Evolutionary learning in networked multi-agent organizations

This study proposes a simple computational model of evolutionary learning in organizations informed by genetic algorithms. Agents who interact only with neighboring partners seek to solve a given problem. We explore the effects of task specialization (transmitters and innovators), organizational culture, and network topology on the efficiency of collective learning. Simulation results indicate that organizations without innovators tend to get stuck in suboptimal equilibria, regardless of organizational culture and network topology. The effect of organizational culture of cherishing the recombination of existing answers is positive if there are no innovators because it helps agents escape from suboptimality, and otherwise negative because too much creativity is introduced into organizations. We also find that agents in highly clustered networks reach local consensus more rapidly, whereas agents are more likely to find the right solution in small-world networks and random networks with ...

• Type: Conference
• Year: 2010
• Where: GECCO
• Authors: Jae-Woo Kim
• Keywords: GECCO 2010, Genetic Algorithms, Network Topology, Optimization, Organizational Culture