T2S: A Rehearsal-Based Approach for Extraction-Resistant Model Watermarking

The paper introduces T2S, a rehearsal-based watermark embedding framework designed to enhance the robustness of model watermarks against extraction attacks. By simulating the extraction process and fine-tuning the watermark knowledge using a trigger set, T2S improves the transferability and detectability of watermarks in stolen models. This advancement is crucial for practitioners as it addresses the growing threat of model extraction, ensuring better protection of intellectual property in AI models.