Localist Connectionist Approaches to Human Cognition

Localist Connectionist Approaches to Human Cognition

Localist Connectionist Approaches to Human Cognition

Localist Connectionist Approaches to Human Cognition

Synopsis

This volume provides an overview of a relatively neglected branch of connectionism known as localist connectionism. The singling out of localist connectionism is motivated by the fact that some critical modeling strategies have been more readily applied in the development and testing of localist as opposed to distributed connectionist models (models using distributed hidden-unit representations and trained with a particular learning algorithm, typically back-propagation). One major theme emerging from this book is that localist connectionism currently provides an interesting means of evolving from verbal-boxological models of human cognition to computer-implemented algorithmic models. The other central messages conveyed are that the highly delicate issue of model testing, evaluation, and selection must be taken seriously, and that model-builders of the localist connectionist family have already shown exemplary steps in this direction.

Excerpt

In the last two decades, connectionism became a major theoretical force in contemporary cognitive psychology. Nobody today would deny the extraordinary impact that this approach has brought to our discipline, although some still question the positive nature of this impact. the present volume provides an overview of a relatively neglected branch of connectionism known as localist connectionism, but not as an attempt to introduce a theoretical divide: We firmly believe that unification rather than division is in order and that connectionism holds much promise as a unifying theoretical framework. Rather, we have singled out localist connectionism because some critical modeling strategies, to be discussed more fully in our two contributions to this volume, have been more readily applied in the development and testing of localist than in distributed connectionist models (i.e., models using distributed hidden-unit representations and trained with a particular learning algorithm, typically back-propagation).

Scientists have made serious attempts to develop a systematic, principled approach in the development, testing, and evaluation of models of human cognition. Model development involves the application of basic strategies such as canonical modeling, functional overlap, and nested modeling (see Jacobs &Grainger, 1994;Grainger &Jacobs, chap. 1;Jacobs,Rey,Ziegler, &Grainger, chap. 5 in the present volume for further details). the application of such strategies facilitates isolation of the key principles underlying a model's performance and helps assign explanatory credit or blame for a given data pattern. It should also lead to the emergence of models with greater empirical content (Popper, 1935/1959).

Search by... Author
Show... All Results Primary Sources Peer-reviewed

Oops!

An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.