mintisan/awesome-kan

A comprehensive collection of KAN(Kolmogorov-Arnold Network)-related resources, including libraries, projects, tutorials, papers, and more, for researchers and developers in the Kolmogorov-Arnold Network field.

What it solves

This repository provides a centralized, organized collection of resources for Kolmogorov-Arnold Networks (KANs). KANs are a new type of neural network architecture designed as an alternative to traditional Multi-Layer Perceptrons (MLPs). While MLPs use fixed activation functions at nodes, KANs use learnable activation functions on the edges (weights), which can lead to better accuracy, faster neural scaling, and improved interpretability in tasks like data fitting and scientific discovery.

How it works

Instead of static weights, KANs use univariate functions (typically parametrized as splines) on their edges. This allows the network to learn the activation functions themselves during training. The repository acts as a curated index, categorizing the rapidly growing ecosystem of KAN research, including academic papers, software libraries, implementation projects, and tutorials.

Who it’s for

This list is designed for researchers and developers interested in exploring, implementing, or improving KAN-based architectures. It is a starting point for anyone looking to understand the current state of KAN research, compare them against MLPs, or find existing code implementations for specific applications like computer vision, time series analysis, or physics-informed machine learning.

Highlights

  • Comprehensive collection of academic papers covering KAN theory and applications.
  • Categorized list of software libraries and implementations, including Python and non-Python versions.
  • Links to specialized variants like Convolutional KANs (ConvKANs), Transformers (KAT), and physics-informed models.
  • Curated tutorials and video resources for learning the fundamentals of KANs.
  • Regular updates tracking the latest developments in the field.

Related

  • Project
  • Project
  • Project
  • Project