Hessian-aware scalar scalings of the gradient yield a local unit step size guarantee and global convergence under weakened smoothness assumptions.
Complexity Guarantees for Nonconvex Newton-MR Under Inexact Hessian Information
1 Pith paper cite this work. Polarity classification is still indexing.
abstract
We consider an extension of the Newton-MR algorithm for nonconvex unconstrained optimization to the settings where Hessian information is approximated. Under a particular noise model on the Hessian matrix, we investigate the iteration and operation complexities of this variant to achieve appropriate sub-optimality criteria in several nonconvex settings. We do this by first considering functions that satisfy the (generalized) Polyak-\L ojasiewicz condition, a special sub-class of nonconvex functions. We show that, under certain conditions, our algorithm achieves global linear convergence rate. We then consider more general nonconvex settings where the rate to obtain first order sub-optimality is shown to be sub-linear. In all these settings, we show that our algorithm converges regardless of the degree of approximation of the Hessian as well as the accuracy of the solution to the sub-problem. Finally, we compare the performance of our algorithm with several alternatives on a few machine learning problems.
citation-role summary
citation-polarity summary
fields
math.OC 1years
2025 1verdicts
CONDITIONAL 1roles
background 1polarities
unclear 1representative citing papers
citing papers explorer
-
First-ish Order Methods: Hessian-aware Scalings of Gradient Descent
Hessian-aware scalar scalings of the gradient yield a local unit step size guarantee and global convergence under weakened smoothness assumptions.