Bayesian model comparison and distinguishability

Julien Diard (1)

1 LPNC - Laboratoire de Psychologie et NeuroCognition

This paper focuses on Bayesian modeling applied to the experimental methodology. More precisely, we consider Bayesian model comparison and selection, and the distinguishability of models, that is, the ability to discriminate between alternative theoretical explanations of experimental data. We argue that this last concept should be central, but is difficult to manipulate with existing model comparison approaches. Therefore, we propose a preliminary extension of the Bayesian model selection method that incorporates model distinguishability, and illustrate it on an example of modeling the planning of arm movements in humans.

Domaines: Informatique; Apprentissage [cs.LG]

Julien Diard. Bayesian model comparison and distinguishability. International Conference on Cognitive Modeling (ICCM 09), Jul 2009, Manchester, United Kingdom. pp.204-209. ⟨hal-00530370⟩