A computational model of eye-movements in category learning

Models of attention in category learning tasks have typically treated attention as a weighting of how influential a feature is to the correct classification of the overall stimulus. Attention shifting is frequently modelled as occurring after the trial is completed (Kruschke, 1992). Recent work has demonstrated in detail how learned attention develops during the course of a single trial. Currently, there is no model which can account for the dynamic attentional shifts that are identified by eye-tracking data. Additionally, research is many fields has identified the need to explore cognitive models that are based on a more naturalistic view of human behaviour. New mathematical techniques utilizing concepts from dynamical systems has greatly increased the tractability of developing such models. This thesis describes two category learning experiments and introduces a new computational model that produces a real-time simulation of eye-movements in these tasks. Human data is compared with the model output and the implications of this model to category learning and related fields is discussed.