First-order Methods Almost Always Avoid Saddle Points

First-order Methods Almost Always Avoid Saddle Points

20 October 2017

Ioannis Panageas

Georgios Piliouras

Max Simchowitz

Michael I. Jordan

Benjamin Recht

Papers citing "First-order Methods Almost Always Avoid Saddle Points"

5 / 5 papers shown

Title
Loss Landscape of Shallow ReLU-like Neural Networks: Stationary Points, Saddle Escape, and Network Embedding Zhengqing Wu Berfin Simsek Francois Ged ODL 30 0 0 08 Feb 2024
Stochastic noise can be helpful for variational quantum algorithms Junyu Liu Frederik Wilde A. A. Mele Liang Jiang Jens Eisert Jens Eisert 8 33 0 13 Oct 2022
Training Generative Adversarial Networks with Adaptive Composite Gradient Huiqing Qi Fang Li Shengli Tan Xiangyun Zhang GAN 10 3 0 10 Nov 2021
Differentiable Game Mechanics Alistair Letcher David Balduzzi S. Racanière James Martens Jakob N. Foerster K. Tuyls T. Graepel 16 78 0 13 May 2019
The Loss Surfaces of Multilayer Networks A. Choromańska Mikael Henaff Michaël Mathieu Gerard Ben Arous Yann LeCun ODL 175 1,182 0 30 Nov 2014