Torben Berndt

Hi, I am Torben. I am a second-year PhD student in the Machine Learning and Artificial Intelligence Group at the Heidelberg Institute for Theoretical Studies, advised by Jan Stühmer. My research focuses on machine learning for data with geometric structure, particularly symmetries and model equivariances. I am currently interested in building models that retain the inductive bias of strictly equivariant models while matching the speed of unconstrained models.

Before moving to Heidelberg, I completed an MSc in Mathematical Sciences at the University of Oxford, where I wrote my master’s thesis under the supervision of Andrey Kormilitzin. Before that, I obtained a BSc in Mathematics from the University of Warwick.

I also have a personal interest in weather modelling, which I explored during two research internships at Ventus.

Feel free to email me if you are interested in discussing research.

News

May 2026: I gave a talk on our recent work on approximate equivariance at the SIMPLAIX workshop.
May 2026: Our paper “Approximate Equivariance via Projection-based Regularisation” was accepted at ICML 2026.
April 2026: I gave a talk on Permutation Equivariant Graph Neural CDEs at the Alan Turing Institute.
September 2025: Our paper “Permutation Equivariant Neural Controlled Differential Equations for Dynamic Graph Representation Learning” was accepted at NeurIPS 2025.

Selected Publications

arXiv

Universal Time Series Generation with Neural Controlled Differential Equations

Torben Berndt*, Elyes Farjallah*, Leif Seute, Raeid Saqur, Benjamin Walker† and Jan Stühmer†

2026

arXiv

Bib

@article{berndt2026universal,
  title = {Universal Time Series Generation with Neural Controlled Differential Equations},
  author = {Berndt, Torben and Farjallah, Elyes and Seute, Leif and Walker, Benjamin and St{\"u}hmer, Jan and Saqur, Raeid},
  journal = {arXiv preprint},
  year = {2026},
  url={http://arxiv.org/abs/2605.28507}
}

GitHub

ICML

Approximate Equivariance via Projection-based Regularisation

Torben Berndt and Jan Stühmer

2026

arXiv

Bib

@article{berndt2026approximate,
  title = {Approximate equivariance via projection-based regularisation},
  author = {Berndt, Torben and St{\"u}hmer, Jan},
  booktitle = {Forty-third International Conference on Machine Learning},
  year = {2026},
  url={https://arxiv.org/pdf/2601.05028}
}

GitHub

NeurIPS

Permutation Equivariant Neural Controlled Differential Equations for Dynamic Graph Representation Learning

Torben Berndt, Benjamin Walker, Tiexin Qin, Jan Stühmer and Andrey Kormilitzin

2025

arXiv

Bib

@article{berndt2025permutation,
  title={Permutation Equivariant Neural Controlled Differential Equations for Dynamic Graph Representation Learning},
  author={Berndt, Torben and Walker, Benjamin and Qin, Tiexin and St{\"u}hmer, Jan and Kormilitzin, Andrey},
  journal={Advances in Neural Information Processing Systems},
  volume={38},
  pages={98276--98311},
  year={2025}
}

GitHub