arxivcs.NEcs.AI2026-07-17
Evolutionary Algorithm-Guided LLMs for Physics-Informed Neural Network Design
Xu Yang, Mingyang Yu, Jing Xu, Keqian Li
Physics-informed neural networks (PINNs) are unusually sensitive to interacting choices of architecture, activation, loss weighting, collocation, optimization, and constraint enforcement. Large language models (LLMs) can propose these choices, but independent recommendations do n…