Applying multi-criteria optimisation to develop cognitive models

Abstract

A scientific theory is developed by modelling empirical data in a range of domains. The goal of developing a theory is to optimise the fit of the theory to as many experimental settings as possible, whilst retaining some qualitative properties such as `parsimony' or `comprehensibility'. We formalise the task of developing theories of human cognition as a problem in multi-criteria optimisation. There are many challenges in this task, including the representation of competing theories, coordinating the fit with multiple experiments, and bringing together competing results to provide suitable theories. Experiments demonstrate the development of a theory of categorisation, using multiple optimisation criteria in genetic algorithms to locate pareto-optimal sets.