Bi-FORK
A generative physics surrogate modeling framework for high-dimensional bifurcation systems. It achieves one-to-many mapping through latent-space flow matching and repulsion-guided sampling, and can recover multiple solution branches in a single forward inference pass.
Timeline (last 90 days)
Bi-FORK is a study of a generative modeling framework for high-dimensional bifurcation systems.
Bi-FORK adopts two key designs, latent-space flow matching and repulsion-guided sampling, to achieve the capability of "one forward pass, multiple solution outputs."
Bi-FORK's evaluation covers three types of bifurcation systems: buckled beams (continuous bifurcation), mechanical metamaterials (discrete bifurcation), and Allen-Cahn phase separation (field-value bifurcation).
The discretization scale of Bi-FORK's evaluation cases reaches up to 260,000 points.
While recovering multimodal solution structures, Bi-FORK improves scalability by several orders of magnitude.
The Bi-FORK framework is still at the research stage and has not yet solved problems such as guaranteeing that generated solutions strictly satisfy physical conservation laws, quantifying the completeness of branch coverage, and extending to more complex spatiotemporal chaotic systems.
All Facts (6)
Bi-FORK is a study of a generative modeling framework for high-dimensional bifurcation systems.
50%UnverifiedBi-FORK adopts two key designs, latent-space flow matching and repulsion-guided sampling, to achieve the capability of "one forward pass, multiple solution outputs."
50%UnverifiedBi-FORK's evaluation covers three types of bifurcation systems: buckled beams (continuous bifurcation), mechanical metamaterials (discrete bifurcation), and Allen-Cahn phase separation (field-value bifurcation).
50%UnverifiedThe discretization scale of Bi-FORK's evaluation cases reaches up to 260,000 points.
50%UnverifiedWhile recovering multimodal solution structures, Bi-FORK improves scalability by several orders of magnitude.
50%UnverifiedThe Bi-FORK framework is still at the research stage and has not yet solved problems such as guaranteeing that generated solutions strictly satisfy physical conservation laws, quantifying the completeness of branch coverage, and extending to more complex spatiotemporal chaotic systems.
50%