I received a Short Term Scientific Mission (STSM) grant from the COST Action CA24136 - InterCoML, which funded a research visit at the Machine Learning Genoa Center (MaLGa) in Genova from 31.08. to 11.09. My hosts were Silvia Villa and Cesare Molinari, whom...
I gave a talk titled “Mean-field models for self-attention dynamics in transformers” at the workshop Gradient Flows Face-to-Face 6 meets Transformers at Ulm University. My talk was based on the following papers:
Together with Christian Fiedler we posted a new preprint on Consensus-based optimization for linearly separable functions. We propose a reparametrization strategy that provably finds coordinate transformations in which the objective becomes separable, which...
Together with José A. Carrillo, Daniel Kelly and Konstantin Riedl I organized the Workshop on the Mathematics of Transformers at the Mathematical Institute of the University of Oxford. Around 40 researchers came together to discuss the mathematical foundati...
I gave an online talk titled “Mean-field Models for Self-attention Dynamics in Transformers” at the BIRS workshop Interacting Particle Systems: Theoretical Innovations and Practical Applications in Hangzhou, China. My talk was based on the following paper:
I gave a talk at the joint seminar of TUM, LMU and KU Ingolstadt, hosted at KU Ingolstadt. My talk was based on the following paper:
I gave a talk at the Workshop on the Mathematics of Transformers, hosted at Julius Maximilians University of Würzburg. You can find more details on the event here and my talk was based on the following paper:
I gave a talk at the 11th International Conference on Curves and Surfaces in Saint-Malo, France. My talk was based on the following paper:
Together with Laura Paul, Holger Rauhut, Martin Burger and Samira Kabri we posted a new preprint on Allure of Craquelure: A Variational-Generative Approach to Crack Detection in Paintings.
At a recent workshop hosted by DESY, around 30 experts from across Europe came together to explore the theoretical foundations of transformers, the core architecture behind modern large language models. Organized by researchers from DESY, Helmholtz Imaging,...
We put a preprint on the arXiv dealing with position-blind pytychography. Compared to the standard setup, here we ask ourselves how well the reconstruction problem can be solved, when the scan position are completely unknown. Furthermore, we explore how to ...
I’m happy to share that our paper Adversarial flows: A gradient flow characterization of adversarial attacks has been published in the European Journal of Applied Mathematics.
Together with Danielle Bednarski we uploaded our lecture notes Introduction to Regularization and Learning Methods for Inverse Problems to arXiv. The notes are based on a course we taught at the University of Hamburg in the winter term 2024/2025.
Together with Leon Bungert and Philipp Wacker we uploaded a paper to arXiv, which shows how CBO can be used to create closed-box adversarial attacks. Beyond that we establish an interesting connection between CBO and evolution strategies. The paper can be f...
I’m happy to share that our paper Analysis of mean-field models arising from self-attention dynamics in transformer architectures with layer normalization has been published in the Philosophical Transactions of the Royal Society A.
I’m happy to share that our paper MirrorCBO: A consensus-based optimization method in the spirit of mirror descent has been published in Mathematical Models and Methods in Applied Sciences (M3AS).
I’m happy to share that our paper Polarized consensus-based dynamics for optimization and sampling has been published in Mathematical Programming.
I am very happy to share that I was awarded the Dr.-Klaus-Körper Prize for my PhD thesis today. The prize is awarded by the GAMM at the annual meeting in Poznań.
Together with Leon Bungert, Franca Hoffmann and Doh Yeon Kim we poste a new preprint on MirrorCBO: A consensus-based optimization method in the spirit of mirror descent.
Together with Martin Burger, Samira Kabri, Yury Korolev and Lukas Weigand we posted a new preprint called Analysis of mean-field models arising from self-attention dynamics in transformer architectures with layer normalization.
Our project COMFORT: Compression Methods for Robustness and Transferability is being funded by the BMBF!
I’m very excited to give a talk in the CMX Lunch seminar at Caltech today. I will present my work on the mathematics of adversarial robustness, you can find the abstract here.
I’m very happy to announce that our paper on the CBXPy package has been published in the journal of open source software and can be found here.
I presented our group poster “Computational Imaging@DESY” at the DESY Photon Science Users’ Days.
I was happy to give a talk on “Neuronale Netze und adversarial attacks: Wie der Panda zum Gibbon wurde” in the student seminar “Data Science in Forschung und Industrie” organized by Daniel Tenbrinck.
Together with Samira Kabri, Lorenz Kuger, Lukas Weigand and Martin Burger, we presented our group poster “Computational Imaging@DESY” at the Digital Total event in Hamburg.
