Paul Setinek

Paul Setinek

Hi, I am Paul! I am a PhD student working on data-driven neural surrogates for large-scale physical simulations, supervised by Johannes Brandstetter at Johannes Kepler University (JKU) in Linz. Currently, I am visiting Steven Brunton at the University of Washington in Seattle.

Before my PhD, I received a bachelor's degree in mechanical engineering from TU Vienna and a master's degree in Applied Computational Science and Engineering from Imperial College London. In my master's thesis, I worked on implicit neural scene representations for underwater maintenance with Aker BP and Imperial College. I also worked on computer vision for automated quality control in manufacturing, using light microscopy, at the Zeiss Solutions Lab in Munich.

Research

Together with the Linz Center of Mechatronics, I worked on distribution shifts and the robustness of neural surrogates in real-world manufacturing simulations. This made me realize the importance of data in SciML on one side, and the cost of generating high-fidelity data on the other. Lately, I have therefore been working on training neural surrogates with multi-fidelity data, enabling compute-efficient scaling of these methods.

Selected publications

  • Compute-optimal data scaling for neural surrogates via multi-fidelity training

    P. Setinek*, F. Paischer, N. Thuerey, J. Brandstetter, F. Koehler*

    NeurIPS 2026 Paper Code Project page Data

  • Towards Multi-Fidelity Scaling Laws of Neural Surrogates in CFD

    P. Setinek, G. Galletti, J. Brandstetter

    NeurIPS 2025 Workshop Paper

  • SIMSHIFT: A Benchmark for Adapting Neural Surrogates to Distribution Shifts

    P. Setinek, G. Galletti, T. Gross, D. Schnürer, J. Brandstetter, W. Zellinger

    Preprint, 2025 Paper Code

  • GyroSwin: 5D Surrogates for Gyrokinetic Plasma Turbulence Simulations

    F. Paischer, G. Galletti, W. Hornsby, P. Setinek, L. Zanisi, N. Carey, S. Pamela, J. Brandstetter

    NeurIPS 2025 Paper