Kolmogorov-Arnold Network Autoencoders for High-Dimensional Data Representation

TMLR Paper6241 Authors

Abstract: In the era of high-dimensional data, traditional machine learning models often face challenges related to computational complexity, overfitting, and suboptimal feature representations. This paper introduces Kolmogorov-Arnold Network Autoencoders (KANAs), a novel framework that leverages Kolmogorov-Arnold Networks (KANs) to transform dimensionality reduction and data reconstruction. Through experiments across diverse datasets, KANA consistently demonstrates superior reconstruction fidelity and linear probing accuracy, establishing itself as a powerful and versatile tool for high-dimensional data processing. The proposed framework shows strong potential for applications in areas such as scientific modeling and data compression.

Submission Type: Regular submission (no more than 12 pages of main content)

Assigned Action Editor: ~Yingnian_Wu1

Submission Number: 6241