报告题目：Energy-Stable Accelerated Optimization using Scalar Auxiliary Variables
报告人：毛志平，宁波东方理工大学
报告时间：2026年4月27日（周一），15:00-18:00
报告地点：长安大学理学院楼308
报告摘要：We propose a family of novel optimization algorithms by using the Scalar/Vector Auxiliary Variable (SAV) method. By reformulating the optimization process as ordinary differential equations, we construct Lyapunov energy functions that enables systematic incorporation of the SAV/VAV framework, leading to energy-stable discrete schemes. We also develop the adaptive extensions. Additionally, to enhance global exploration in highly nonconvex landscapes, we design a stochastic variant, combining ideas from Langevin dynamics and simulated annealing. The energy stability is guaranteed for all proposed methods. Extensive experiments on benchmark functions and deep learning tasks show that, overall, our methods outperform state-of-the-art optimizers in convergence speed, accuracy, and stability.
报告人简介：毛志平教授2009年本科毕业于重庆大学，2015年博士毕业于厦门大学计算数学专业，国家高层次青年人才，2015年10月至2020年9月在美国布朗大学应用数学系从事博士后研究，国家级青年人才计划入选者。毛志平教授主要从事深度学习与偏微分方程数值解，特别是谱方法研究以及深度学习求解复杂系统方面的研究，其目前在SIREV, JCP， SISC， SINUM、 CMAME等国际高水平杂志上发表论文 40 余篇。
