报告人：杜嘉伟 高级科学家
主持人：殷赵霞 教授
时间：2026年1月13日 16:00
地点：钉钉会议 322-980-606
报告人简介：
Dr. Jiawei Du is currently a Senior Scientist at the Agency for Science, Technology and Research (A*STAR), Singapore. His research focuses on data synthesis, optimization algorithms, and intelligent agent systems. He received his Ph.D. degree from the National University of Singapore (NUS) in 2023. In the past five years, he has published nearly 40 papers in these areas, including works in top-tier venues such as NeurIPS, ICLR, IEEE TPAMI, and CVPR. He has authored 10 papers as the first author. Dr. Du serves as an Area Chair for IJCAI and has continuously contributed as a reviewer for premier conferences including ICML, NeurIPS, ICLR, and CVPR.
报告内容介绍：
With the rapid advancement of Large Language Models, agent systems are evolving from simple conversational tools into action-oriented entities. However, their deployment in real-world environments faces two fundamental challenges. Traditional agents relying on the "instruction-following" paradigm effectively become static snapshots after deployment; they depend heavily on human priors and are unable to learn from execution, which limits their ability to autonomously evolve in dynamic environments. Furthermore, as agents move beyond these strict constraints to pursue autonomous optimization in open-ended settings, ensuring safety becomes critical, since increased autonomy without effective monitoring often introduces uncontrollable risks. Addressing these challenges, this talk argues that the future of agent systems lies not in elaborate pre-specification but in robust mechanisms for growth, exploring evolutionary strategies for experience-driven learning alongside defensive boundaries to reconcile autonomy with reliability.
