Cycles in adversarial regularized learning - Inria - Institut national de recherche en sciences et technologies du numérique Access content directly
Conference Papers Year : 2018

Cycles in adversarial regularized learning

Abstract

Regularized learning is a fundamental technique in online optimization, machine learning, and many other fields of computer science. A natural question that arises in this context is how regularized learning algorithms behave when faced against each other. We study a natural formulation of this problem by coupling regularized learning dynamics in zero-sum games. We show that the system's behavior is Poincare recurrent, implying that almost every trajectory revisits any (arbitrarily small) neighborhood of its starting point infinitely often. This cycling behavior is robust to the agents' choice of regularization mechanism (each agent could be using a different regularizer), to positive-affine transformations of the agents' utilities, and it also persists in the case of networked competition (zero-sum polymatrix games).
Fichier principal
Vignette du fichier
1709.02738.pdf (1.26 Mo) Télécharger le fichier
Origin Files produced by the author(s)
Loading...

Dates and versions

hal-01643338 , version 1 (09-10-2018)

Identifiers

  • HAL Id : hal-01643338 , version 1

Cite

Panayotis Mertikopoulos, Christos Harilaos Papadimitriou, Georgios Piliouras. Cycles in adversarial regularized learning. SODA '18 - Twenty-Ninth Annual ACM-SIAM Symposium on Discrete Algorithms, Jan 2018, New Orleans, United States. pp.2703-2717. ⟨hal-01643338⟩
167 View
66 Download

Share

Gmail Mastodon Facebook X LinkedIn More