Explore a 16-minute IEEE presentation on an innovative approach to tracing text provenance using data hiding techniques. Delve into the concept of an Adversarial Watermarking Transformer, understanding its potential applications and implications. Learn about the envisioned scenario, model architecture, and background models that form the foundation of this research. Examine the experimental setup, including training methodologies and evaluation metrics. Discover the verification process and analyze qualitative examples that demonstrate the effectiveness of the proposed system. Investigate the robustness of the watermarking technique against both basic and advanced attacks. Gain insights into the future of text provenance tracing and its significance in the era of AI-generated content.
Adversarial Watermarking Transformer: Towards Tracing Text Provenance with Data Hiding
Overview
Syllabus
Introduction
Our envisioned scenario
Our model
Background models
Experimental setup
Training
Evaluation
Verification
Qualitative examples
Robustness
Basic Attacks
Advanced Attacks
Conclusion
Taught by
IEEE Symposium on Security and Privacy
