.png){kind=icon}
.png){kind=icon}
Return
Curvature-aware dynamic precision approach for physics-informed neural networks
.png){kind=icon}
OA
AI
Y
I
G
T
DOI：10.1016/j.neucom.2026.134698.png)
.png){kind=icon}
.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
• We propose a dynamic precision approach for training physics-informed neural networks and showcase its applicability on benchmark failure-mode equations. • We reuse L-BFGS curvature information to build a precision-switching controller. • The proposed dynamic approach preserves double precision, FP64-level, accuracy at lower cost. • The proposed controller is architecture-agnostic.
Keywords:
Physics-informed neural networks
Dynamic precision
Loss curvature signal
Computational cost
AI Summary
Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.
Journal
IF:
6.5
Papers:
2.5W
Citations:
6.5W
Organization
Cited Papers