基于自适应共生进化的自主体快速强化学习研究

Autonomous Agent Fast Reinforcement Learning Method Based On Adaptive Symbiotic Evolution

文章研究了一类基于自适应共生进化模型的自主体构造方法，提出了基于自适应共生进化算法（ＡＳＥ）的一种进化强化学习方法，该方法不仅可以高效地进行神经网络决策系统的设计，而且在多自主体组成的自律系统的群体行为进化中可快速学习而收敛，并且通过多样度调节维持了群体多样性，克服了未成熟收敛现象。研究者将该方法用于人工生物中求偶通讯规范的研究，通过在环境中的生存和行为学习，雌雄个体都可有效地学习到成功的求偶通讯规范。

A sort of autonomous agent construction method based on adaptive symbiotic evolutionary model was proposed, and by using adaptive symbiotic evolution algorithm (ASE) as an evolutionary reinforcement learning method, Which can design neural network decision making systems effectively. while at the same time this method can accelerate the learning and convergence speed in the behavior evolution problem of multi agent autonomous system, and it also prevents premature convergence and promotes population diversity. It uses this model in the study of artificial creature matching criterion, through an efficient living and behavior learning period in the environment, the male and female individuals will all be able to learn a successful matching communication strategy.