KAN or MLP: A Fairer Comparison

TL;DR

This paper provides a comprehensive comparison between KAN and MLP models across multiple tasks, revealing MLP's general superiority except in symbolic formula tasks where B-spline activation helps KAN.

Contribution

It offers a controlled, fair comparison of KAN and MLP, highlighting the impact of B-spline activation and the forgetting issue in continual learning settings.

Findings

MLP outperforms KAN in most tasks.

B-spline activation improves MLP performance in symbolic tasks.

KAN exhibits more severe forgetting than MLP in continual learning.

Abstract

This paper does not introduce a novel method. Instead, it offers a fairer and more comprehensive comparison of KAN and MLP models across various tasks, including machine learning, computer vision, audio processing, natural language processing, and symbolic formula representation. Specifically, we control the number of parameters and FLOPs to compare the performance of KAN and MLP. Our main observation is that, except for symbolic formula representation tasks, MLP generally outperforms KAN. We also conduct ablation studies on KAN and find that its advantage in symbolic formula representation mainly stems from its B-spline activation function. When B-spline is applied to MLP, performance in symbolic formula representation significantly improves, surpassing or matching that of KAN. However, in other tasks where MLP already excels over KAN, B-spline does not substantially enhance MLP's performance. Furthermore, we find that KAN's forgetting issue is more severe than that of MLP in a standard class-incremental continual learning setting, which differs from the findings reported in the KAN paper. We hope these results provide insights for future research on KAN and other MLP alternatives. Project link: https://github.com/yu-rp/KANbeFair