This repository implements Convolutional KANs (CKAN), a novel deep learning architecture that extends the Kolmogorov-Arnold Network paradigm to convolutional layers. Unlike standard CNNs that use fixed linear transformations, CKANs apply learnable non-linear basis functions at each spatial location during convolution. The project provides Jupyter notebook implementations building on the efficient-kan library, with full mathematical justification and experiments detailed in an associated arXiv paper.