Learning to Play Approximate Nash Equilibria in Games with Many Players
نویسنده
چکیده
We illustrate one way in which a population of boundedly rational individuals can learn to play an approximate Nash equilibrium. Players are assumed to make strategy choices using a combination of imitation and innovation. We begin by looking at an imitation dynamic and provide conditions under which play evolves to an imitation equilibrium; convergence is conditional on the network of social interaction. We then illustrate, through example, how imitation and innovation can complement each other; in particular, we demonstrate how imitation can help a population to learn to play a Nash equilibrium where more rational methods do not. This leads to our main result in which we provide a general class of large game for which the imitation with innovation dynamic almost surely converges to an approximate Nash, imitation equilibrium.
منابع مشابه
Learning in and about Games
We study learning in finitely repeated 2× 2 normal form games, when players have incomplete information about their opponents’ payoffs. In a laboratory experiment we investigate whether players (a) learn the game they are playing, (b) learn to predict the behavior of their opponent, and (c) learn to play according to a Nash equilibrium of the repeated game. Our results show that the success in ...
متن کاملImitation and the Emergence of Nash Equilibrium Play in Games with Many Players
Wemodel a learning dynamic in which players imitate and innovate. Of interest is to question whether Nash equilibrium play emerges, and if so, the role that imitation plays in this emergence. Our main result provides a general class of coordination game for which approximate Nash equilibrium play does emerge. Important conditions include that players imitate ‘similar’ individuals. The role of i...
متن کاملDeep Reinforcement Learning from Self-Play in Imperfect-Information Games
Many real-world applications can be described as large-scale games of imperfect information. To deal with these challenging domains, prior work has focused on computing Nash equilibria in a handcrafted abstraction of the domain. In this paper we introduce the first scalable endto-end approach to learning approximate Nash equilibria without any prior knowledge. Our method combines fictitious sel...
متن کاملOn Learning Algorithms for Nash Equilibria
Can learning algorithms find a Nash equilibrium? This is a natural question for several reasons. Learning algorithms resemble the behavior of players in many naturally arising games, and thus results on the convergence or nonconvergence properties of such dynamics may inform our understanding of the applicability of Nash equilibria as a plausible solution concept in some settings. A second reas...
متن کاملAspects of communication complexity for approximating Nash equilibria
Since it was shown that finding a Nash equilibrium is PPAD-complete [6], even for 2-player normal form games [2], a lot of attention has been given to -approximate Nash equilibria. Almost all results on -approximate Nash equilibria assume full knowledge of the game that is played. This thesis will focus on -approximate Nash equilibria in an uncoupled setup, players only have knowledge of their ...
متن کامل