基于Q-Learning适应度函数的演化艺术  

Evolutionary Art based on Q-Learning fitness function

作　　者：仇颉[1] 

机构地区：[1]南京工业大学信息科学与工程学院,南京210009

出　　处：《微处理机》2010年第6期45-48,共4页Microprocessors

摘　　要：演化艺术通过演化计算演绎出众多妙趣横生的图片,使人们的生活更加多姿多彩。传统演化艺术系统,多采取基于树的表示法,由人工筛选父代基因优良的种群,经由交叉或突变操作,生成子代种群。自动演化艺术系统的出现,使用户从适应度函数的工作中解脱,系统可以自动寻求目标最优解。重点研究了基于图表示法的笛卡尔遗传规划(CGP),采用Q-Learning机器方法重新定义了选择策略,使得系统不仅可以独立寻找目标最优解,还能自动寻找到达最优解的最优路径。Artificial creative systems produced lots of interesting aesthetics photos.Based on ＂tree-based＂ genetic programming,traditional evolutionary Art system passes the good gens of the father photos to the offspring using mutation and crossover.The appearance of auto systems frees users from the work of fitness function.The programs can search for the optimizing solutions automatically.We start this paper with the theory of genetic programming and then compared the different between ＂tree-based＂ GP and ＂Graph-based＂ GP.An application of Q-Learning to the fitness function was proposed in this paper.That will take the jobs of humans during the evolve processes.The system values the actions of the agents according to the Q-values and find the best way to get the optimizing solutions all by itself.

关 键 词：演化艺术 笛卡尔遗传规划 再励学习 Q学习 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]