DP-KAN: Differentially Private Kolmogorov-Arnold Networks

TL;DR

This paper explores the application of the Kolmogorov-Arnold Network (KAN) in differentially private training, showing it can be made private with performance comparable to MLPs under privacy constraints.

Contribution

It introduces a method to apply differential privacy to KAN using DP-SGD and evaluates its effectiveness across multiple datasets.

Findings

KAN can be privatized with DP-SGD easily

Performance of KAN under privacy constraints is comparable to MLP

Accuracy deterioration due to privacy is similar for KAN and MLP

Abstract

We study the Kolmogorov-Arnold Network (KAN), recently proposed as an alternative to the classical Multilayer Perceptron (MLP), in the application for differentially private model training. Using the DP-SGD algorithm, we demonstrate that KAN can be made private in a straightforward manner and evaluated its performance across several datasets. Our results indicate that the accuracy of KAN is not only comparable with MLP but also experiences similar deterioration due to privacy constraints, making it suitable for differentially private model training.