Reproduce “Neurally Integrated Finite Elements for Differentiable Elasticity on Evolving Domains” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Gilles Daviet, Tianchang Shen, Nicholas Sharp, David I. W. Levin
A differentiable elastic simulator is introduced that leverages neurally integrated quadrature within a Mixed Finite Element (FEM) framework to robustly simulate and optimize evolving implicit geometries. This approach facilitates end-to-end optimization of an object's shape, topology, and material properties, enabling physically-aware 3D reconstructions directly from image-based inputs.
Key insights
- Training a neural network to predict quadrature rules provides a continuous, differentiable, and computationally efficient method for numerical integration over complex, evolving implicit domains.
- The Mixed FEM formulation offers inherent robustness against large material stiffness variations and locking phenomena, which is crucial for optimization tasks where material properties may change significantly.
- Coupling the differentiable simulator with an implicit surface representation, like FlexiCubes, enables a holistic optimization framework that jointly refines geometry, topology, and material properties directly from image-based inputs.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.