A frame for modelling collective learning behaviors based on cellular automata

In this paper, we present a general frame for modelling and simulating collective learning behaviors of human groups with interactions. In our ldquobottom-uprdquo method, we assume that individuals learn to make their decisions repeatedly, following their studies on the performances of their former behaviors. Behavior of each individual is driven by temporal situation of the system and individualpsilas anticipation with respect to future decisions of other individuals, which can be calculated from individualspsila properties, such as strategy-sets, profit functions, beliefs about history information and future situation, and so on. Some learning rules, e.g. reinforcement learning, are also discussed for updating individualspsila properties during evolution of the system. Cellular Automata is employed in the frame to compute and observe the visualized long-term behaviors of the system at higher group level. Experiments show that the modelled collective systems with certain learning schemes can reach some balanced state under restrictive conditions. Our framework is helpful for study of modelling and simulating social dynamic systems, and researching higher-level behaviors of collective system with learning techniques.