Jump to content

Draft:Kolmogorov-Arnold Network (KAN)

From Wikipedia, the free encyclopedia


A Kolmogorov-Arnold Network (KAN) is a new class of feedforward neural networks inspired by the Kolmogorov–Arnold representation theorem. Unlike Multilayer Perceptrons (MLPs) which use a fixed activation function and weights at each neuron, KANs use learnable weight functions on their edges, modeled as splines.[1]




References[edit]

  1. ^ Liu, Ziming; Wang, Yixuan; Vaidya, Sachin; Ruehle, Fabian; Halverson, James; Soljačić, Marin; Hou, Thomas Y.; Tegmark, Max (2024-05-02). "KAN: Kolmogorov-Arnold Networks". arXiv:2404.19756 [cs.LG].