Saved in:
Bibliographic Details
Main Authors: Hong, Chengjie, He, Feixiang, Zeng, Yiheng, Kang, Lulu, Wang, He
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2605.17985
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917507658416128
author Hong, Chengjie
He, Feixiang
Zeng, Yiheng
Kang, Lulu
Wang, He
author_facet Hong, Chengjie
He, Feixiang
Zeng, Yiheng
Kang, Lulu
Wang, He
contents We propose a new method for compressing physics foundation models (PFMs) which is a new trend in AI for Science. While model compression is essential for reducing memory use and accelerating inference in large foundation models, it remains under-explored for PFMs, where preserving physical fidelity is crucial. The challenge lies in the functional nature of physics data, where partial derivatives encode spatiotemporal dynamics and exhibit high sensitivity to compression. Conventional compression methods ignore this structure, often causing severe performance degradation or failure. To address this, we introduce a sensitivity-aware fidelity-enforcing compression framework that explicitly models loss-aware layer sensitivity in the output function space during compression. This provides a new route to compressing scientific foundation models while preserving accuracy and physical fidelity. Experiments show substantial gains over existing methods across multiple models and datasets, achieving significantly higher compression ratios while maintaining accuracy, in some cases by orders of magnitude. More broadly, the work potentially leads to a new subfield of efficient, deployable, and sustainable scientific foundation models in AI for Science.
format Preprint
id arxiv_https___arxiv_org_abs_2605_17985
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle SAFE-SVD: Sensitivity-Aware Fidelity-Enforcing SVD for Physics Foundation Models
Hong, Chengjie
He, Feixiang
Zeng, Yiheng
Kang, Lulu
Wang, He
Machine Learning
Artificial Intelligence
We propose a new method for compressing physics foundation models (PFMs) which is a new trend in AI for Science. While model compression is essential for reducing memory use and accelerating inference in large foundation models, it remains under-explored for PFMs, where preserving physical fidelity is crucial. The challenge lies in the functional nature of physics data, where partial derivatives encode spatiotemporal dynamics and exhibit high sensitivity to compression. Conventional compression methods ignore this structure, often causing severe performance degradation or failure. To address this, we introduce a sensitivity-aware fidelity-enforcing compression framework that explicitly models loss-aware layer sensitivity in the output function space during compression. This provides a new route to compressing scientific foundation models while preserving accuracy and physical fidelity. Experiments show substantial gains over existing methods across multiple models and datasets, achieving significantly higher compression ratios while maintaining accuracy, in some cases by orders of magnitude. More broadly, the work potentially leads to a new subfield of efficient, deployable, and sustainable scientific foundation models in AI for Science.
title SAFE-SVD: Sensitivity-Aware Fidelity-Enforcing SVD for Physics Foundation Models
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2605.17985