Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
Bi-FORK addresses the challenge of modeling high-dimensional physical systems exhibiting symmetry-breaking bifurcations, where standard generative models fail due to their one-to-one assumption. The framework employs latent flow matching for coherent trajectory generation and repulsion-guided sampling to recover multiple distinct solution branches in a single pass. Evaluated on diverse physical phenomena, Bi-FORK successfully captures multimodal solution structures and scales significantly beyond existing methods for these complex systems.
Generative models can now capture the full spectrum of solutions in high-dimensional physical systems with symmetry-breaking bifurcations, a feat previously intractable.
Bifurcations are ubiquitous in physical systems, from structural buckling to fluid and climate dynamics, yet they remain largely unexplored in deep learning. At a symmetry-breaking bifurcation, a single input admits multiple equally valid solutions, violating the one-to-one assumption underlying most learned physical surrogates. We introduce Bi-FORK, a generative framework for learning these one-to-many solution maps in high-dimensional systems. Bi-FORK generates complete trajectories through latent flow matching, preserving space and time coherence, and uses repulsion-guided sampling to recover distinct solution branches in a single amortized pass. We evaluate Bi-FORK on buckling beams, mechanical metamaterials, and Allen-Cahn phase separation, spanning continuous, discrete, and field-valued bifurcations with discretizations up to 260,000 points. Bi-FORK recovers the multimodal solution structure while scaling several orders of magnitude beyond prior approaches, opening generative modeling to high-dimensional bifurcating physical systems.