报告题目: Non-Convex Global Optimization as an Optimal Stabilization Problem

报告人：黄宇扬（帝国理工老虎机 ）

邀请人&主持人：黄辉

报告时间与地点：2月26日 4: 30pm， 腾讯会议：846-713-458

报告摘要:  Global optimization of non-convex functions is fundamental to machine learning and engineering design. For non-convex, continuous, and possibly non-smooth objective functions with multiple global minimizers, classical gradient-based methods lack global convergence guarantees. We reformulate the global minimization problem as a discounted infinite-horizon optimal control problem, where the value function of the associated Hamilton-Jacobi-Bellman equation serves as a globally informative surrogate for the objective landscape. Under minimal structural assumptions requiring neither convexity, differentiability, nor Łojasiewicz conditions—we establish explicit exponential convergence rates to global minimizers with computable constants. Numerical experiments on challenging non-convex landscapes demonstrate the practical effectiveness of this control-theoretic approach.

报告人介绍：Yuyang Huang is a PhD candidate in Applied Mathematics at Imperial College London, supervised by Professors Dante Kalise and Nikolas Kantas, supported by the Roth Scholarship. Her research focuses on non-convex optimization, computational optimal control, and Hamilton-Jacobi-Bellman equations, with applications to machine learning and agent-based modelling.