Viktor Stein

Postdoctoral researcher in applied mathematics working on Wasserstein gradient flows, regularized divergences, optimal transport, and geometry-inspired machine learning.

Recent News

  • In September 2026, I gave the expository talk Moving Mass, Decreasing Energy: Learning Probability Distributions with Wasserstein Gradient Flows at the BMS–BGSMath Junior Meeting in Barcelona.

  • In winter semester 2026/27, together with Francesco Mattesini, I am again organizing the hybrid Oberseminar of the Applied Numerical Analysis Group at TUM.

  • Two papers were accepted at ICML 2026: Approximating $f$-Divergences with Rank Statistics and Well-Posed KL-Regularized Control via Wasserstein and Kalman–Wasserstein KL Divergences. I presented both as posters at ICML 2026 in Seoul.

  • On 25 June 2026, I defended my PhD thesis, Metric Gradient Flows in Measure Spaces: Kernelization and Acceleration.

  • In March 2026, I gave a talk on SympFormer: Accelerated Attention Blocks via Inertial Dynamics on Density Manifolds at the MFO workshop Flows on Measure Spaces and Applications in Machine Learning.

  • In March 2026, the poster Wasserstein Gradient Flows for Moreau Envelopes of $f$-Divergences in Reproducing Kernel Hilbert Spaces received the Best Poster Award at TUM’s research retreat Advanced Mathematical Methods in Data Analysis and Signal Processing.

  • New preprint: SympFormer: Accelerated Attention Blocks via Inertial Dynamics on Density Manifolds, joint work with Wuchen Li and Gabriele Steidl. Code

Upcoming

  • SIAM Imaging Science and SIAM Mathematics of Data Science, Salt Lake City, Utah, USA, November 2026.