Minimum Description Length and Cognitive Modeling
Yong R. Su, In Jae Myung, Mark A. Pitt · The MIT Press eBooks · 2005
The question of how one should decide among competing explanations of data is at the heart of the scientific enterprise. In the field of cognitive science, mathematical models are increasingly being advanced as explanations of cognitive behavior. In the application of Minimum Description Length (MDL) principle to the selection of these models, one of the major obstacles is to calculate Fisher information. In the present study we provide a general formula to calculate Fisher information for models of cognition that assume multinomial or normal distributions. We also illustrate the usage of the formula for models of categorization, information integration, retention, and psychophysics. Further, we compute and compare the complexity penalty terms of two recent versions of MDL [Rissanen 1996; Rissanen 2001] for a multinomial model. Finally, the adequacy of MDL is demonstrated in the selection of retention models. The study of cognition is concerned with describing the mental processes that un-derly behavior and developing theories that explain their operation. Often the the-