New Coevolution Dynamic as an Optimization Strategy in Group Problem Solving

TL;DR

This paper introduces a new coevolution dynamic with multiple rewirings in social agent networks, enhancing problem-solving efficiency in NK fitness landscapes by adjusting network rewiring strategies.

Contribution

It proposes a novel coevolution dynamic allowing multiple rewiring events, improving the optimization process in agent-based models solving complex landscapes.

Findings

More rewirings reduce computational cost on easy landscapes at low average degree.

Optimal rewirings (3-4 neighbors) improve global maxima search in medium-sized systems.

Effectiveness varies with landscape difficulty and network connectivity.

Abstract

Coevolution on social models couples the time evolution of the network with the time evolution of the states of the agents. This paper presents a new coevolution dynamic allowing more than one rewiring on the network. We explore how this coevolution can be employed as an optimization strategy for problem-solving capability of task-forces. We used an agent-based model study how this new coevolution dynamic can help a group of agents whose task is to find the global maxima of NK fitness landscapes. Each agent can replace more than one neighbors, and this quantity is a tunable parameter in the model. These rewirings is a way for the agent to obtain information from individuals that were not previously part of its neighborhood. Our results showed that this tunable coevolution can indeed produce gain on the computational cost under certain circunstances. At low average degree (using a random network to connect the agents) and on easy landscape, more rewirings return lower cost, helping the agents to find the global maximum faster. However, at larger average degree and difficult landscape, the effect is more complex. At medium size systems, the coevolution brings optimal results when 3 or 4 neighbors are replaced.