Introduction to Connectionist Modelling of Cognitive Processes

by McLeod, Peter; Plunkett, Kim; Rolls, Edmund T.

ISBN13: 9780198524274

ISBN10: 0198524277

Format: Hardcover

Pub. Date: 1998-04-23

Publisher(s): Oxford University Press

Availability: This title is currently not available.

Other versions by this Author

This Item Qualifies for Free Shipping!*

*Excludes marketplace orders.

List Price: ~~$230.39~~

Rent Textbook

Select for Price

Add to Cart

There was a problem. Please try again later.

New Textbook

We're Sorry
Sold Out

Used Textbook

We're Sorry
Sold Out

eTextbook

We're Sorry
Not Available

Summary

Connectionism is a way of modelling how the brain uses streams of sensory inputs to understand the world and produce behaviour, based on cognitive processes which actually occur. This book describes the principles, and their application to explaining how the brain produces speech, formsmemories and recognises faces, how intellect develops, and how it deteriorates after brain damage. Part I explores the basic concepts, the architecture and properties of the most common connectionist models, and how connectionist learning rules work. Part II describes and evaluates connectionistmodels of a variety of cognitive processes, including the learning and production of speech, the formation of episodic memories and visual representations, the development of cognitive processes in infancy, and their breakdown in brain-damaged patients. The models range from some well-known classicsto others at the frontiers of current research. Each chapter ends with a list of recommended further reading. Also included is a disk with the software for running tlearn, a user-friendly simulator for connectionist modelling of cognitive processes, which will run on either PCs or Macs. Thesoftware includes exercises to introduce the simulator, and working copies to explore some of the models described in the text. A reference handbook for tlearn is included to enable readers to build their own models. The authors, as well as being leading researchers in their field, have extensiveexperience of teaching connectionism to undergraduates. They have written the first comprehensive, up-to-date textbook on connectionist modelling, designed specifically for advanced undergraduates, and accessible to those with only limited knowledge of mathematics. This will be an essentialintroductory text for all students in psychology or cognitive science taking a course on connectionism.

Prologue

(7)

What is the problem?

(1)

What connectionist models can do

(6)

Part I Principles

(147)

1 The basics of connectionist information processing

(22)

Neurally inspired information processing

(3)

Five assumptions about computation in the brain on which connectionist models are based

(4)

Symbols and elementary equations

(5)

Connectionism in a nutshell

(1)

Exercises with tlearn

(9)

2 The attraction of parallel distributed processing for modelling cognition

(21)

The representation of knowledge in connectionist networks is distributed

(1)

Distributed representations are damage resistant and fault tolerant

(2)

Connectionist networks allow memory access by content

(1)

Retrieving information from a distributed database

(10)

Constraint satisfaction in connectionist networks

(3)

There is no distinction between `memory' and `processing' in connectionist models

(1)

Problems for distributed representations

(2)

3 Pattern association

(21)

The architecture and operation of a pattern associator

(10)

A pattern association network

(2)

The Hebb rule

(1)

Learning with the Hebb rule

(2)

Recall from a Hebb trained matrix

(1)

Learning different associations on the same weight matrix

(3)

Recall reflects the similarity of retrieval pattern and stored patterns

(2)

Properties of pattern associators

(4)

Generalisation

(1)

Fault tolerance

(1)

The importance of distributed representations for pattern associators

(1)

Prototype extraction and noise removal

(1)

Speed

(1)

Interference is not necessarily a bad thing

(1)

Further reading

(1)

Training a pattern associator with tlearn

(7)

4 Autoassociation

(24)

The architecture and operation of an autoassociator

(3)

Architecture

(2)

Learning with the Delta rule

(1)

Properties of autoassociator memories

(8)

What an autoassociator learns

(2)

Storage of different memories on the same connections

(2)

Pattern completion

(1)

Noise resistance

(2)

Forming categories and prototypes from individual experiences

(5)

Discovering a prototype from exemplars with an autoassociator

(3)

Learning different prototypes on the same matrix

(1)

Further reading

(1)

Autoassociation exercises with tlearn

(8)

5 Training a multi-layer network with an error signal: hidden units and backpropagation

(31)

The perceptron convergence rule

(2)

Gradient descent

(6)

Gradient descent with a sigmoid activation function

103

(2)

Linear separability

105

(3)

Solving the XOR problem with hidden units

107

(1)

Hidden units and internal representation

108

(4)

Hinton's family tree problem

108

(2)

What the hidden units represent in the family tree task

110

(2)

Backpropagation

112

(5)

The problem

113

(1)

An informal account

114

(1)

Local minima

114

(2)

Backpropagation and biological plausibility

116

(1)

Exercise: learning Exclusive OR with tlearn

117

(10)

6 Competitive networks

127

(12)

The architecture and operation of a competitive network

128

(8)

Excitation

128

(1)

Competition

129

(1)

Weight adjustment

129

(3)

Limiting weight growth

132

(1)

Competitive learning in the brain

133

(3)

Pattern classification

136

(2)

Correlated teaching

137

(1)

Further reading

138

(1)

7 Recurrent networks

139

(16)

Controlling sequences with an associative chain

139

(1)

Controlling sequences with a recurrent net

140

(2)

State units and plan units

141

(1)

Simple recurrent networks (SRNs)

142

(3)

Learning to predict the next sound in a sequence

144

(1)

Attractors

145

(3)

Learning sequences with tlearn

148

(7)

Part II Applications

155

(176)

8 Reading aloud

155

(23)

The traditional `2-route' model of reading aloud

156

(1)

The connectionist approach

157

(1)

The Seidenberg and McClelland model of reading aloud

158

(9)

Replicating the results of word naming experiments

161

(4)

What has the model learnt?

