AI RESEARCH
On the Generalization in Topology Optimization via Sensitivity-Conditioned Bernoulli Flow Matching
arXiv CS.AI
•
ArXi:2606.02179v1 Announce Type: cross Surrogate models for topology optimization (TO) exhibit highly variable out-of-distribution (OOD) generalization under distribution shifts such as changing loads or boundary conditions, yet the source of this variability remains unclear. We hypothesize that OOD performance is governed by how much information the conditioning signal preserves about the adjoint sensitivity (reduced gradient) that drives classical TO.