KAN Explorer

Choose a function, train a KAN with B-spline functions on its edges, and compare its error and parameter count with an MLP trained on the same data.

Loading the lab… If it does not appear, make sure JavaScript is enabled in your browser.

Training runs in the browser with gradient descent on a single-variable function. The result is from a small experiment and does not generalize.

Architectures related to this lab

Advanced2024

Kolmogorov–Arnold Networks

Kolmogorov–Arnold Networks (KAN)

Networks that place learnable activation functions (B-spline curves) on the edges instead of fixed weights, while the nodes only sum their inputs, inspired by the Kolmog…

Advanced Neural Architectures