Uncovering Latent Behaviors in Ant Colonies

Many biological systems exhibit collective behaviors that strengthen their adaptability to their environment, compared to more solitary species. Describing these behaviors is challenging yet necessary in order to understand these biological systems. We propose a probabilistic model that enables us to uncover the collective behaviors observed in a colony of ants. This model is based on the assumption that the behavior of an individual ant is a time-dependent mixture of latent behaviors that are specific to the whole colony. We apply this model to a large-scale dataset obtained by observing the mobility of nearly 1000 Camponotus fellah ants from six different colonies. Our results indicate that a colony typically exhibits three classes of behaviors, each characterized by a specific spatial distribution and a level of activity. Moreover, these spatial distributions, which are uncovered automatically by our model, match well with the ground truth as manually annotated by domain experts. We further explore the evolution of the behavior of individual ants and show that it is well captured by a second order Markov chain that encodes the fact that the future behavior of an ant depends not only on its current behavior but also on its preceding one.

Published in:
Proceedings of the 2016 SIAM International Conference on Data Mining, 9
Presented at:
2016 SIAM International Conference on Data Mining, Miami, Florida, USA, May 5-7, 2016

 Record created 2017-01-05, last modified 2018-09-13

Publisher's version:
Download fulltext

Rate this document:

Rate this document:
(Not yet reviewed)