Get the latest tech news
The Annotated Kolmogorov-Arnold Network (Kan)
An annotated guide to the Kolmogorov-Arnold Network
class KANConfig: """ Configuration struct to define a standard KAN. """ residual_std = 0.1 grid_size = 5 spline_order = 3 grid_range = [-1.0, 1.0] device = torch.device("cuda" if torch.cuda.is_available() else "cpu") If you understand how MLPs work, then the following architecture should look familiar. We will run through a simple example of training a KAN on the canonical MNIST handwritten digits dataset to show how easy it is to adapt these models for standard deep learning settings. !wget www.di.ens.fr/~lelarge/MNIST.tar.gz !tar -zxvf MNIST.tar.gz -C data/ In the interest of reusing the existing train logic we created earlier, we write a function to turn a torch.Dataset with MNIST into the dictionary format.
Or read this on Hacker News