We explore how interacting micro-mechanism with simple behavior lead to macro collective behavior in multi-agent system. We analyze the micro-interaction mechanism of multi-agent, from the perspective of macro collective behavior. We study the multi-agent system with learning function. We explore the symmetry breaking process of how homogeneous individual evolve into heterogeneous individual. We study the mutual trust mechanism and reputation mechanism of agent. We also study complexity of evolutionary algorithm.

Micro-mechanism of collective behavior

Cooperation behavior, widely existing in real system, is a typical manifestation of collective emergence behavior. We build evolutionary model with cooperation and specialization of agents, put forward the descriptive index and evolution rules of agents, and also conclude a symmetry breaking micro-mechanism with division of labor. At the same time, we study the evolutionary model with competitive cooperation, describe the measurement index and evolution rules of the competition and cooperation among agents, and conduct model simulation based on agent. We focus on the relation between information share level and macro collective behavior (formation of cooperation).

Complex adaptive system

A distinct manifestation of multi-agent system’s adaptability is to build trust and reputation. At the same time, the trust mechanism and reputation mechanism of agent group is also a major research area of P2P system. We study the trust mechanism and reputation mechanism of overall distributed multi-system. We introduce the approach of overall environmental information into P2P system to study the characteristic of non-scaling distribution of overall information. And it has been discovered that the mechanism that considers overall system information is better in quality than the original ones and also better meet real system’s demand on trust mechanism and reputation mechanism.  

Evolutionary algorithm

Evolutionary algorithm is a bionics algorithm that mimics natural genetic evolutionary laws. It can deal with the overall searching and optimization problem. It is also an important approach to study phylogeny. Our focus is to study the algorithm complexity of evolutionary algorithm. The study on the convergence complexity of existing evolutionary algorithm all use expected first passage time. As an average quantity, expected first passage time actually misses much information about the distribution of convergence time. Therefore, we introduce a concept that approximate to convergence in sense of probability, which could provide the whole time distribution. We use it, as a supplement of expected first passage time, to solve evolutionary algorithm problems that can apply hierarchical computation.