English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Tools annual

Global: Zendy Free Plan

Global: Zendy Tools monthly

Global: Zendy Plus monthly

Recently, the research on protecting the intellectual properties (IP) of deepneural networks (DNN) has attracted serious concerns. A number of DNN copyrightprotection methods have been proposed. However, most of the existingwatermarking methods focus on verifying the copyright of the model, which donot support the authentication and management of users' fingerprints, thus cannot satisfy the requirements of commercial copyright protection. In addition,the query modification attack which was proposed recently can invalidate mostof the existing backdoor-based watermarking methods. To address thesechallenges, in this paper, we propose a method to protect the intellectualproperties of DNN models by using an additional class and steganographicimages. Specifically, we use a set of watermark key samples to embed anadditional class into the DNN, so that the watermarked DNN will classify thewatermark key sample as the predefined additional class in the copyrightverification stage. We adopt the least significant bit (LSB) imagesteganography to embed users' fingerprints into watermark key images. Each userwill be assigned with a unique fingerprint image so that the user's identitycan be authenticated later. Experimental results demonstrate that, the proposedmethod can protect the copyright of DNN models effectively. On Fashion-MNISTand CIFAR-10 datasets, the proposed method can obtain 100% watermark accuracyand 100% fingerprint authentication success rate. In addition, the proposedmethod is demonstrated to be robust to the model fine-tuning attack, modelpruning attack, and the query modification attack. Compared with three existingwatermarking methods (the logo-based, noise-based, and adversarial frontierstitching watermarking methods), the proposed method has better performance onwatermark accuracy and robustness against the query modification attack.

Protecting the Intellectual Properties of Deep Neural Networks with an Additional Class and Steganographic Images