Learned Monkeys: Emergent Properties of Deep Reinforcement Learning Generated Networks

Abstract

Graph generation methods used in the study of animal social networks suffer from the inability to fully explain individual agent behaviors and values. However, recent advancements in complex networks and machine learning offer a novel way of artificially simulating network formations. We use deep reinforcement learning (DRL) to model the proximity network of white-faced capuchin monkeys. This process of constructing a graph provides insight into the unique, individual social strategies of a network’s agents depending on the initial DRL parameters. We generated a network to closely match the characteristics of the proximity network constructed from an observational dataset. For example, our model-generated graph and the observed graph consistently showed a few members with significantly higher betweenness centrality than all other members despite each agent starting with the same parameters.