Parallel genetic algorithm optimization of a cognitive model: investigating group and individual performance on a math stressor task

Open Access
Kase, Sue Ellen
Graduate Program:
Information Sciences and Technology
Doctor of Philosophy
Document Type:
Date of Defense:
July 21, 2008
Committee Members:
  • Frank Edward Ritter, Committee Chair
  • Mary Beth Rosson, Committee Member
  • Suzanne Michelle Shontz, Committee Member
  • Laura Klein, Committee Member
  • Xiaolong Zhang, Committee Member
  • high-performance computing
  • genetic algorithm
  • cognitive modeling
  • cognitive science
  • stress
  • mathematics anxiety
The purpose of the thesis was to enable an individual differences investigation of how stress affects cognitive performance in a mental serial subtraction task. To do this a new optimization approach was developed for computing the fit of a serial subtraction cognitive model to human performance data by varying architectural parameters. The new optimization approach utilizes a modified parallel genetic algorithm (PGA) on a high performance computing (HPC) cluster together with the ACT-R cognitive architecture and a serial subtraction cognitive model. Modifications to the PGA were necessary because of stochasticity embedded in the subsymbolic processes of the model’s architecture. The optimization approach in this thesis was highly successful in fitting the serial subtraction model to three different levels of human data: average across subjects, challenge and threat task appraisal groups, and single-subject level performance. The optimization results revealed several interesting patterns in the parametric values found to produce best fits to the human data. Two of the patterns are supported by individual differences theories of stress and anxiety from cognitive performance research. If the field’s traditional manual optimization process had been used instead, these patterns would have gone undiscovered. Contributions of the thesis include: development of a software system for individual differences modeling; increased understanding of ACT-R modeling, and the effects of anxiety on cognitive performance; a new application of PGAs; and a new paradigm for model creation and validation as the purposefulness of the thesis optimization approach was shown applicable in both model development and model validation.