Bi-FORK: Generative Modeling of High-Dimensional Bifurcating Systems
Bi-FORK targets physical systems where one input can legitimately split into multiple solutions.
The paper frames that symmetry-breaking case as a failure mode for standard learned physical surrogates, which tend to assume a one-to-one map. Bi-FORK uses latent flow matching to generate full trajectories while keeping space and time coherence, then applies repulsion-guided sampling to recover distinct branches in one pass. The authors test it on buckling beams, mechanical metamaterials, and Allen-Cahn phase separation, with discretizations up to 260,000 points. ArXiv · AI/CL/LG's note
The paper frames that symmetry-breaking case as a failure mode for standard learned physical surrogates, which tend to assume a one-to-one map. Bi-FORK uses latent flow matching to generate full trajectories while keeping space and time coherence, then applies repulsion-guided sampling to recover distinct branches in one pass. The authors test it on buckling beams, mechanical metamaterials, and Allen-Cahn phase separation, with discretizations up to 260,000 points. ArXiv · AI/CL/LG's note
score 4