Influence of learning behavior on evolution in colonies of artificial organisms

摘要

In this paper, we add learning behavior to artificial evolution simulation and evaluate the effect of learning behavior. Each individual establishes its own neural network with its genetic information. To train artificial organisms, we used improved Hebb's law in each generation. Also, we propose a reward function to take reinforcement learning in complicated and dynamically-propagation algorithm using the error between the output and object patterns. However, Nolfi's work is limited to a simple environment case where the exact object pattern can be calculated for the behavior of artificial organism, worked on the cross evolution through competitions among groups of artificial organisms. The artificial organisms established their own neural network with its genetic information and they competed each other under a simple rule. Network is the behavior of the artificial organism. The individual level learning behavior used the back Gruau Through the cross evolution among the groups of artificial life, various competition strategies appeared. However, Gruau's work did not include the learning technique determined environment. When individual-level learning behavior was introduced, evolution of each simulation model got faster and the effectiveness of evolution was significantly improved.