Study on Multi-agent Based Simulation of Team Machine Learning

Abstract In today's large-scaled distributed learning, it often involves a large number of machines. Coordination between them can be very complicated. In order to emphasize the importance of the organic relationships between machines, we introduce the organization theories of human society, such as cooperation and competition, to machine learning. We design two type of multi-agents along with their interaction rules, and then perform the simulation on Swarm platform. The dynamic processes of the simulated team machine learning are examined and the results show that, by elaborately designed interaction rules, the overall performance of team learning can be promoted dramatically and coordination structure of the machines can be optimized.