2026年第71期(总第1215期)

演讲主题：AI-Assisted Discovery of Symbolic Laws and Decision Formulas

主讲人：周源 清华大学丘成桐数学科学中心副教授

主持人：关旭 供应链管理与系统工程系主任、教授

活动时间：2026年10月09日（周五）15:30-17:30

活动地址：管院大楼105教室

主讲人简介：

周源是清华大学丘成桐数学科学中心副教授。他于2009年在清华大学计算机科学与技术系获得工学学士学位，2014年在卡内基梅隆大学计算机科学系获得博士学位。在加入清华大学之前，他曾任麻省理工学院应用数学讲师，以及伊利诺伊大学厄巴纳-香槟分校和印第安纳大学伯明顿分校助理教授。

他的研究聚焦于数据驱动的决策与面向科学的人工智能，涵盖运筹学、机器学习与优化等方向。研究成果发表于Operations Research、Management Science、Mathematics of Operations Research、Production and Operations Management、SIAM Journal on Optimization、Nature Machine Intelligence、Journal of Machine Learning Research、ICML、NeurIPS、ICLR、COLT、STOC、FOCS、SODA等运筹学、管理科学、机器学习与理论计算机科学领域的顶级期刊与会议。他目前担任Operations Research与Operations Research Letters副主编。

活动简介：

This talk presents two applications of AI-driven symbolic search. Part 1 introduces PhyE2E, an end-to-end symbolic regression method that integrates physical dimensions, formula complexity, candidate operators, and constants. It learns dimensional consistency during formula synthesis, decomposes multivariate expressions via variable splitting, and refines locally with search algorithms. Experiments show substantial gains in symbolic and dimensional accuracy as well as formula simplicity. Applied to space physics, PhyE2E characterizes short- and long-term sunspot cycles, reveals a physically interpretable relationship governing near-Earth plasma pressure, and describes differential solar rotation at high latitudes where observations are limited. Part 2 applies symbolic search to operational decision-making, using AI to discover concise, interpretable, directly implementable decision formulas, and to analyze their theoretical properties toward establishing provable performance guarantees for the discovered decision rules.