时间：2025年6月27日（周五） 14:00-17:00
地点：普陀校区理科大楼A1114
题目：Responsible Generative AI Deployment in High-Stakes Domains
主讲人：刘晓钟 伍斯特理工学院（WPI）计算机科学与数据科学系副教授
主持人：贺国秀
内容简介：Generative AI is reshaping critical sectors such as law, healthcare, drug development, and biomedical research, offering unprecedented potential to accelerate innovation and decision-making. However, the deployment of these powerful models in high-stakes environments introduces profound challenges related to trust, transparency, and safety. In contexts where even minor system failures can have severe or irreversible consequences, ensuring the reliability of generative models becomes paramount. This talk presents our recent work on advancing responsible AI practices, including integrating human feedback to improve model robustness, designing effective frameworks for human–AI collaboration, and developing novel techniques for model calibration and control. We will also explore the ethical and operational risks of generative AI in sensitive domains, highlighting the urgent need for safeguards that align model behavior with real-world safety and accountability demands.
报告人简介：Xiaozhong LIU（刘晓钟）博士是伍斯特理工学院（WPI）计算机科学与数据科学系的副教授。在此之前，他曾任印第安纳大学布卢明顿分校信息与计算机工程学院的副教授。他的研究兴趣包括自然语言处理（NLP）、文本与图挖掘、信息检索与推荐、元数据处理，以及计算社会科学。他在雪城大学完成的博士论文（导师为Elizabeth D. Liddy 教授）提出了一种创新的排序方法，利用动态社区兴趣对检索结果进行加权。近年来，他的研究重点包括人工智能的可解释性与可信性、基于深度学习的自然语言生成、网络犯罪与安全，以及以人为中心的数据异质性等方向。除学术研究外，他还积极参与产业界的项目。例如，他曾任阿里巴巴达摩院的高级顾问兼自然语言处理研究总监，主导多个前沿且具有高度市场驱动性的研究项目。
