Performance-lossless Black-box Model Watermarking

arXiv:2312.06488v2 Announce Type: replace
Abstract: With the development of deep learning, high-value and high-cost models have become valuable assets, and related intellectual property protection technologies have become a hot topic. However, existing model watermarking work in black-box scenarios mainly originates from training-based backdoor methods, which probably degrade primary task performance. To address this, we propose a branch backdoor-based model watermarking protocol to protect model intellectual property, where a construction based on a message authentication scheme is adopted as the branch …

address arxiv assets backdoor box cost cs.cr deep learning development high hot hot topic intellectual property performance property protection task technologies topic training value watermarking work