New method enables label-free training for finite-element surrogate models

A new research paper proposes a label-free training objective for finite-element surrogate models, utilizing discrete energy minimization. This method eliminates the need for reference solutions, directly using the assembled discrete potential energy as a training signal. The paper details identities that ensure this signal is exact for linear elastostatics, showing that minimizing discrete energy and supervised regression in the stiffness norm yield identical unique minimizers and gradients. It also includes a conditioning lemma bounding displacement error by energy gap and discusses limitations for elastodynamics. AI

IMPACT This research could streamline the development of surrogate models by removing the need for extensive labeled data, potentially accelerating scientific simulations.

RANK_REASON Research paper detailing a novel method for training AI models. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• alphaXiv
• arXiv
• CatalyzeX
• Chebyshev
• conjugate gradient method
• cs.LG
• DagsHub
• finite element method
• Focus Entertainment
• Gotit.pub
• Hugging Face
• linear elastostatics
• ScienceCast

• paper
• infra

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →