Generating an Agent Taxonomy Using Topological Data Analysis

One of the challenges with the interpretability of large and complex multiagent simulations is understanding the kinds of agents that emerge from the interactions in the simulation, in terms of agent states and behaviors. We address one aspect of this challenge, which is to generate an agent taxonomy by analyzing the simulation outputs. We show that topological data analysis (TDA) can be used for this problem by applying it to agent trajectories, and present some promising results from the analysis of a large-scale disaster simulation. The results show a taxonomy of multiple types of agents that emerge, and which can be tracked over time through this taxonomical description.

Authors

• Samarth Swarup University of Virginia
• Reza Rezazadegan University of Virginia

Keywords

• multiagent simulation
• topology
• taxonomy
• simulation analytics