Learning in dynamic hierarchical network structures in complex systems: A study on the effect of learning in improving resilience in large scale social networks

A Cultural Algorithm (CA) framework was developed for the Mesa Verde Village multi-agent simulation in Swarm. Going beyond the basic farming resources of agents, we implemented two main social networks: a kinship relation network for generalized reciprocal exchange, and an economic network for balanced reciprocal exchange. Agents, or households, are able to procure several resources. They include agriculture, the hunting of deer, rabbits, and hares, the collection of wood for fuel, and acquisition of water. Individuals can exchange surplus good for needed goods through the exchange network.; Intelligent artifacts that describe the attributes of an agent as they participate in these networks can be produced and reside in the cultural space. One such intelligent artifact that can be produced relates to an individual agents' reputation. Reputation can be modified through the individuals' participation in each of the two exchange networks. Acquired reputation can be used to determine how new communal networks form within the community. Agents can learn how to interact with others within the network based upon reputation as well as other factors. Agents can learn both procurement strategies in conjunction with exchange strategies in order to survive.; The agent can learn an individual procurement and exchange strategy based on its own experience under dynamic environmental conditions. The result of individual learning can be generalized at a global level using a Cultural Algorithm. The impact that these new extensions to the model have on the systems' overall resiliency and reliability are then examined and compared to earlier versions without these additional features. Overall, the emerging communities overtime will attain certain measured levels of quality of life. Low quality of life will present motive for the population to move outside the study area.