Introduction

This repository contains the code required to reproduce the experimental results and plots for the paper "A Dynamic Framework for Grid Adaptation in Kolmogorov–Arnold Networks".

Getting Started

After cloning the repository,

git clone https://github.com/srigas/kan_grid.git grids
cd grids

create a Python virtual environment, activate it and install all dependencies:

python3 -m venv env
source env/bin/activate  # On Windows use: env\Scripts\activate
pip3 install -r requirements.txt

Note that the jaxKAN package is not installed via PyPI, but is instead provided locally, as this version includes code that (during release) is not yet available inside the package.

Then launch JupyterLab:

jupyter lab

The data for the synthetic functions described in the first experiment have already been generated using the benchmarks/generate_custom_dataset.py file.

The code for each family of experiments can be found in the corresponding Jupyter Notebook, with the synthetic function benchmarks in Benchmark 1 - Custom Functions.ipynb, the function regression on a subset of the Feynman dataset in Benchmark 2 - Feynman Subset.ipynb and the forward PDE problem in Benchmark 3 - Helmholtz PDE.ipynb. The corresponding results can be found in the results folder, which also includes the plots shown in the paper. The analysis of these results is handled via the Analysis.ipynb notebook, while the Example.ipynb notebook contains the code required to generate the plot shown in Figure 1 of the paper.

Citation

If the code and/or results presented in this work helped you for your own work, please cite our work as:

@article{kan_grid,
   title = {A Dynamic Framework for Grid Adaptation in Kolmogorov–Arnold Networks},
   author = {Rigas, Spyros and Papaioannou, Thanasis and Trakadas, Panagiotis and Alexandridis, Georgios},
   year = {2026},
   eprint = {2601.18672},
   archivePrefix = {arXiv},
   primaryClass = {cs.LG},
   url = {https://arxiv.org/abs/2601.18672}, 
}