Complexity in a brain-inspired agent-based model

An agent-based model consists of a set of agents representing the components of a system. These agents interact with each other according to rules designed with knowledge of the system in mind. Although rules control the low-level interactions of agents, these models often exhibit emergent behavior at the system level. We apply the agent-based modeling framework to functional brain imaging data. In this model, agents are defined by network nodes and represent brain regions, and links representing functional connectivity between nodes dictate which agents interact. A link between two regions may be positive or negative, depending on the correlation in functional activity between the two regions. Agents are either active or inactive, and systematically update based on the activity of their immediate neighbors. Their dynamics are observed over a certain time period starting from predetermined initial configurations. While the information received by each node is limited by the number of other nodes connected to it, we have shown that this model is capable of producing emergent behavior dependent on global information transfer. Specifically, the system is capable of solving well-described test problems, such as the density classification and synchronization problems. The model is capable of producing a wide range of behaviors varying greatly in complexity, including oscillations with cycles ranging from a few steps to hundreds, and non-repeating patterns over hundreds of thousands of time steps. We believe this wide dynamic range may impart the potential for this system to produce a myriad of brain-like functional states.