A computational model of feature formation, event prediction, and attention switching
- Eman Awad, School of Computer Science, University College Dublin, Dublin, Ireland
- Fintan Costello, University College Dublin, Dublin, Ireland
AbstractIn this paper we present a model of three central aspects of probabilistic cognition: event prediction, feature formation, and attention allocation. While most models of probabilistic reasoning take a parameter estimation and error minimisation approach (sometimes referred to as 'predictive coding', and often described in terms of Bayesian updating), our model takes a contrasting frequentist hypothesis-testing approach. This choice is motivated by a series of recent results suggesting that people's probabilistic reasoning follows frequentist probability theory. In simulation tests we demonstrate that this frequentist model, in which predictive features are formed by a process of null hypothesis significance testing, can give a successful account of event prediction and attentional switching behaviour.