Heavy-ball methods with random starts provably escape saddle points via a new state-space mapping that allows larger steps than plain gradient descent.
On gradients of func- tions definable in o-minimal structures
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
math.OC 1years
2019 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Heavy-ball Algorithms Always Escape Saddle Points
Heavy-ball methods with random starts provably escape saddle points via a new state-space mapping that allows larger steps than plain gradient descent.