PARENT SESSION
Oral Session 16: Statistics and Biometrics.
Presiding: E Garton and T Simons
Monday, August 2, 8:00 AM to 11:30 AM, Meeting Room D 139.

Using machine learning to learn nature: Bayesian Learning Networks and the philosophy of studying communities.

McMahon, Sean^*,1, ¹ University of Tennessee - Knoxville, Knoxville, TN, US

ABSTRACT- The community scale - interactions between populations and their biotic and abiotic environment - constitutes the nexus between ecological patterns, evolutionary trajectories, and the arena of nature's disruption due to human activities. A fundamental challenge to ecologists lies in the fact that the community scale has often proved resistant to predictive understanding of its behaviors because communities contain many components with complex and dynamic relationships. There have been calls to abandon this scale as idiosyncratic, a mire into which scientific efforts are wasted. This view is a lamentable overreaction to the challenges and potential of studying communities. Bayesian Learning Networks (BLNs), a 15 year old member of the statistical family of graphical models, have been developed and refined largely by scientists that work with artificial intelligence and machine learning. Because BLNs map the relationships between multiple variables with uncertain dependencies, and can incorporate many variable structures, they are well-suited for the measurement of ecological communities and offer a method of predicting system behaviors. This talk gives a brief introduction to BLNs, describes a simple application of this method to a forest understory plant population, and explores how a BLN model can offer inference into the various direct and indirect pathways that may affect a forest community, with implications for evolutionary hypotheses, population structure, and conservation.

Key words: community ecology, graphical models, bayesian learning networks, statistical methods