Time：2024年7月11日（星期四）14:30-16:00,Thursday July 11, 2024
Venue：E4-233, Yungu Campus
Host：Chuanhao Wei, ITS
Speaker：Tao Luo, Shanghai Jiao Tong University
Title：Geometry and Local Recovery of Global Minima of Two-layer Neural Networks at Overparameterization
Abstract:In this talk, we investigate the geometry of the loss landscape for two-layer neural networks in the vicinity of global minima. Utilizing novel techniques, we demonstrate: (i) how global minima with zero generalization error become geometrically separated from other global minima as the sample size grows; and (ii) the local convergence properties and rate of gradient flow dynamics. Our results indicate that two-layer neural networks can be locally recovered in the regime of overparameterization.
