Graphical models and human causal learning

Tom Griffiths
University of California, Berkeley (UC Berkeley)

Graphical models can be extended into a formalism that can be used to represent not just probabilistic dependencies, but also the causal relationships that hold between a set of variables. This provides a way to represent and reason about causality, as well as the basis for exploring questions about causal learning. I will outline the basic ideas behind causal graphical models, and the approaches that can be taken to learning causal relationships from a statistical perspective.
This will include a brief discussion of the distinction between learning causal strength and learning causal structure, and the constraint-based and Bayesian approaches to structure learning.

Presentation (PDF File)

Back to Graduate Summer School: Probabilistic Models of Cognition: The Mathematics of Mind