Causal discovery based on non-gaussianity
Patrick Hoyer, University of Helsinki


The estimation of linear causal models from non-experimental data is a well-known problem which has received much attention in the past. Most previous work has been, either explicitly or implicitly, based exclusively on conditional independencies in the data. I will describe some recently developed methods utilizing non-gaussianity (when available) to infer causal relationships. Some problems and limitations of both families of methods will be pointed out, and I will discuss current work that aims to solve these weaknesses by combining the two approaches. Finally, I outline some future directions.


Patrik Hoyer is a postdoc at the Department of Computer Science, University of Helsinki, who has been working on machine learning methods related to independent component analysis and causal inference. He will be visiting the CMU Department of Philosophy until Dec 12th 2007.

Venue, Date, and Time

Venue: NSH 1507

Date: Monday, December 3

Time: 12:00 noon