Entropic Penalties in Finite Games

The main objects here are finite-strategy games in which entropic terms are subtracted from the payoffs. After such subtraction each Nash equilibrium solves an explicit, unconstrained, nonlinear system of smooth equations. That system, while characteristic of perturbed best responses, is amenable in computation. It also facilitates analysis of fictitious play, learning by reinforcement, and evolutionary dynamics. Key words: finite games, Nash equilibrium, fictitious play, stimulusresponse, gradient methods, evolutionary dynamics, entropy.

• Type: Journal
• Year: 2005
• Where: ANOR
• Authors: Sjur Didrik Flåm, E. Cavazzuti
• Keywords: ANOR 2005, Evolutionary Dynamics, Fictitious Play, Nash Equilibrium