题  目：AI-Driven Copyright Protection for 3D Molecular Structures

内容简介：Generating molecules with AI has advanced scientific discoveries in designing new materials and drugs. However, it also raises concerns regarding copyright and biosecurity, in which the generation of harmful content highlights the urgent need to protect generated molecules. Protecting molecules is a new research topic since directly applying traditional protection methods destroys the structure of molecules and renders them ineffective. Additionally, molecules are so sensitive to noise, such that slight changes of atom positions may cause atom distances to exceed chemical bond lengths, leading to molecular instability. This study proposes a novel AI-based copyright protection method for 3D molecules, embedding copyright information as a watermark within the molecular structure to ensure protection and traceability. With the optimization ability of AI, the molecules’ stability can be guaranteed by limiting the maximum jitter distance of atoms, ensuring the validity of chemical bonds. The method extracts the geometric invariants features of the molecular structure by using AI, achieving robustness against post-processing (e.g., rotation and translation). Preliminary experimental results have demonstrated that this study can effectively protect 3D molecules while maintaining their functionality, marking a novel application of AI in scientific research.

报告人：胡润文

报告人简介：香港城市大学博士后，研究兴趣包括可逆信息隐藏、鲁棒水印及其应用。代表性成果有基于深度学习预测的可逆信息隐藏算法和可抗几何攻击的可逆鲁棒水印算法。在IEEE Transactions on Pattern Analysis and Machine Intelligence、IEEE Transactions on Image Processing、IEEE Transactions on Information Forensics and Security、IEEE Transactions on Circuits and Systems for Video Technology等期刊发表多篇原创性工作。

时  间：2025年4月3日（周四）下午14：00至17：00

地  点：石牌校区南海楼507会议室

热烈欢迎广大师生参加!

  信息科学技术学院

2025年4月1日