PUB - Publikationen an der Universität Bielefeld

In order to make machines more intelligent, it is inevitable to understand human-like cognitive development, in which adaptive, autonomous and progressive evolution of cognitive decision-making in interacting with the environment plays a key role. Inspired by enactive artificial intelligence and evolutionary sampling learning, a new cognitive development learning model termed evolutionary enactive learning (EEL) is proposed in this paper. The proposed model is constructed by extending the reinforcement learning framework and introducing the utility-selection theory to guide the coevolution of pattern representation and decision-making policies. Theoretical analysis on the model's validity of EEL is given. To further demonstrate the effectiveness of the proposed method, two simulated cognitive decision-making tasks are designed, in which pattern representation and decision-making must be jointly developed to achieve good cognitive performance. Our experimental results clearly demonstrate that the resulting learning process is rational and effective. Finally, we indicate that the proposed EEL could be readily further extended by introducing existing machine learning techniques to solve more practical applications.