Local Convergence of Gradient Methods for Min-Max Games: Partial Curvature Generically Suffices
Local Convergence of Gradient Methods for Min-Max Games: Partial Curvature Generically Suffices
About this item
Full title
Author / Creator
Publisher
Ithaca: Cornell University Library, arXiv.org
Journal title
Language
English
Formats
Publication information
Publisher
Ithaca: Cornell University Library, arXiv.org
Subjects
More information
Scope and Contents
Contents
We study the convergence to local Nash equilibria of gradient methods for two-player zero-sum differentiable games. It is well-known that such dynamics converge locally when \(S \succ 0\) and may diverge when \(S=0\), where \(S\succeq 0\) is the symmetric part of the Jacobian at equilibrium that accounts for the "potential" component of the game. W...
Alternative Titles
Full title
Local Convergence of Gradient Methods for Min-Max Games: Partial Curvature Generically Suffices
Authors, Artists and Contributors
Author / Creator
Identifiers
Primary Identifiers
Record Identifier
TN_cdi_proquest_journals_2820822835
Permalink
https://devfeature-collection.sl.nsw.gov.au/record/TN_cdi_proquest_journals_2820822835
Other Identifiers
E-ISSN
2331-8422
