Synthesising cognitive models with genetic programming

Abstract

Developing, understanding and verifying the behaviour of cognitive models is a non-trivial task. A good cognitive model explains and predicts human behaviour in a particular experimental setting. Cognitive models are often in the form of computer programs which need to be designed and written for the given experiment and behaviour: time constraints or natural bias (oversights) often lead to models written by human programmers being constrained to particular theoretical assumptions. We apply program synthesis to this task, introducing novel training and post-processing techniques; our experiments automatically create good quality models, and help visualise the structure of the solution space.