Hello, I'm
Francesco Pio Ramunno
AI Research Scientist | PhD candidate in Deep Learning & Astrophysics
I’m an AI Research Scientist and a PhD candidate in Deep Learning and Astrophysics at the University of Geneva and FHNW. On 1 December 2026 I’m joining Sony AI in Zurich as a Research Scientist in the AI for Chip Design team.
I work on scientific machine learning: diffusion models, transformers for PDE surrogates, video generation, and long-horizon spatio-temporal forecasting. I spent my PhD teaching generative models to understand the Sun, with work on solar image synthesis, super-resolution of magnetograms and forecasting of solar evolution. Since then I’ve scaled this work up to foundation models for science at Polymathic AI and to semiconductor device simulation at Sony AI.
Before machine learning, my background was in mathematics and astrophysics. I care about models that are accurate, physically consistent and interpretable. On this site you’ll find my work, slides, code and notes. Feel free to reach out!
Experience
Dec 2026 – present
AI Research Scientist
Sony AI · Zurich, Switzerland
AI for Chip Design team, working on machine learning for semiconductor devices.
Feb 2026 – Aug 2026
Research Scientist Intern in Machine Learning
Sony AI · Zurich, Switzerland
- Designed a scalable dual-attention transformer for PDE surrogate modeling on large semiconductor meshes (500k+ nodes), combining global token attention with adaptive sparse local refinement.
- Improved the state of the art by more than 30% on open-source 2D/3D PDE benchmarks and on Sony's in-house semiconductor simulation datasets.
- Integrated the architecture into Sony AI's production semiconductor simulation framework. First-author paper submitted to ICLR 2027.
Mar 2025 – Aug 2025
AI Foundation Models for Science Intern
Polymathic AI (Simons Foundation) · New York City, USA
Supervisors: Rudy Morel, David Fouhey
- Built and scaled data pipelines for a 100+ TB multimodal spatio-temporal NASA solar dataset, synchronizing 14 modalities.
- Designed diffusion-based video models for long-horizon scientific forecasting, reducing exposure bias compared with autoregressive baselines. Published at NeurIPS 2025 (Main Track).
- Scaled high-resolution multi-channel diffusion models to distributed training on up to 32 NVIDIA H200 GPUs.
Mar 2022 – Aug 2022
Data Science and Machine Learning Intern
European Space Agency (ESA) · Frascati, Italy
Supervisors: Marco Micheli, Juan-Luis Cano
- Built end-to-end pipelines combining orbital, physical and chemical properties of near-Earth objects from heterogeneous databases.
- Trained ML models (XGBoost, Random Forests) that halved the RMSE of classical regression baselines for asteroid composition prediction.
- Delivered reproducible workflows and documentation to the ESA Near-Earth Object Coordination Centre (NEOCC).
Mar 2021 – Aug 2021
Data Science Intern
University of Rome Tor Vergata · Rome, Italy
Morphological analysis of XMM-Newton X-ray data with wavelet filtering, improving parameter measurement accuracy by more than 30%.
Education
2022 – 2026
PhD in Deep Learning and Astrophysics
University of Geneva & FHNW · Switzerland
Supervisors: Prof. Slava Voloshynovskiy, Prof. André Csillaghy
2020 – 2022
MSc in Astrophysics and Space Science
University of Rome Tor Vergata · Rome, Italy
110/110 with honours · Supervisors: Dario Del Moro, Juan-Luis Cano, Marco Micheli
2017 – 2020
BSc in Mathematics
Università degli Studi della Basilicata · Potenza, Italy
110/110 with honours · Thesis advisor: Prof. Ermenegildo Caccese
Publications
2026
F. P. Ramunno, S. Mestici, I. Coco, M. Walach, S. Massetti, M. F. Marcucci, B. Panos, A. Csillaghy
arXiv preprint (submitted to Journal of Geophysical Research: Machine Learning and Computation)
Paper
Abstract
2025
R. Morel, F. P. Ramunno, J. Shen, A. Bietti, K. Cho, M. Cranmer, S. Golkar, et al., D. Fouhey, S. Ho
Advances in Neural Information Processing Systems 38 (NeurIPS 2025), Main Track
Paper
Abstract
2025
F. P. Ramunno, P. Massa, V. Kinakh, B. Panos, A. Csillaghy, S. Voloshynovskiy
Astronomy and Astrophysics
Paper
Code
Abstract
2024
F. P. Ramunno, H.-J. Jeong, S. Hackstein, A. Csillaghy, S. Voloshynovskiy, M. K. Georgoulis
Proceedings of SPAICE2024: The First Joint European Space Agency / IAA Conference on AI in and for Space, 17-19 September 2024
Paper
Code
Demo
Abstract
2024
F. P. Ramunno, S. Hackstein, V. Kinakh, M. Drozdova, G. Quétant, A. Csillaghy, S. Voloshynovskiy
Astronomy & Astrophysics
Paper
Abstract
All publications & conference abstracts →
Skills
- Research areas
- Scientific ML, Generative Modeling, Diffusion Models, Transformers, PDE Surrogates, Graph Neural Networks, Computer Vision, Video Generation, Uncertainty Quantification
- ML frameworks
- PyTorch, JAX, Scikit-learn, Hugging Face, WandB
- Engineering
- Distributed Training, HPC, SLURM, Linux, Docker, Git, Large-scale Data Pipelines
- Programming
- Python, MATLAB, IDL, Bash
- Languages
- Italian (native), English (C2), German (A1)