Title: Learning about our planet's climate through structure learning of Bayesian networks 

Abstract:
This talk discusses a new application of machine learning in climate science, namely 
learning "graphs of information flow". The key idea is to interpret large-scale 
atmospheric dynamical processes as information flow around the globe and to identify their 
pathways from observed data using causal discovery algorithms.

We use constraint-based structure learning of dynamic Bayesian networks for the causal 
discovery process, i.e. to learn the graphs of information flow.  The resulting graphs 
allow climate scientists to gain new insights, for example to identify subtle changes of 
the planet's dynamics or to evaluate climate models. We will introduce the basic method, 
discuss obstacles encountered and overcome, and present the first insights gained into 
earth's climate.