.

Summary:

An automated test suite makes it much easier to maintain, extend and debug your Python code. In this course we will learn how to write automated tests in Python using the pytest library. After introducing the key concepts, the majority of the course will be hands-on, writing and running tests.
Learning Objectives

After the course participants should be able to:

- Install and run pytest
- Write simple tests
- Use temporary files in tests
- Use fixtures to manage resources
- Parametrize tests
- Add an automated test suite to their existing python projects

You will learn to visually communicate your complex research ideas and results so your messages are effortlessly understood by any specific audience (scientists or non-scientists). We will not focus on aesthetics but on how understanding human visual perception can inform your design decision for better comprehension of your scientific images, posters, and slides. You will also design a graphical abstract of your research, discuss it with peer scientists in a group exercise, and get actionable advice and feedback on your own materials. It is an immersive workshop, comprehensive, structured, memorable, easy to follow, useful and fun. More at https://www.seyens.com

Content & Method:

The training is offered as blended learning that combines a self-study module and a live online workshop. All participants get 12 month access to all materials.

Teaching methods:

The course is divided into six blocks (half days). Each block will consist of a mixture of presentations and hands-on exercises. In the presentations we take starting point in real-world datasets and scientific problems and discuss the relevant statistical methods. Focus is on the link between data and methods, implementation, and interpretation of results. Exercises will be practical, with exploration and analysis of real-world datasets. Participants are expected to use the statistical software environments R and RStudio for the exercises.

Teaching material:

The teachers will provide slides, programs/output from R Markdown, data, and exercises. Apart from this, notice that the book R for Data Science (https://r4ds.had.co.nz or https://r4ds.hadley.nz) is useful for data manipulation and visualization.

Tentative plan:

Each block (A-F) is expected to last for a morning/afternoon.

A. Recap (getting a common language)

• Basic statistical principles (data types, random variation, estimation, hypothesis tests, etc.)
• Workflow with R Markdown

B. Models for continuous outcomes

• Models for regression, ANOVA, clustered/blocked data (linear mixed models)
• Model diagrams
• Hypothesis tests for overall effects in regression, ANOVA, linear mixed models

C. Elaborate analysis of data with continuous outcome

• Estimated marginal means (emmeans)
• Pairwise comparisons and adjustment for multiple testing
• Statistical power

D. Modifications of models and analyses for continuous outcomes

• Model validation and data transformation
• Bootstrap computation of standard errors and p-values

E. Models with non-linear relationships

• Additive models (GAMs)
• Non-linear regression

F. Prediction with multiple/many covariates

• Preprocessing
• Training/validation/testing
• Regression-based models for prediction
• Principal component analysis (PCA)

This second seminar is planned as follow-up session to the Introduction seminar for those researchers who would like to review their own work in terms of potential patent protection. For that purpose, participants of the Introduction seminar will be asked to provide a short summary of their own already-disclosed innovations. Three to four selected cases will be analyzed and discussed with all participants with regard to the patentability requirements.
Concerning the patentability of not-yet-disclosed innovations, at the end of the workshop there will be the possibility to plan specific one-to-one meetings to avoid compromising the novelty of the underlying subject matter.

11:00 Antonia Papapostolou (Supervisor A. Dreuw)
“Development of Quantum Chemical Methods for the Calculation of Molecular Response Properties”

11:20 Adrian Müller (Supervisor A. Dreuw)
“Quantum mechanical method development for MOST systems”

11:40 Friederike Schneider (Supervisor A. Dreuw)
"High-Performance Quantum Chemical Computing"

12:00 Panasun Manorost (Supervisor P. Bastian)
“Surface-subsurface flow simulation. Creating Method and Application”

12:20 Pascal Lafrenz (Supervisors H. Kaessmann, B. Velten)
"Multi-omics data integration and evolutionary models to understand de novo gene birth and human evolution"

14:00 Leonie Kreis (Supervisor R. Herzog)
„Multilevel architectures and algorithms in deep learning“

14:20 Yannick Pauler (Supervisor C. Rother)
“Recovering 3D Information from a Moving Camera"

14:40 Leo Nguyen (Supervisor U. Koethe)
„Advancing Lithium Battery Design through Machine Learning using Invertible Neural Networks“

This seminar illustrates the strategic relevance of patents in the field of software for academic institutions and explains the requirements for software-based innovation to be considered as patentable software inventions. In particular, a series of examples will demonstrate opportunities and pitfalls when protecting software-based innovation by means of patent applications. One goal of the seminar is to provide researchers with a tool which they can use for assessing their own research results with regard to the potential that the underlying innovation can be protected via a patent.

