Geometry of the Loss Landscape in Overparameterized Neural Networks:
Symmetries and Invariances
Berfin Simsek 1 2 Francois Ged 1 Arthur Jacot 1 Francesco Spadaro 1
Clement Hongler 1 * Wulfram Gerstner 2 * Johanni Brea 2 *
Abstract 1. Introduction
Neural network landscapes were traditionally thought of
as highly non-convex landscapes, where non-global crit ...


雷达卡




京公网安备 11010802022788号







