摘要： 人工生命进化模型设计的关键问题是学习与进化之间的关系，在自主体生存期内的学习过程可以通过不同的遗传方式指导个体行为的进化。该文利用进化算法和人工神经网络的研究方法，设计了两种不同的人工生命自主体的进化模型，模型解决了先天的遗传进化和后天的个体神经系统强化学习的有机结合问题，并且得出结论认为，强化学习有助于自主体适应复杂的外部环境，同时学习也可以直接或间接地使该适应性成为自主体遗传基因上的固定成分。

关键词: 人工生命, 强化学习, Baldwin效应

Abstract: One key problem in artificial life evolutionary model is the interaction between learning and evolution. The learning process that occurs over the lifetime of an individual can snide the evolution of that individual's species and its behavior through two distinctive methods: Baldwin Effect and Lamarckian Inheritance with different genetic propagation. We proposed two evolutionary models combined with reinforcement learning, and interpreted the importance of the changing role of learning and evolution in the whole process. By comparing the effectiveness of our models, we conclude that reinforcement learning helps individual adapting to complex environment, and learning can directly or indirectly make that adaptation become genetically fixed.

中图分类号: 

• TP18