Multiagent Diffusion and Distributed Optimization

Distributed problem solving by a multiagent system represents a promising approach to solving complex computational problems. However, many multiagent systems require certain degree of planning, coordination and negotiation to achieve the given goal. This paper presents a multiagent framework for tackling global optimization tasks inspired by diffusion in nature. The framework is designed for situations where agent communication must be kept to a minimal. Hence, complicated coordination and negotiation is not possible. Distributed agents in this framework share the common goal of finding the global optimal solution. They cooperate to achieve this common goal by sharing and updating a common belief that captures their estimation of the whereabouts of the optimal solution. To facilitate this, agents are naturally organized in families with a parent and its offsprings as members. This paper also presents an algorithm called Evolutionary Diffusion Optimization, which is implemented base on the proposed agent framework. Experimental results on some benchmark problems are presented together with performance comparison with a simulated annealing algorithm.