报告题目 Title：Machine Learning and Learning Systems

报告人 Speaker：Enrique Zuazua

报告人所在单位 Affiliation：Friedrich-Alexander-Universität Erlangen-Nürnberg

时间 Time：2027-01-11 10:00-11:30

地点 Venue：Room 2201, Guanghua Eastern Main Tower, Fudan University (Handan Campus)

报告摘要 Abstract：Machine learning is creating a new generation of systems that compute, learn, interact, and act. Understanding these systems, and making them reliable and secure, raises fundamental mathematical questions about approximation, inference, dynamics, optimization, and control. This colloquium will explore how applied mathematics helps us understand and design intelligent learning systems. AI safety and security provides an overarching theme, from understanding how errors propagate in adaptive systems to studying how communicated training updates can reveal private data. Control theory and inverse problems offer tools for identifying both the capabilities and the vulnerabilities of learning systems. I will also discuss neural networks as dynamical systems and generative AI through diffusion, time reversal, and transport. I will then turn to scientific machine learning and digital twins, where physical models and learned components evolve together through observations and feedback. The HYCO framework (Hybrid-Cooperative Learning) provides an approach in which physics-based and data-driven models learn through mutual interaction. This leads naturally to questions of stability, robustness, and the reliability of predictions and decisions. The lecture is intended for a broad mathematical audience. Its aim is to show how classical mathematics contributes to the foundations of trustworthy AI, while learning systems open new mathematical questions.

个人简介 Bio：Enrique Zuazua is an applied mathematician working at the interface of partial differential equations, control theory, numerical analysis, machine learning and AI. Since September 2019 he has held the Alexander von Humboldt Professorship and Chair for Dynamics, Control, Machine Learning and Numerics at Friedrich-Alexander-Universität Erlangen–Nürnberg (FAU), while also holding part-time appointments at Universidad Autónoma de Madrid and the University of Deusto in Bilbao. He also serves as Chief Algorithm Scientist at Sherpa.ai and is a member or fellow of several academies and scientific societies, including Jakiunde and Academia Europaea. He has received three ERC Advanced Grants NUMERIWAVES (2010), DyCon (2016), and CoDeFeL (2022), and the W. T. and Idalia Reid Prize from SIAM (2022).He was an invited section speaker at ICM 2006 in Madrid and a Special Section Lecturer at ICM 2026 in the areas of Control and Artificial Intelligence. His contributions are widely cited across mathematics and neighboring disciplines, and he holds leading editorial roles and serves on scientific committees for journals, centers, and funding agencies.

海报 Poster: Enrique Zuazua  Distinguished Lecture.jpg