
Hi, I’m Lenard.
I’m a PhD student at the ELLIS Institute Tübingen and the MPI for Intelligent Systems in Tübingen and Stuttgart, where I work with Maximilian Dax and Renate Sachse on machine learning methods for the inverse design of metamaterials. Before that, I studied Physics at the University of Tübingen (M.Sc.). My interests sit at the intersection of metamaterials, Bayesian inference, scientific machine learning, and linear operators.
Projects
laplax
Contributor & Co-authorScalable Laplace approximations and function-space posterior approximations in JAX. Efficient curvature-based uncertainty estimates for high-dimensional models.
Finite differences in JAX — automatic differentiation and numerical methods for gradient computation.
linox
AuthorLinear operator abstractions and matrix-free solvers in JAX. Designed for scientific ML where explicit matrix formation is infeasible. Available on PyPI (pip install linox), with documentation at linox.lenardrommel.com.
Talks
Interactive visual essays on my research.
IT for Medical Practices
I currently work as an IT administrator at a medical practice, where I’ve been learning hands-on what actually matters in day-to-day operations — from infrastructure and data security to building a practice website. I’d be happy to sit down with other practices and work on solutions together.
A project I built along the way
The website for the practice I work at — built and maintained by me.
If any of this sounds relevant, feel free to reach out.
lenardrommel@icloud.comExperience
Machine learning methods for the inverse design of metamaterials. Supervised by Maximilian Dax (MPI-IS Tübingen) and Renate Sachse (MPI-IS Stuttgart). Part of MPI-AIX.
Uncertainty quantification for neural operators. Developed GP regression kernels for operator learning. Contributed to laplax.
Reservoir computing with swarms, optimizers for dynamical systems, reproducible pipelines (Snakemake, Docker).
Time-series prediction using LSTMs and Echo State Networks.
Education
Thesis: OSP-Laplace — Bayesian Neural Operators with Operator-Space Priors. Focus: ML, Probabilistic Numerics, Laplace Approximation, Neural Operators, Gaussian Processes.
Thesis: Finite Volume Neural Network for simple vortex models.
Publications
Skills
About
I’m interested in practical approaches to uncertainty quantification, including Laplace approximations and function-space perspectives on posterior approximation. Reliability and calibration matter — especially when ML meets real-world applications.
Much of my recent work involves developing matrix-free operator abstractions and scalable solvers in JAX. These methods enable Bayesian inference and optimization in high-dimensional settings where explicit matrix formation is infeasible.
I believe ML methods need to work reliably outside toy settings. My focus is on developing tools and methods that practitioners can actually trust in scientific and engineering applications.
PhD · since Nov 2026
Machine learning for the inverse design of metamaterials
ELLIS Institute Tübingen / MPI-IS (Tübingen & Stuttgart)
with Maximilian Dax & Renate Sachse
Education
M.Sc. & B.Sc. Physics
University of Tübingen