Orestis Oikonomou

ooikonomou [at] ethz.ch

ETH Zürich

neurosymbolic AI · scientific machine learning · earth observation

I work on bridging physics, symbolic structure, and large neural models — building systems that don’t just fit data, but reason about the equations and processes underneath it. My current focus is neurosymbolic methods for partial differential equations and foundation models for SAR / multimodal Earth observation.

Recently I’ve been:

training neurosymbolic models that combine PDE structure with deep learning — work accepted to ICML 2026.
building ZeroFlood, a zero-shot flood mapping pipeline for SAR — accepted to EUSAR 2026.
competing in the IBM × HuggingFace TerraMind Challenge (1st-round winner) — presenting at the 2nd NASA / ESA workshop in the US, mid-May 2026.

If any of that sounds like your kind of thing, I’d love to chat. Drop me an email or find me on the socials below.

news

May 15, 2026	🛰️ Won the 1st round of the IBM × HuggingFace TerraMind Challenge — work will be presented at the 2nd NASA / ESA Workshop in the US, mid-May 2026.
May 01, 2026	🌊 ZeroFlood — our zero-shot SAR flood-mapping pipeline — was accepted to EUSAR 2026.
Apr 15, 2026	🎉 Our neurosymbolic PDE paper was accepted to ICML 2026!

selected publications

ICML
A Neurosymbolic Framework for Partial Differential Equations

Orestis Oikonomou, and others

In Proceedings of the 43rd International Conference on Machine Learning (ICML), 2026

Abs Bib

We present a neurosymbolic framework that integrates symbolic representations of PDE structure with deep learning, improving generalization, interpretability, and physical consistency of learned solvers across families of partial differential equations.
@inproceedings{oikonomou2026neurosymbolic, title = {A Neurosymbolic Framework for Partial Differential Equations}, author = {Oikonomou, Orestis and others}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning (ICML)}, year = {2026}, }
EUSAR
ZeroFlood: Zero-Shot Flood Mapping from SAR

Orestis Oikonomou, and others

In European Conference on Synthetic Aperture Radar (EUSAR), 2026

Abs Bib

ZeroFlood performs zero-shot flood extent mapping directly from SAR imagery without requiring labeled flood examples, enabling rapid response in regions and events with no available training data.
@inproceedings{oikonomou2026zeroflood, title = {ZeroFlood: Zero-Shot Flood Mapping from SAR}, author = {Oikonomou, Orestis and others}, booktitle = {European Conference on Synthetic Aperture Radar (EUSAR)}, year = {2026}, }
Workshop
Foundation-Model Approaches to Earth Observation: Lessons from the TerraMind Challenge

Orestis Oikonomou, and others

2026

1st-round winner, IBM × HuggingFace TerraMind Challenge

Abs Bib

Approach and findings from the IBM x HuggingFace TerraMind Challenge, where our submission won the first round. Presented at the 2nd NASA/ESA workshop in the US, May 2026.
@misc{oikonomou2026terramind, title = {Foundation-Model Approaches to Earth Observation: Lessons from the TerraMind Challenge}, author = {Oikonomou, Orestis and others}, howpublished = {2nd NASA / ESA Workshop on AI for Earth Observation}, year = {2026}, note = {1st-round winner, IBM × HuggingFace TerraMind Challenge}, }