A new protocol called PDE-Dirichlet has been developed to address data heterogeneity in federated learning for scientific machine learning tasks involving partial differential equations (PDEs). This protocol quantifies client separation using optimal transport and establishes a link between allocation heterogeneity and optimization divergence. Experiments across various PDE tasks and neural operator families demonstrated that increased heterogeneity leads to greater solution distance and optimization dispersion, with the most significant impact observed in low-viscosity Burgers' equation. AI
IMPACT Introduces a standardized method for evaluating non-IID federated learning in scientific machine learning, potentially improving model robustness and generalization.
RANK_REASON Academic paper detailing a new protocol for federated learning. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
