Robust Watermark
Robust watermarking aims to embed imperceptible identifiers into digital content (images, text, audio, video, and AI models) to verify authenticity and ownership, resisting various attacks like editing, compression, and model extraction. Current research focuses on developing watermarking techniques robust to diverse perturbations, exploring methods across various modalities and model architectures (e.g., diffusion models, transformers, neural radiance fields), and establishing rigorous benchmarks for evaluating watermark strength and resilience. This field is crucial for addressing the growing concerns of copyright infringement, deepfakes, and malicious use of AI-generated content, impacting both intellectual property protection and the trustworthiness of digital information.
Papers
WAVES: Benchmarking the Robustness of Image Watermarks
Bang An, Mucong Ding, Tahseen Rabbani, Aakriti Agrawal, Yuancheng Xu, Chenghao Deng, Sicheng Zhu, Abdirisak Mohamed, Yuxin Wen, Tom Goldstein, Furong Huang
Probabilistically Robust Watermarking of Neural Networks
Mikhail Pautov, Nikita Bogdanov, Stanislav Pyatkin, Oleg Rogov, Ivan Oseledets