From local growth to global optimization in insect built networks
Abstract
Social insect colonies build large net-like systems: gallery and trail networks. Many such networks appear to show near-optimal performance. Focusing on the network system inside termite nests we address the question how simple agents with probabilistic behaviour can control and optimize the growth of a structure with size several magnitude orders above their perceptual range. We identify two major classes of mechanisms: (i) purely local mechanisms, which involve the arrangement of simple motifs according to predetermined rules of behaviour and (ii) local estimation of global quantities, where sizes, lengths, and numbers are estimated from densities, concentrations, and traffic. Theoretical considerations suggest that purely local mechanisms work better during early network formation and are less likely to fall into local optima. On the contrary, estimation of global properties is only possible on functional networks and is more likely to work through pruning. This latter mechanism may contribute to restore network functionalities following unpredicted changes of external conditions or network topology. An analysis of the network properties of Cubitermes termite nests supports the role of both classes of mechanisms, possibly in interplay with environmental conditions acting as a template.
Domains
Machine Learning [cs.LG]Origin | Files produced by the author(s) |
---|