Model selection versus parameter estimation in Bayesian inference: An application to causal learning

Hongjing Lu
University of Hong Kong

Formulating causal learning problem in Bayesian framework helps differentiate two fundamental questions in causality: (1) What likelihood functions do humans use in causal inference? (2) What prior knowledge do humans assume in causal inference? Furthermore, two major types of queries used in human causal learning studies present different computational goals for causal modeling. Specifically, a causal structure query can be formulated as model selection in Bayesian inference, whereas a causal strength query can be formulated as parameter estimation in Bayesian inference. A successful inference model is expected to provide coherent answers to a variety of causal queries and causal judgments. This talk will illustrate how to construct Bayesian models for various experimental designs and then test the validity of computational models by comparing their predictions with human performance.

Presentation (PDF File)

Back to Long Programs