Open AccessSearch for a saddle point of a convex-concave stochastic game by the adaptive method of mirror descent.
Author(s) -
Alexander Nazin
Publication year - 2020
Publication title -
trudy kolʹskogo naučnogo centra ran
Language(s) - English
Resource type - Journals
ISSN - 2307-5252
DOI - 10.37614/2307-5252.2020.8.11.025
Subject(s) - saddle point , saddle , descent (aeronautics) , stochastic gradient descent , regular polygon , mathematical optimization , zero sum game , point (geometry) , computer science , function (biology) , convex function , gradient descent , mathematics , artificial intelligence , nash equilibrium , geometry , engineering , aerospace engineering , evolutionary biology , artificial neural network , biology
A stochastic game problem of 2 persons with a zero sum is considered, leading to the search for a saddle point of the game function based on the gradient approach. We study mirror descent algorithms, both adaptive and non-adaptive. The main results are proved. An illustrative example is discussed.