Studies of problems related to parallel distributed associative models of memory

Parallel distributed associative (PDA) models are a new approach to the study of cognition. They lie between neuroscience and traditional psychological level computation. The PDA paradigm departs from the Rationalist tradition which is a logical, language-based approach to cognition and aims to provide a formal description of cognitive processes at the subsymbolic level using distributed memory structures and distributed control. The first part of this thesis describes an analysis of a PDA model, the Memory Surface Model (MSM) developed by Goldschlager (1984), and compares it to other models in the same paradigm. The second part concerns the implications of FDA models in developing high level primitives for cognitive processes. A method of representing concepts in PDA models is described, and it is shown that the acquisition of optimal concept prototypes and the communication of concepts in the model are intractable in general. A preliminary study using Johnson-Laird's theory of mental models to provide rational reasoning without an innate logical inference mechanism is described. This is discussed with respect to results of experiments described in the psychological literature. This thesis investigates the feasibility of FDA models as an approach to studying cognition. It discusses how a distributed architecture could support high level primitives, and some effects of the use of those primitives in cognitive studies