学习对进化的影响研究及仿真验证  

How Learning Affects Evolution:Research and Simulation Validation

作　　者：张家奇[1] 陈启军[1] 

机构地区：[1]同济大学电子与信息工程学院,上海201804

出　　处：《系统仿真学报》2007年第24期5849-5851,5855,共4页Journal of System Simulation

基　　金：国家自然科学基金(60643001);教育部新世纪优秀人才计划和上海市曙光计划项目(04SG22)。

摘　　要：研究了学习与进化的关系,特别是学习如何影响进化。基于Hinton和Nowlan在学习与进化研究领域的第一个计算模型,验证了学习能够指导进化。相对于Hinton和Nowlan较理想化的结果,得到了一个更加客观的结果。通过分析两个实验结果中需要学习来确定的基因所占比率不同,重新设计了系统新的适应度函数,证明了学习指导下进化的具体发展方向:进化向着完成这种学习能力的基因方向发展,学习到的性状最终按照达尔文的进化机制被遗传。The relationship between learning and evolution was discussed, especially how learning affects evolution. Based on the original computation model in learning and evolution by Hinton and Nowlan, it is proved learning can guide evolution. By contrast with Hinton and Nowlan＇s idealized results, more objective results were acquired. To analyze the reason why the ratio of the genes, which need learning to change, is different, a new fitness function was put forward. And it is proved when there is a learning which can guide the evolution, the evolution should go in the way to complete the genes which the learning need, that is, learned traits can be passed to the offspring by a Darwinian mechanism.

关 键 词：学习 进化 遗传算法 鲍德温效应 

分 类 号：TP273.3[自动化与计算机技术—检测技术与自动化装置]