Exploring Vision-Based Models Using Kolmogorov-Arnold Networks: A Comparative Study

This study aims to integrate Kolmogorov Arnold Networks with vision based deep learning tasks. We hope to provide a comparative analysis between vision based deep learning models implemented using traditional methods and using Kolmogorov Arnold Networks. We have attempted this task specifically for two models: Image Classification and Image captioning. Our goal is to create these alternative KAN models and determine if they stand as a viable alternative. Our findings shed light on the scenarios where KAN's are applicable and adds to their capability in processing image data compared to pre-existing models further highlighting their value as an addition to the deep learning toolkit.