: Because the functions are univariate splines, they are easier for humans to visualize and understand, making KANs particularly useful for AI for Science . The pykan Library
: Nodes in a KAN simply sum the incoming signals; they do not have their own activation functions like ReLU or Sigmoid. kan.py
: It offers built-in plotting functions to visualize the "shape" of the learned functions on every edge, helping researchers "see" what the model has learned. Key Features and Limitations Description Language Built on Python and PyTorch. Efficiency : Because the functions are univariate splines, they
from kan import KAN import torch # Create a KAN with 2 inputs, 5 hidden neurons, and 1 output model = KAN(width=[2, 5, 1], grid=5, k=3) # Training follows a standard loop structure # model.train(dataset, opt="LBFGS", steps=20) Use code with caution. Copied to clipboard Key Features and Limitations Description Language Built on
). In a KAN, each connection is a small, learnable spline function (
The pykan repository, maintained by the original researchers, provides the tools to build, train, and visualize these networks.
A basic setup for a KAN involves importing the library and defining the layer structure: