Causal Discovery of Latent-Variable Models from a Mixture of Experimental and Observational Data

Document Type

Presentation Abstract

Presentation Date

12-9-2004

Abstract

I will describe a Bayesian method for learning causal Bayesian networks through networks that contain latent variables from an arbitrary mixture of observational and experimental data. Observational data are passively observed. Experimental data, such as those produced by randomized controlled trials, result from the experimenter's manipulating, typically randomly, one or more variables and observing the states of other variables. I will also present Bayesian methods (including two new methods) for learning the causal structure and parameters of the underlying causal process that is generating the data, given that (1) the data contain a mixture of observational and experimental case records, and (2) the causal process is modeled as a causal Bayesian network. These learning methods were applied using as input various mixtures of experimental and observational data that were generated from the ALARM causal Bayesian network. In these experiments, the absolute and relative quantities of experimental and observational data were varied systematically. For each of these training datasets, the learning method was applied to predict the causal structure and to estimate the causal parameters that exist among randomly selected pairs of nodes in ALARM. I report how these structure predictions and parameter estimates compare with the true causal structures and parameters as given by the ALARM network. I show that one of the new methods for learning Bayesian network structure from a mixture of data, the implicit latent variable modeling method, is asymptotically correct and efficient.

Additional Details

Thursday, 9 December 2004
4:10 p.m. in Math 109

This document is currently not available here.

Share

COinS