Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

Surya: Foundation Model for Heliophysics

Created by
  • Haebom

Author

Sujit Roy, Johannes Schmude, Rohit Lal, Vishal Gaur, Marcus Freitag, Julian Kuehnert, Theodore van Kessel, Dinesha V. Hegde, Andr es Mu noz-Jaramillo, Johannes Jakubik, Etienne Vos, Kshitiz Mandal, Ata Akbari Asanjan, Joao Lucas de Sousa Almeida, Amy Lin, Talwinder Singh, Kang Yang, Chetraj Pandey, Jinsu Hong, Berkay Aydin, Thorsten Kurth, Ryan McGranaghan, Spiridon Kasapis, Vishal Upendran, Shah Bahauddin, Daniel da Silva, Nikolai V. Pogorelov, Anne Spalding, Campbell Watson, Manil Maskey, Madhulika Guhathakurta, Juan Bernabe-Moreno, Rahul Ramachandran

Outline

Surya is a 366 million-parameter foundation model for solar physics. It is designed to learn a universal solar representation from multi-instrument SDO observations (including eight AIA channels and five HMI products). It employs a space-time converter architecture, spectral gating, and short- and long-range attention. It was pretrained on high-resolution solar image prediction tasks and optimized through autoregressive expansion fine-tuning. Zero-shot evaluation demonstrates its ability to predict solar dynamics and flare events, and downstream fine-tuning using LoRA demonstrates robust performance in solar wind prediction, active region segmentation, solar flare prediction, and EUV spectra. Surya is the first solar physics foundation model to use time progression from full-resolution SDO data as a pretext task.

Takeaways, Limitations

•
Takeaways:
◦
The first fundamental model in solar physics to utilize time progression as a pretext task using full-resolution SDO data.
◦
It shows strong performance in predicting various solar phenomena (solar dynamics, flares, solar wind, etc.) and analyzing them (active region segmentation, EUV spectrum, etc.).
◦
We demonstrate generalization ability to various solar phenomena through zero-shot and LoRA-based parameter-efficient fine-tuning.
◦
Suggests that the model can learn the fundamental physics of solar evolution.
•
Limitations:
◦
Limitations is not explicitly mentioned in the paper. Further research may be needed to further analyze the model's generalization performance and physical interpretability.
◦
Possibility of performance degradation due to data bias.
◦
Computational cost and interpretation difficulty due to the model's complexity.
👍
Made with Slashpage