165

(1)

Limitations of the model

166

(1)

The Plaut, McClelland, Seidenberg and Patterson model

167

(4)

Input coding

167

(1)

Pronunciation

168

(1)

Reading with an attractor network

169

(2)

Componential attractors

171

(1)

What have these models achieved?

171

(1)

Further reading

172

(1)

Reading aloud with tlearn

172

(6)

9 Language acquisition

178

(32)

Learning the English past tense

179

(8)

A symbolic account of past tense learning

180

(3)

A connectionist account of past tense learning

183

(4)

Early lexical development

187

(7)

A connectionist model of early lexical development

188

(4)

Evaluation of the model

192

(2)

The acquisition of syntax

194

(8)

Further reading

202

(1)

Learning the English past tense with tlearn

202

(8)

10 Connectionism and cognitive development

210

(33)

Stages in development-a challenge for connectionism?

210

(2)

The development of object permanence

212

(7)

Modelling the development of representations which could produce object permanence

212

(6)

Evaluating the model

218

(1)

The balance beam problem

219

(15)

Modelling the balance beam problem

222

(2)

Running the model

224

(7)

Evaluating the model

231

(3)

Stage-like behaviour from continuous change

234

(1)

Variability in learning

234

(6)

Individual differences

235

(3)

Critical periods

238

(2)

Further reading

240

(1)

Modelling the balance beam problem with tlearn

240

(3)

11 Connectionist neuropsychology-lesioning networks

243

(25)

The simulation of deep dyslexia

245

(10)

Hinton and Shallice (1991)

246

(2)

Attractors

248

(1)

Attractor basins

249

(1)

Lesioning an attractor

250

(1)

Is the result dependent on fine details of the model?

251

(3)

The interpretation of double dissociation

254

(1)

Modelling a deficit in semantic memory

255

(5)

Modality and category specificity in semantic memory

255

(3)

Modelling an asymmetrical double dissociation

258

(2)

Modelling an information processing deficit in schizophrenia

260

(8)

Selective attention

260

(2)

Modelling the Stroop task

262

(1)

Lesioning the model

263

(2)

Further reading

265

(1)

Exercises with tlearn

265

(3)

12 Mental representation: rules, symbols and connectionist networks

268

(10)

Learning minority default rules

268

(5)

Default mapping

269

(1)

Minority defaults

270

(3)

Symbols and distributed representations

273

(3)

Representing mental types

274

(1)

Symbolic attractors

275

(1)

Levels of explanation

276

(1)

Further reading

277

(1)

13 Network models of brain function

278

(25)

Memory formation in the hippocampus

278

(14)

The role of the hippocampus in memory formation

279

(1)

Information flow to and from the hippocampus

280

(1)

The internal structure of the hippocampus

281

(2)

A computational theory of hippocampal operation

283

(3)

A neural network simulation of hippocampal operation

286

(1)

The performance measure

287

(1)

Running the model

288

(1)

Performance of the network

288

(4)

Invariant visual pattern recognition in the inferior temporal cortex

292

(10)

How not to achieve position invariant object recognition

292

(2)

The flow of visual information from retina to temporal lobe

294

(1)

VisNet-an approach to biologically plausible visual object identification

294

(3)

Testing the network

297

(2)

The importance of the trace rule for forming invariant representations

299

(1)

Brains, networks and biological plausibility

300

(2)

Further reading

302

(1)

14 Evolutionary connectionism

303

(11)

The evolution of goal directed behaviour

304

(4)

The evolutionary advantage of the capacity to learn

306

(2)

Innately guided learning in speech perception

308

(6)

The Nakisa and Plunkett model

309

(1)

Network training and evolution

310

(1)

Speed

311

(1)

Cross-linguistic performance

311

(1)

Categorical perception

312

(1)

Nativism or constructivism?

313

(1)

15 A selective history of connectionism before 1986

314

(17)

McCulloch and Pitts (1943)

314

(4)

Logical operations with neuron-like computational units

314

(1)

Computing AND, OR and NOT

315

(1)

Producing the sensations of hot and cold

316

(2)

Hebb (1949)

318

(2)

Neuronal inspiration in psychological modelling

318

(1)

The Hebb synapse

319

(1)

Rosenblatt (1958)-the perceptron

320

(3)

Minsky and Papert (1969)-a critique of perceptrons

323

(2)

The XOR problem and the perception of connectedness

323

(2)

Hinton and Anderson (1981)

325

(1)

Hopfield (1982)

326

(5)

Content-addressable memory in networks with attractor states

326

(1)

Input patterns and energy

327

(2)

Novel inputs produce higher values of E than memories

329

(1)

Changes in state lead to a reduction in E

329

(2)

Appendix 1 Installation procedures for tlearn

331

(2)

Appendix 2 An introduction to linear algebra for neural networks

333

(7)

Appendix 3 User manual for tlearn

340

(36)

Bibliography

376

(7)

Index

383

Introduction to Connectionist Modelling of Cognitive Processes

Rent Textbook

New Textbook

Used Textbook

eTextbook

Summary

Table of Contents

Digital License