报告题目 (Title)：Optimization and generalization analysis for shallow neural networks
（浅层神经网络的优化与泛化分析）
报告人 (Speaker)：顾亦奇 教授（电子科技大学）
报告时间 (Time)：2025年12月12日 (周五) 13:30
报告地点 (Place)：校本部F309教室
邀请人(Inviter)：涂一辉
主办部门：理学院数学系
报告摘要：
We study the behavior of stochastic gradient descent (SGD) in solving least-squares regression with shallow neural networks, including fully-connected neural networks, convolutional neural networks and physics-informed neural networks. Past work on this topic has been based on the over-parameterization regime, whose convergence may require the network width to increase vastly with the number of training samples. We perform new optimization and generalization analyses, showing that the training loss and expected risk can be reduced below any target accuracy without the overparameterization hypothesis.
