12:00 Wed 19 Jun 1996, WeH 7220
-----------------------------------------------------
Title: Causal Ordering and Causal Discovery
Speaker: Marek J. Druzdzel
         Department of Information Science
           and Intelligent Systems Program
         University of Pittsburgh

Abstract:
In this talk, I will briefly review the mechanism-based view of
causality (Simon 1953; Iwasaki and Simon 1986; Druzdzel and
Simon 1993) and the link between causal ordering and directed
probabilistic graphs, such as Bayesian belief networks.
I will apply this view to the study of assumptions underlying
the conditional independence search approach to causal discovery
from observation (Spirtes, Glymour and Scheines; Pearl and Verma).
(A close relative of this work is known as learning Bayesian
belief networks from data.)
I will propose a deterministic definition of independence and show
a link between the work in econometrics on the problem of
identifiability and the causal discovery work, discussing the
assumptions made in the latter and their identifying strength.
I will also sketch the search-based causal discovery algorithms
in terms of discovery of structure in structural equation models.
---
(This is joint work with Herb Simon.)