Watermarking Kolmogorov-Arnold Networks for Emerging Networked Applications via Activation Perturbation

Fuente: arXiv
Enregistré dans:
Détails bibliographiques
Auteurs principaux: Lu, Chia-Hsun, Wu, Guan-Jhih, Ho, Ya-Chi, Shen, Chih-Ya
Format: Preprint
Publié: 2025
Sujets:
Accès en ligne:
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866916889891962880
author Lu, Chia-Hsun
Wu, Guan-Jhih
Ho, Ya-Chi
Shen, Chih-Ya
author_facet Lu, Chia-Hsun
Wu, Guan-Jhih
Ho, Ya-Chi
Shen, Chih-Ya
contents With the increasing importance of protecting intellectual property in machine learning, watermarking techniques have gained significant attention. As advanced models are increasingly deployed in domains such as social network analysis, the need for robust model protection becomes even more critical. While existing watermarking methods have demonstrated effectiveness for conventional deep neural networks, they often fail to adapt to the novel architecture, Kolmogorov-Arnold Networks (KAN), which feature learnable activation functions. KAN holds strong potential for modeling complex relationships in network-structured data. However, their unique design also introduces new challenges for watermarking. Therefore, we propose a novel watermarking method, Discrete Cosine Transform-based Activation Watermarking (DCT-AW), tailored for KAN. Leveraging the learnable activation functions of KAN, our method embeds watermarks by perturbing activation outputs using discrete cosine transform, ensuring compatibility with diverse tasks and achieving task independence. Experimental results demonstrate that DCT-AW has a small impact on model performance and provides superior robustness against various watermark removal attacks, including fine-tuning, pruning, and retraining after pruning.
format Preprint
id arxiv_https___arxiv_org_abs_2508_06676
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Watermarking Kolmogorov-Arnold Networks for Emerging Networked Applications via Activation Perturbation
Lu, Chia-Hsun
Wu, Guan-Jhih
Ho, Ya-Chi
Shen, Chih-Ya
Machine Learning
With the increasing importance of protecting intellectual property in machine learning, watermarking techniques have gained significant attention. As advanced models are increasingly deployed in domains such as social network analysis, the need for robust model protection becomes even more critical. While existing watermarking methods have demonstrated effectiveness for conventional deep neural networks, they often fail to adapt to the novel architecture, Kolmogorov-Arnold Networks (KAN), which feature learnable activation functions. KAN holds strong potential for modeling complex relationships in network-structured data. However, their unique design also introduces new challenges for watermarking. Therefore, we propose a novel watermarking method, Discrete Cosine Transform-based Activation Watermarking (DCT-AW), tailored for KAN. Leveraging the learnable activation functions of KAN, our method embeds watermarks by perturbing activation outputs using discrete cosine transform, ensuring compatibility with diverse tasks and achieving task independence. Experimental results demonstrate that DCT-AW has a small impact on model performance and provides superior robustness against various watermark removal attacks, including fine-tuning, pruning, and retraining after pruning.
title Watermarking Kolmogorov-Arnold Networks for Emerging Networked Applications via Activation Perturbation
topic Machine Learning
url https://arxiv.org/abs/2508.06676