A comparison-oracle algorithm couples comparison-based gradient direction estimation with normalized gradient descent and claims O(nD^2/epsilon^2 log(nD/epsilon)) queries for smooth strictly quasi-convex minimization, on the strength of an unattainable uniform gradient norm bound.
Zeroth-order nonconvex stochastic optimization: Handling constraints, high dimensionality, and saddle poi nts, Found Comput Math
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
math.OC 1years
2024 1verdicts
REJECT 1representative citing papers
citing papers explorer
-
On quasi-convex smooth optimization problems by a comparison oracle
A comparison-oracle algorithm couples comparison-based gradient direction estimation with normalized gradient descent and claims O(nD^2/epsilon^2 log(nD/epsilon)) queries for smooth strictly quasi-convex minimization, on the strength of an unattainable uniform gradient norm bound.