Solution-space heterogeneity shapes federated learning dynamics across partial differential equations
Read the original on arXiv Machine Learning →The paper introduces a new federated learning protocol for partial differential equations called solution-space PDE-Dirichlet, which transforms continuous supervised responses into reusable solution bins and measures client separation via optimal transport. It establishes an exact inverse relationship between population allocation heterogeneity and Dirichlet concentration, and shows how response heterogeneity can cause gradient disagreement, local-update dispersion, and parameter divergence. Experiments on seven PDE tasks, three neural-operator families, and five random seeds demonstrate that lower concentration consistently increases solution distance and optimization heterogeneity, with the most pronounced error increase observed in low-viscosity Burgers equations.
Machine-generated by The Flow from the publisher's headline and feed description — not written or checked by a human. The full article lives at arXiv Machine Learning.