The following publications are possibly variants of this publication:
- Efficiently avoiding saddle points with zero order methods: No gradients requiredEmmanouil-Vasileios Vlatakis-Gkaragkounis, Lampros Flokas, Georgios Piliouras. nips 2019: 10066-10077 [doi]
- Riemannian stochastic optimization methods avoid strict saddle pointsYa-Ping Hsieh, Mohammad Reza Karimi Jaghargh, Andreas Krause 0001, Panayotis Mertikopoulos. nips 2023: [doi]
- Accelerated Multiplicative Weights Update Avoids Saddle Points Almost AlwaysYi Feng, Ioannis Panageas, Xiao Wang 0036. IJCAI 2022: 1811-1817 [doi]
- First-order methods almost always avoid strict saddle pointsJason D. Lee, Ioannis Panageas, Georgios Piliouras, Max Simchowitz, Michael I. Jordan, Benjamin Recht. mp, 176(1-2):311-337, 2019. [doi]
- First-order methods almost always avoid saddle points: The case of vanishing step-sizesIoannis Panageas, Georgios Piliouras, Xiao Wang. nips 2019: 6471-6480 [doi]