GroupeDeLecture090121RS

Causal Structure Search: Philosophical Foundations and Problems

Richard Scheines, Department of Philosophy, Carnegie Mellon University, USA

Abstract: We briefly discuss the recent history of causal discovery and the core philosophical foundations underlying Causal Bayes Nets: the Causal Markov Axiom and Faithfulness. We then discuss the problems that arise around how one defines the variables for analysis. We discuss the problem of discretizing variables, the problem of variables that are logically related, the problem of automatically finding meaningful and interpretable variables, and a decision theory problem between defining variables that are uncertain but perhaps strong causes vs. variables that are more certain but weak causes.

Video presentation

You will see the recorded video of the talk.

The pdf slides

The presentations slides are also available for off-line browsing.

About the speaker: Richard Scheines' research is on causal discovery, in particular the problem of learning about causation from statistical evidence. The theoretical and computational dimensions of this work have come to be called the TETRAD project, which represents nearly 25 years of collaboration with Clark Glymour, Peter Spirtes and many, many others. Building efficient and practically useful algorithms for causal discovery is as much computer science as philosophy, and thus Richard Scheines has a courtesy appointment in the Machine Learning Department.
He has also put a lot of effort into building and researching the effectiveness of educational software, ranging from intelligent proof tutors to virtual causality labs to a full semester course on Causal and Statistical Reasoning. Because of this work he has a courtesy appointment in the Human-Computer Interaction Institute.