Comparing causal explanations with Bayesian methods

How can researchers decide which causal structure best explains observed data? In his guest talk, Joe Suzuki will present a Bayesian approach that considers not only how well a model fits the data, but also how complex it is.

Suzuki is a Professor Emeritus of Statistics at University of Osaka. He will visit the Chair of Causality, led by Prof. Alexander Marx at TU Dortmund University’s Department of Statistics and affiliated with the Research Center Trustworthy Data Science and Security (RC Trust), for a presentation titled “Bayesian ICA for Causal Discovery.”

Independent component analysis, or ICA, provides a framework for causal discovery when observed variables are generated from independent, non-Gaussian sources. Suzuki will discuss how this approach can be formulated within a Bayesian framework.

The central idea is to compare possible causal structures using their marginal likelihoods. This enables researchers to consider both model fit and model complexity. Suzuki will explain how Bayesian model selection can help identify causal orderings and dependency structures. He will also discuss computational strategies for reducing the number of possible structures that need to be examined. 

Event Details

Talk: Bayesian ICA for Causal Discovery
Speaker: Prof. Joe Suzuki, Division of Mathematical Science, Osaka University
Date: Monday, August 10, 2026
Time: 11:00 a.m.
Location: Joseph-von-Fraunhofer-Straße 25, Room 303
Online: Zoom