Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

SAFE-SVD: Sensitivity-Aware Fidelity-Enforcing SVD for Physics Foundation Models

About

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.

Chengjie Hong, Feixiang He, Yiheng Zeng, Lulu Kang, He Wang• 2026

Related benchmarks

TaskDatasetResultRank
Incompressible Navier-Stokes (NS-2D)VICON (test)
Base Loss0.51
58
Compressible Navier-Stokes (Comp-2D)VICON (test)
Base Loss0.75
51
NS-2DVicon
Base Loss0.51
44
2D Diffusion-Reaction SimulationDiffRe-2D
Base Loss3.06
29
Compressible Euler Equations (Euler-2D)VICON (test)
Base Loss2.55
29
Compressible Navier-Stokes SimulationCompNS
Base Loss0.04
29
Shallow Water Equations SimulationSWE
Base Loss17.22
29
Multi-physics SimulationPoseidon NS-BB (test)
Base Loss0.00e+0
25
Multi-physics SimulationPoseidon CE-RPUI (test)
Base Loss0.00e+0
25
Multi-physics SimulationPoseidon Wave-Layer (test)
Base Loss0.00e+0
25
Showing 10 of 16 rows

Other info

Follow for update