Collective Exploration / Search

2011–2013  |  ongoing 

 

To examine how groups of fish navigate complex environments, I conducted a series of pilot experiments with schools of Golden Shiners. Groups were trained to the general location of a food reward in an empty tank. Then, a complex labyrinth was inserted, and fish had to search for a path to the reward. I tested different group sizes and compositions to see which performed best.

I hope to expand on these preliminary experiments along with theoretical models combining multi-agent reinforcement learning and game theory to assess when schools stay together, and under what conditions individuals tend to split off. Do free riders play a role in the fission/fusion dynamics seen in these experiments? Are there consistent strategies that emerge within groups in terms of exploration and exploitation of information?

A consistent strategy of following the crowd could allow some individuals to devote their attention to close observation of the environment, which may ultimately lead to their discovering novel routes, and their sudden departure from the group as they speed towards a (possible) reward. In addition to these questions regarding individual strategies and inter-group dynamics, I am also interested in analyzing this process of group exploration as a multi-agent form of Novelty Search, which may provide insights for the design of AI systems focused on collective search (either physical or virtual).