{"id":"arxiv/2606.14708","url":"https://arxiv.org/abs/2606.14708","tags":["math.OC","cs.AI"],"score":0,"title":"PH-KAN: Port-Hamiltonian Kolmogorov-Arnold Network","domain":"math.OC","source":"arxiv","authors":["Achraf El Messaoudi","Karim Cherifi","Yann Le Gorrec","Yongxin Wu"],"reviews":0,"subject":"math.OC","summary":"Data-driven machine learning approaches have become increasingly attractive for nonlinear system identification, but standard models often fail to preserve the underlying physical structure and remain difficult to interpret, especially when no analytical model is available. In this context, port-Hamiltonian (pH) models provide a natural physics-informed representation. However, when these models are parameterized with standard multilayer perceptrons (MLPs), the learned constitutive components often remain poorly interpretable. In this paper, we propose a structure-preserving identification…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-04-07","reactions":{}}