Using HPC and PGAs to Optimize Noisy Computational Models of Cognition

Cognitive modeling on high performance computing platforms is an emerging field. A preliminary analysis i s presented on the use of parallel processing and genetic algorithms for optimizing the fit of non-linear, multivariable symbolic models of human cognition to experimental data. The effectiveness of this experimental optimization methodology i s illustrated with a prototype model of a serial arithmetic task built in the ACT-R cognitive architecture. The results confirm that HPC-based optimization techniques could replace the manual optimization techniques used by cognitive modelers up until the present.