New Investigator Award Recipient Presentation

Dr. Michael D. Lee

(University of Adelaide, Australia)

Inferring Stimulus Representations from Similarity Data

A basic problem for cognitive psychology is to understand the way people mentally represent stimuli. One widely

used approach for inferring stimulus representations from data relies on modeling measures of stimulus similarity.

A significant challenge in making inferences about stimulus representations from similarity data, however, is that it

requires striking the right balance between data-fit and model complexity. Alternative assumptions about how

stimuli are represented, and how similarity is measured across these structures, lead to different representational

models with different inherent complexities.

This talk considers the application of Bayesian model selection methods to a number of different representational

models. Initially, we focus on featural models, including additive clustering, additive trees, Tversky's contrast

model, and a recently proposed modification of the contrast model. Later, we considers how model selection helps

constrain even more flexible model classes, such as a combined approach that uses both features and dimensions,

and an individual differences approach that uses different representations for different groups of subjects.