ㅇ 제목 : 진화 게임 이론 소개

ㅇ 일시 : 2017.11.23.(목) 10:00 ~ 12:00

ㅇ 장소 : ETRI 7동 481호

ㅇ 강사 : 김은연 교수(한밭대학교)

ㅇ 요약 : Evolutionary Mathematical Game Theory

In the economic applications of game theory, the goal is to solve the games, to find the Nash equilibrium, and to find the evolutionary stable strategies. These properties of a game are supposed to permit the games, functioning as models of behavior, to make predictions about the behavior of agents that are engaged in the games. The problem is that human agents usually don't arrive at the predicted points. Even artificial agents with a modicum of complexity don't arrive at the stable points predicted by theory.

We go over evolutionary games and evolving their game playing strategies. The original application of evolutionary computation was to the optimization of solutions to fixed problems. This meant that, when we are determining the quality of structures used to drive selection, reproduction, and ultimately evolution, we are searching a fixed landscape that is in effect the graph of the function we are using to estimate quality. This goes out the window when we train agents to

play games by having them play against one another. The quality of an agent is measured by having it play other agents from those currently present in the population. Since this agent population is updated in each iteration of the algorithm, this means that the standard of quality changes along with the agent population. The standard form of evolution, against a fixed standard, is called evolution. The evolution of game playing agents against a changing population of other game playing agents is one type of co-evolution. Changing the rate at which the agent population is updated is enough to change the population of agents that result from evolution.

In this talk, the general mathematical game theory will be introduced and several games including the prisoner’s dilemma will be discussed.