Researchers have developed Norm-PCGrad, a novel method to improve the training of physics-informed neural networks (PINNs) and physics-informed Kolmogorov-Arnold Networks (PIKANs) when using domain decomposition. This technique addresses issues with conflicting gradients that arise from combining residual, boundary, and interface terms in the loss function. Norm-PCGrad demonstrates state-of-the-art accuracy across various 2D and 3D problems with minimal computational overhead. Additionally, the study proposes using separable architectures like SPINN to enhance computational efficiency within domain decomposition frameworks. AI
IMPACT Enhances the scalability and accuracy of scientific machine learning models for solving complex physical simulations.
RANK_REASON The cluster contains an arXiv preprint detailing a new algorithmic approach for scientific machine learning. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
