Cheby-KANs Dataset

These are the codes and models used in our experiments regarding our submitted article “Cheby-KANs: Advanced Kolmogorov-Arnold Networks for Applying Geometric Deep Learning in Quantum Chemistry Applications”. The code is developed using python programming language. In our paper we haedeveloped the B-spline based KANs with a more powerful and much faster polynomials “shifted-Chebyshev polynomials” of the first kind. Also, we integrated our new architecture with geometric deeplearning to predict quantum properties. This was done by modifying the famous Schnet model by usingour new model rather than the ordinary multi-layer-perceptrons. Attached are the files for the codes and the pretrained models.

We are using python for our experiments, and make sure to install the latest version of Pytorch and Pytorch-Geometric to be ble to load the pretrained models to reproduce the results.

本数据集包含我们在投稿论文《Cheby-KANs：将几何深度学习（Geometric Deep Learning）应用于量子化学（Quantum Chemistry）场景的进阶科尔莫戈罗夫-阿诺德网络（Kolmogorov-Arnold Networks，KANs）》相关实验中所使用的代码与模型。本次实验代码基于Python编程语言开发。在本论文中，我们提出了基于B样条的KANs，并采用了性能更优异、运算速度更快的第一类移位切比雪夫多项式（shifted-Chebyshev polynomials）。此外，我们将所提出的新型架构与几何深度学习相结合，用于预测量子化学相关性质。具体实现方案为：将经典的Schnet模型中的普通多层感知机替换为我们提出的新型模型。本数据包附带了实验代码与预训练模型的相关文件。

本次实验基于Python开展，请务必安装最新版本的PyTorch与PyTorch几何（PyTorch Geometric），方可加载预训练模型并复现实验结果。