A review on linking cognitive models from different levels of analysis.

Abstract

Over the last decades, many cognitive models of the human mind have been made. These models are not always comparable to each other, one model might model the behaviour of individual neurons, whilst more abstract models can model how a more global cognitive process might work. To create some structure in the separation and comparability of different cognitive models, David Marr introduced his theory of different levels of analysis. Separating the models on different levels of analysis simplified the comparison between models on the same level of analysis, but it complicated the comparison between models from different levels of analysis. In this paper I have proposed and tested the method of comparing models from different levels of analysis through the means of a linking hypothesis. A linking hypothesis tests if a model from one level of analysis is a realisation of a model from another level of analysis by testing whether the fundamental differences between the two models result in two fundamentally different functioning model. To create the optimal conditions for the testing of a linking hypothesis, I have first conducted a literature research around memory models on the computational level of analysis and then select one model which is most suited for the testing of a linking hypothesis. Then I conducted two small simulations, testing the linking hypothesis on this model and a model that hypothetically realizes from it. From this small experiment I can conclude that a linking hypothesis is a successful method for comparing models from different levels of analysis.