Adaptive Regularization for Nonconvex Optimization sing Inexact Function Values and Randomly Perturbed Derivatives

Stefania Bellavia, Gianmarco Gurioli, Benedetta Morini, Philippe TOINT

Résultats de recherche: Papier de travailArticle de travail

2 Téléchargements (Pure)

Résumé

  A regularization algorithm allowing random noise in derivatives and inexact
  function values is proposed for computing approximate local critical points
  of any order for smooth unconstrained optimization problems. For an
  objective function with Lipschitz continuous p-th derivative and given an
  arbitrary optimality order q <= p, it is shown that this algorithm will, in
  expectation, compute such a point in at most  
         O((min_{j=1,...,q}epsilon_j)^{-(p+1)/(p-q+1)})
  inexact evaluations of f and its derivatives whenever q in {1,2}, where
  \epsilon_j is the tolerance for j-th order accuracy.  This bound becomes at most
         O((min_{j=1,...,q}epsilon_j)^{-(q(p+1))/(p)})
  inexact evaluations if q>2 and all derivatives are Lipschitz
  continuous. Moreover these bounds are sharp in the order of the accuracy
  tolerances. An extension to convexly constrained problems is also outlined.
langue originaleAnglais
ÉditeurArxiv
Nombre de pages22
Volume2005.04639
Etat de la publicationPublié - 13 mai 2020

Empreinte digitale

Examiner les sujets de recherche de « Adaptive Regularization for Nonconvex Optimization sing Inexact Function Values and Randomly Perturbed Derivatives ». Ensemble, ils forment une empreinte digitale unique.

Contient cette citation