Abstract
Inspired by the Kolmogorov-Arnold representation theorem, we proposeKolmogorov-Arnold Networks (KANs) as promising alternatives to Multi-LayerPerceptrons (MLPs). While MLPs have fixed activation functions on nodes("neurons"), KANs have learnable activation functions on edges ("weights").KANs have no linear weights at all -- every weight parameter is replaced by aunivariate function parametrized as a spline. We show that this seeminglysimple change makes KANs outperform MLPs in terms of accuracy andinterpretability. For accuracy, much smaller KANs can achieve comparable orbetter accuracy than much larger MLPs in data fitting and PDE solving.Theoretically and empirically, KANs possess faster neural scaling laws thanMLPs. For interpretability, KANs can be intuitively visualized and can easilyinteract with human users. Through two examples in mathematics and physics,KANs are shown to be useful collaborators helping scientists (re)discovermathematical and physical laws. In summary, KANs are promising alternatives forMLPs, opening opportunities for further improving today's deep learning modelswhich rely heavily on MLPs.