Using Model Tracing and Evolutionary Algorithms to Determine Parameter Settings for Cognitive Models From Time Series Data such as Visual Scanpaths

Time-series data such as eye movements or mouse movements contain rich information about the dependencies between successive human actions. This information can be potentially very useful for examining model assumptions and constraining parameter search. This paper explores the use of model tracing, which simulates a task by tracking observable human behaviors, to time-series data. We explore two aspects of tracing that are different from conventional cognitive modeling: (a) tracking the observed behaviors and (b) estimating the likelihood of the observed events. We demonstrate how these two features of tracing, along with the use of an evolutionary optimization algorithm, led to accurate and robust estimates for parameters of visual acuity functions needed by visual search models.