Experience & Education

Courant Institute, NYU Postdoctoral Researcher Jun 2023 - Apr 2026
Tech. Univ. Berlin Scientific Assistant Oct 2018 - Mar 2023
Mitsubishi Electric Research Labs Research Intern Oct 2018 - Mar 2023
Continental AG R&D Intern Nov 2015 - Mar 2016
Ph.D. in Mathematics TU Berlin, Germany Oct 2018 - Jan 2023
M.S. in Engineering Science TU Berlin, Germany Oct 2016 - Apr 2018
B.S. in Engineering Science TU Berlin, Germany Oct 2012 - Apr 2016

Recent Projects

Highly Compressible Sequence Models

Highly Compressible Sequence Models

Regularizing Hankel singular value distributions for compressible dynamics

Deep state space models (SSMs) are a new class of sequence models that with inference costs that scale logarithmically in the sequence length (compared to quadratically for transformers).

Contributions

  • Regularizing Hankel singular values to train SSMs that compress well via balanced truncation, with a provably differentiable nuclear norm objective
  • O(n²) algorithm for computing Hankel singular values by exploiting the matrix block-structure reducing state-dimension cost from O(n³)
  • Up to 10× accuracy improvement compared to previous compression methods

Related Publications

  1. P. Schwerdtner, J. Berman, B. Peherstorfer, Hankel Singular Value Regularization for Highly Compressible State Space models, NeurIPS 2025

Quadratic Manifolds

Quadratic Manifolds

Utilizing nonlinear dependencies in latent space

Quadratic manifolds augment linear approximations with quadratic correction terms. This is useful for finding efficient representations of high-dimensional data points to capture fine details with low-dimensional surrogates.

Contributions

  • Greedy algorithm leading to orders of magnitudes improvement in accuracy
  • Streaming-based manifold computation to process petabyte-scale simulation data
  • Applications to complex flow problems in collaboration with National Labs (LANL & NREL)

Related Publications

  1. P. Schwerdtner, B. Peherstorfer, Greedy construction of quadratic manifolds for nonlinear dimensionality reduction and nonlinear model reduction, 2024
  2. P. Schwerdtner, P. Mohan, A. Pachalieva, J. Bessac, D. O’Malley, B. Peherstorfer, Online learning of quadratic manifolds from streaming data for nonlinear dimensionality reduction and nonlinear model reduction, Proceedings of the Royal Society A, 2025
  3. Video: Talk at MORE-Conference (San Diego, 2024)
    4.
Show more

Multilevel Monte Carlo wildfire simulation

Multilevel Monte Carlo wildfire simulation

Enabling uncertainty quantification at operational scales

Developed methodology for merging multiple data sources for estimating burned areas in large-scale coupled atmospheric wildfire simulations.

Contributions

  • Contributed uncertainty quantification method to Google-research wildfire simulation framework swirl_lm
  • Multilevel Monte Carlo simulation to enable uncertainty quantification at operational scales
  • Designed large-scale simulation study using Google-Cloud TPUs to simulate wildfire scenarios in California

Related Publications

  1. P. Schwerdtner, F. Law, Q. Wang, C. Gazen, Y.-F. Chen, M. Ihme, and B. Peherstorfer, Uncertainty quantification in coupled wildfire–atmosphere simulations at scale, PNAS Nexus, 2024