In the 21st century, society faces multiple and highly interwoven global health challenges, linked to accelerating climate change, antimicrobial resistance, food and water insecurity, and an increasing number of infectious diseases. Threatening convergence, these health issues call for more effective and sustainable solutions to ensure that the global population continues to thrive. Scientists and researchers can massively contribute to discovering, co-designing, and evaluating these solutions.

In this context, Nature-based Solutions (NbS), while not yet commonplace, are increasingly gaining recognition alongside more conventional engineering solutions and healthcare interventions. The concept of NbS describes the innovative use of existing knowledge about natural systems to assist society in effectively addressing contemporary environmental, social and economic challenges while simultaneously providing ecological and health benefits. These win-win NbS include the management, restoration and protection of ecosystems, strategies for climate adaptation and mitigation and the greening of infrastructure.

The Summer School will explore the innovative potential of NbS to tackle two pressing global health challenges – climate change and pandemic prevention, which are highly interconnected and provide opportunities to design, apply and evaluate a range of win-win NbS to inform future policies based on scientific evidence.

The guest speakers will share their expertise and experience with the participants and will collaboratively work together on a capstone project, relevant to the development and presentation of a research project on NbS to tackle climate change and pandemics.The Summer School will promote interdisciplinary thinking and will encourage participants to bring examples of NbS beyond these two challenges for discussion to consider how to develop NbS that address concrete challenges and implement solutions in very diverse settings. As a result, the Summer School will allow participants to further connect, interact, and build a worthwhile network of expertise on NbS.

Advanced statistical methods are rapidly impregnating many scientific fields, offering new perspectives on long-standing problems. In materials science, data-driven methods are already bearing fruit in various disciplines, such as hard condensed matter or inorganic chemistry, while comparatively little has happened in soft matter. I will describe how we use multiscale simulations to leverage data-driven methods in soft matter. We aim at establishing structure-property relationships for complex thermodynamic processes across the chemical space of small molecules. Akin to screening experiments, we devise a high-throughput coarse-grained simulation framework. Coarse-graining is an appealing screening strategy for two main reasons: it significantly reduces the size of chemical space and it can suggest a low-dimensional representation of the structure-property relationship. I will briefly mention a biological application of our methodology that led to the discovery of in vivo active compounds. Finally, I will mention a number of ways machine learning can help fulfill the promise of connecting models at different scales.

Biosketch:

Tristan Bereau is a computational physicist working at the interface between multiscale modeling and machine learning for soft matter and biomolecules. He earned a Ph.D. in Physics at Carnegie Mellon University. After a postdoc at the University of Basel, he led an Emmy Noether research group at the Max Planck Institute for Polymer Research. He then moved to the University of Amsterdam as an assistant professor in chemistry and computer science, followed by a role in Industry. Tristan serves on the editorial boards of the journals Machine Learning: Science & Technnology and Computational Science and Engineering. He is currently a professor at the Institute for Theoretical Physics at the University of Heidelberg.

Convolutional Neural Networks (CNN) are the current backbone of deep learning architectures in a wide range of applications which process array data, like 2D/3D images. Despite this overwhelming success of the application of CNNs in terms of qualitative (visual) results and classification test set accuracies, this talk will point out some servere inherent problems of CNNs regarding their insufficient signal processing capabilities.
While these “signal processing flaws” have been largely ignored as test accuracies on many problems have been increasing over many years, recent research showed that current models are highly vulnerable to even the slightest changes in input distributions. In our latest works [1-5], we showed that the missing “robustness” of CNNs is not only due to insufficient training data, but to a large extend is also caused by faulty operators and architectures which are failing to adhere to basic signal processing demands. The aim of this talk is to give an overview of the dominant problems in the context of image processing and analysis and discuss possible counter measures.

References:
[1] Durall, R., Keuper, M., & Keuper, J. (2020). Watch your up-convolution: Cnn based generative deep neural networks are failing to reproduce spectral distributions. CVPR 2020
[2] Grabinski, J., Keuper, J., & Keuper, M. (2022). Aliasing and adversarial robust generalization of CNNs. Machine Learning, 111(11), 3925-3951.
[3] Grabinski, J., Jung, S., Keuper, J., & Keuper, M. (2022). FrequencyLowCut Pooling--Plug & Play against Catastrophic Overfitting. ECCV 2022
[4] Grabinski, J., Gavrikov, P., Keuper, J., & Keuper, M. Robust Models are less Over-Confident. NeurIPS 2022
[5] Gavrikov, P., & Keuper, J. (2022). CNN Filter DB: An Empirical Investigation of Trained Convolutional Filters. CVPR 2022 (oral)