Together with Samira Kabri, Lorenz Kuger and Martin Burger we organized a course at the Helmholtz Incubator Summer Academy.
I’m currently in Kelowna at the BIRS event Leveraging Model- and Data-Driven Methods in Medical Imaging. I will give a talk on our Bregman learning framework.
I will present the work on continuum limits of Lipschitz learning at the FoCM 2023 as a poster. You can find the poster here. It summarizes the main results of the following papers I wrote together with Jeff Calder and Leon Bungert:
I am currently presenting our work on Bregman learning at the Mathematics and Image Analysis conference 2023 in Berlin.
I’m currently attending the winter school on advanced methods for mathematical image analysis in Bologna. The stay is financed via the Erasmus BIP program.
A beamer template
📈 Bregman Learning
Very happy to be in NewOrleans right now at NeurIPS 22 presenting our paper on Bregman learning as a poster.
Together with Leon Bungert and Philipp Wacker we upload our preprint Polarized consensus-based dynamics for optimization and sampling. The code for the algorithm can be found in our repo.
Consensus Based Optimization
Check out our new preprint Ratio convergence rates for Euclidean first-passage percolation: Applications to the graph infinity Laplacian on arXiv. We prove the first quantitative convergence rates for the graph infinity Laplace equation fo...
This week I am visiting my colleague Leon Bungert at the Mittag-Leffler insitute in Stockholm. He is staying there for the program Geometric Aspects of Nonlinear Partial Differential Equations and I am very happy to have the chance to visit him there.
I am very happy to announce that our paper on Bregman learning will be presented as a poster at NeurIPS 22. You can checkout the poster on ResearchGate or on GitHub.
I am very happy to announce that our paper on Uniform Convergence Rates for Lipschitz Learning on Graphs has appeared in the IMA journal of numerical analysis.
fau-book
Today I was given the chance to present our work on convergence rates for the Lipschitz learning problem at the GAMM Annual Meeting 2022 in Aachen.
The paper on Bregman learning for sparse neural networks which I worked on together with Leon Bungert, Daniel Tenbrinck and Martin Burger just got published in the Journal of Machine Learning Research.
Today I had the chance to present our work on a kernelized version of consensus based optimization at the 15th International Conference on Monte Carlo and Quasi-Monte Carlo Methods in Scientific Computing.
Today I presented our work on convergence rates for the Lipschitz learning problem at the conference on calculus of variation in Lille.
Today I presented our work on convergence rates for the Lipschitz learning problem in our joint junior seminar. This seminar takes place within the network of the MoD community.
I was very happy to join the Curves and Surfaces conference in Arcachon. My supervisor Martin Burger organized a Minisymposium together with Daniel Tenbrinck which can be found here.
Today I had the chance to present our work on convergence rates for the Lipschitz learning problem at the HCM - Workshop Synergies between Data Science and PDE Analysis.
Mathematik für Physikstudierende C
Today I had the chance to present our work about stochastic Bregman iterations at the East Coastern Optimization Meeting (ECOM).
Together with Leon Bungert and Daniel Tenbrinck we organized a three-part Minisymposium for the SIAM conference IS22 on Recent Advances on Stable Neural Networks.
I am very happy to join the hackathon “Math Meets Image” in Berlin. It takes place from 17th to 19th March 2022 at The Classroom and is part of MATH+ Thematic Einstein Semester on “Mathematics of Imaging in Real-World Challenges”.
Together with Leon Bungert and Jeff Calder we published a preprint on Uniform Convergence Rates for Lipschitz Learning on Graphs.
Looking forward to join the workshop Dynamics and Discretization: PDEs, Sampling, and Optimization at the Simons Institute in Berkeley.
Was very happy to give a talk for the “Mädchen- und Technik 2021”-week together with Daniel Tenbrinck. This event allows young girls to gain an insight into the academic world.
📈 CLIP
Happy to announce that I will be presenting our paper about Bregman iterations at the 2nd Alps-Adriatic Inverse Problems Workshop which is taking place at Klagenfurt during September 22-24, 2021. I am especially thrilled that my talk is scheduled to be live...
Together with Leon Bungert, Daniel Tenbrinck and Martin Burger we published a new preprint called Neural Architecture Search via Bregman Iterations.
Together with Leon Bungert, Daniel Tenbrinck and Martin Burger we published a new preprint called A Bregman Learning Framework for Sparse Neural Networks.
Published a new preprint called CLIP: Cheap Lipschitz Training of Neural Networks, together with Leon Bungert, Daniel Tenbrinck, Leo Schwinn and René Raab.
Published my first preprint called Continuum Limit of Lipschitz Learning on Graphs, together with Leon Bungert.