A Simple Learning Approach for Endogenous Network Formation

We operationalize the theory associated with an economic model of network formation and conduct experiments to determine the feasibility of endogenous, dynamic network formation in multi-agent organizations. We develop a learningbased, decentralized network formation strategy that allows the agents to make network adaptation decisions based on past performance. We compare our method with a dynamic network formation process proposed in the economics literature that relies on a global computation over the entire network structure. Our findings demonstrate that local decisions based solely on prior experience perform as well as local decisions based on perfect knowledge and a global computation.