Analysis of Symbolic Parameter Models

Abstract

Abstract : Model-fitting, the problem of finding parameter settings that cause a model to fit given data as closely as possible, is a hard but important problem in cognitive science in general, and in cognitive diagnosis in particular. Efficient solutions have been found for certain types of model-fitting problems (e.g., linear and integer programming) that involve specific types of parameters (usually continuous) and models (usually linear). But these techniques usually do not apply to computational cognitive models whose parameters are often discrete and symbolic and whose internal workings must be treated as a black box for the purposes of fitting. We have constructed ASPM (Analysis of Symbolic Parameter Models), a suite of computational tools for fitting and analyzing such symbolic parameter models. We have shown how it can be used to fit and analyze computational models with well over 10 billion combinations of parameter settings. We have made a good start towards making ASPM robust and user friendly in preparation for releasing it for public use. (AN)