Natural Selection of Game Playing Agents

Kovács, Dániel László [Kovács, Dániel László (intelligens rends...), szerző] Méréstechnika és Információs Rendszerek Tanszék (BME / VIK)

Angol nyelvű Tudományos Konferenciaközlemény (Könyvrészlet)
    In this article we present an evolutionary simulation framework for the natural selection of game theoretic player agents. The framework is inspired by evolutionary game theory and the work of Robert Axelrod. Our incentive was to develop an intuitive, realistic implementation of these models where agents are represented individually, each having its own lifecycle, resources, properties and decision mechanism, and where selection emerges from their mutual interaction. Thus selection is not pre-programmed, but emerges naturally (as agents interact, win or lose, are born or die). Interaction among agents is modeled with general n-player games. The framework can be used to model real world multi-agent scenarios to predict the fittest agent programs responsible for selecting agents’ interaction strategies. Several experiments were conducted with generalized Hawk-Dove and Tragedy of the Commons games. Agents playing constantly, randomly, according to Nash equilibrium and a generalized Tit-for-Tat program were studied. Experiments indicate that some fundamental evolutionary game theoretic assumptions might be questioned.
    Hivatkozás stílusok: IEEEACMAPAChicagoHarvardCSLMásolásNyomtatás
    2021-05-11 18:27