We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

math.OC

Change to browse by:

References & Citations

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Mathematics > Optimization and Control

Title: Softplus Penalty Functions for Constrained Optimization

Authors: Stefan Meili
Abstract: Penalty functions are widely used to enforce constraints in optimization problems and reinforcement leaning algorithms. Softplus and algebraic penalty functions are proposed to overcome the sensitivity of the Courant-Beltrami method to strong objective function gradients. These penalty functions are shown to converge in one fifth as many iterations on problems with fifty optimized parameters and be less sensitive to objective function gradient and scaling than the Courant-Beltrami penalty function. The softplus and algebraic penalty functions developed in this work allow optimized solutions to be found in fewer iterations with greater control over constraint accuracy.
Subjects: Optimization and Control (math.OC)
Cite as: arXiv:2107.04541 [math.OC]
  (or arXiv:2107.04541v1 [math.OC] for this version)

Submission history

From: Stefan Meili [view email]
[v1] Fri, 9 Jul 2021 16:58:15 GMT (955kb,D)

Link back to: arXiv, form interface, contact.