Research

What I work on

I build differentiable methods that respect the physics of the systems they describe, from planetary atmospheres to seismic waves to the human brain.

PhD 路 ongoing

Exoplanet atmospheres with JWST

JWST measures spectra of planets hundreds of light-years away. Inferring atmospheric composition from those measurements requires accounting for the telescope, detector and the planet鈥檚 transit.

I develop physics-based, differentiable methods for analysing JWST observations of transiting exoplanets. More details will follow once the first papers are out. In the meantime, my interactive explainer on transmission spectroscopy covers the foundations this work builds on.

Artist's impression of a hot Jupiter exoplanet orbiting close to its host star
Artist's impression of a hot Jupiter, a gas giant orbiting very close to its star. Credit: NASA

HomanLab 路 Z眉rich

How the brain follows a story

Diagram of the method: converting stories into drift and shift signals via a large language model and mapping them to brain activity
From stories to brain signals via a large language model.

At the HomanLab within the Psychiatric University Hospital of Z眉rich, I explored how our brains follow the flow of stories. Working with a language model, we distilled two simple signals from narratives: a drift signal capturing the gradual build-up of meaning, and a shift signal that spikes when the story moves to a new event or scene.

Compared against high-resolution fMRI recordings of a volunteer listening to crime stories, the burst-like shift signal lit up the brain's speech and hearing centres, while the slow drift signal was strongest in the default-mode network, in regions like the angular gyrus and precuneus that support memory and imagination. The auditory response was associated with event boundaries. Changes in the broader networks tracked the story鈥檚 evolving context.

Brain map showing shift responses in peri-Sylvian regions
Shift: event-boundary responses in auditory鈥搇anguage cortex.
Brain map highlighting drift responses in default-mode regions
Drift: slow accumulation responses in default-mode regions.

Master's thesis 路 ETH Z眉rich

Physics-informed neural networks for seismology

My master's thesis investigated physics-informed neural networks (PINNs) for solving the elastic wave equation, which describes how seismic waves travel through the Earth. By embedding wave physics directly into the architecture, using custom wavelet and plane-wave layers with encoder and decoder components, the networks were roughly twice as accurate as standard PINNs.

Conditioning the networks on the seismic source location lets them infer the wavefield for any source position in a single forward pass, far faster than traditional finite-difference solvers. The thesis was featured in the British Seismology Meeting 2024 report in Astronomy & Geophysics.

Simulated radial elastic wavefields produced by a physics-informed neural network
Elastic wavefields simulated by a physics-informed neural network.

Publications

arXiv:2512.20481
2025
Coherence in the brain unfolds across separable temporal regimes. D. Staub, F. Rabe, A. Misra, Y. Pauli, R. H眉ppi, N. Yang, N. Lang, L. Michels, V. Edkins, S. Fr眉hholz, I. Sommer, W. Hinzen, P. Homan. arXiv
arXiv:2605.21049
2026
Cross-lingual robustness of LLM-brain alignment and its computational roots. N. Yang, R. He, P. Homan, I. Sommer, D. Staub, W. Hinzen. arXiv
Master's thesis
ETH Z眉rich 路 2024
Solving the elastic wave equation with physics-informed neural networks: a robust and critical assessment. DOI 10.3929/ethz-b-000668359

Full list on Google Scholar.