Paulo von Petersenn

About

Figure: Venus with Earth at the centre traces a rosette, a sum of sines; a learned change of frame puts the Sun at the centre, where it moves on a simple ellipse
  • Name: Paulo von Petersenn
  • Occupation: PhD student, University of Zurich
  • Research: algorithms that learn physical equations from data
  • Based in: Zurich, Switzerland

I'm Paulo von Petersenn, a PhD student at the University of Zurich, where I develop new algorithms that learn physical equations from data. My background is in physics and mathematics, and I'm drawn to the mathematical foundations of AI and to what AI can do for the sciences. Outside research I'm an avid language learner and I play the piano.

Contact

CV

CV of Paulo von Petersenn, page 1

Download CV (PDF)

Now PhD student, Sbalzarini lab, DM3L, University of Zurich

Developing new algorithms that learn physical equations from data, with a focus on the mathematical foundations of AI and on AI for the sciences. Supervised in the group of Prof. Ivo Sbalzarini. I started at the Center for Systems Biology Dresden (MPI-CBG / TU Dresden) and moved with the group to Zurich.

Before Guest researcher, Max Planck Institute for Mathematics in the Sciences, Leipzig

Complex Systems group. Neural network generalization through the lens of information theory and Solomonoff induction, and ℓ₀ regularization for pruning.

Abroad Hong Kong Baptist University and National Taiwan University

A semester in Hong Kong during my bachelor's, then a DAAD-funded year in Taipei during my master's studying quantum computing, quantum information theory, and Mandarin.

Teaching Faculty of Mathematics & physics didactics and labs, Leipzig University

Teaching assistant at the Faculty of Mathematics. Earlier, as a research assistant for physics didactics and labs, I built remote-learning solutions for physics labs during the COVID-19 pandemic, which led to a co-authored publication, A Personal, Portable Electric Circuit Lab (The Physics Educator).

Education M.Sc. Mathematical Physics and B.Sc. Physics, Leipzig University

Selected work

Efficient compression of neural networks and datasets, with Lukas Silvester Barth (arXiv:2505.17469). Compares and improves ℓ₀-regularization methods for shrinking neural networks while keeping test accuracy, and shows that minimizing description length yields strong data compression.

Probabilistic and nonlinear compressive sensing, with Lukas Silvester Barth (arXiv:2509.15060).

Pleco-to-Anki, an open-source Anki add-on for Chinese learners that turns Pleco dictionary bookmarks into nicely formatted flashcards.

News

Oct 2026 New homepage.

Oct 2026 Talk at the SciML lunch, DM3L, Zurich: "What does it mean to explain data?".

Sep 2026 Moved with the Sbalzarini lab to the University of Zurich.

Jul 2026 Attended the AI for Science European Summer School at the Acropolis Museum, Athens.

Apr 2026 Community outreach talk at the MPI-CBG, Dresden: "Teaching computers to think: why large language models work so well".

Nov 2025 Talk at the AI lunch, CSBD, Dresden: "Compression of neural networks and datasets through sparse learning".

Oct 2025 Started my PhD at the Center for Systems Biology Dresden, in the group of Ivo Sbalzarini.

Sep 2025 Preprint: Probabilistic and nonlinear compressive sensing (arXiv:2509.15060).

May 2025 Preprint: Efficient compression of neural networks and datasets (arXiv:2505.17469).