| New Bulgarian University > | Center for Cognitive Science > | Summer Schools > | 2001 > | Course Description |
New Bulgarian University
This
course is intended to give initial theoretical and practical background in the
connectionist approach to Cognitive Science. The basic concepts and methods of
connectionism will be presented with their pros, cons and limitations. On this
basis, connectionist models of different cognitive tasks will be discussed.
During the small group sessions the participants will be able to get some
hands-on experience in connectionist modeling using the simulation interface
Tlearn.
Class 1: Models in Cognitive Science: symbolic, connectionist
and dynamical system approaches. The connectionist approach to Cognitive
Science. Connectionist architectures. Basic concepts.
Readings:
·
Chs. 1, 2 and 15 in McLeod, P., Plunkett, K., and
Rolls, E.T. (1998), “Introduction to Connectionist Modelling of Cognitive
Processes”, Oxford
Optional readings:
·
Ch. 5 in Varela
F. J., Thompson, E., Rosch, E., “Emergent Properties and
Connectionism”, “The
embodied mind: Cognitive Science and Human Experience”, MIT.
·
Ch. 1 in
Rolls E. T. and Treves A. (1998), “Neural Networks and Brain
Function”,
Oxford
· Ch. 2 in PDP:
Explorations in the Microstructure of Cognition (1986), “A General Framework
for Parallel Distributed Processing”, Rumelhart, D. E., Hinton, G. E. and
McClelland, J. L.
·
“The Symbolic
and Connectionist Paradigms”, Dinsmore J., ed. (1992), Lawrence Erlbaum
Associates, Ch. 1: “Thunder in the
Gap”, Dinsmore J.
·
Port, R. F. and
van Gelder, T. (1995), “It’s About Time: An Overview of the Dynamical
Approach to Cognition”, ch. 1 in “Mind as Motion: Exploration in the
Dynamics of Cognition”, eds. Port,
R. F. and van Gelder, T. (1995)
Class 2: Learning paradigms: supervised and unsupervised
learning. Learning in single-layered networks. The Hebbian learning rule.
Pattern association and auto association. The ‘delta’ learning rule.
Readings:
·
Chapter 3 and
Chapter 4
in
McLeod, P., Plunkett, K., and Rolls, E.T. (1998), “Introduction to
Connectionist Modelling of Cognitive Processes”,
Oxford
Optional readings:
· Chapter 2 and Chapter 5 in Rolls E. T. and Treves A. (1998), “Neural Networks and Brain Function”, Oxford
Class 3: Supervised
learning in multi-layered networks. The Backpropagation learning rule.
Readings:
·
Ch. 5 in McLeod, P., Plunkett, K., and Rolls, E.T.
(1998), “Introduction to Connectionist Modelling
of Cognitive Processes”, Oxford
Optional readings:
·
Seijnowski, T.
J. and Rosenberg, C. R. (1987), NETtalk: a parallel network that learns to read
aloud, Complex Systems 1:145-168
·
Chapter 5 in Rolls E. T. and Treves A.,(1998), “Neural Networks and Brain Function”,
Oxford
·
Ch. 8 in PDP:
Explorations in the Microstructure of Cognition, “Learning Internal
Representations by Error Propagation”, Rumelhart, D. E., Hinton, G. E. and
Williams, R. J.
Class 4: Unsupervised learning. Competitive learning. Architecture and operation.
Readings:
·
Ch. 6 in McLeod, P., Plunkett, K., and Rolls, E.T.
(1998), “Introduction to Connectionist Modelling of Cognitive
Processes”,
Oxford
Optional readings:
·
Ch. 4 in Rolls, E. T. and Treves, A. (1998), “Neural
Networks and Brain Function”, Oxford
Class 5: Recurrent networks. Simple recurrent networks. The constraint satisfaction problem. Hopfield networks.
Readings:
·
Chs. 2,
7 and 15
in
McLeod, P., Plunkett, K., and Rolls, E.T. (1998), “Introduction to
Connectionist Modelling of Cognitive Processes”, Oxford
Optional readings:
·
Ch. 3 in Rolls, E. T. and Treves, A. (1998), “Neural
Networks and Brain Function”, Oxford
·
Elman J. L.
(1990), “Finding Structure in
Time”, Cognitive Science v. 14, 179-211
·
McClelland, J.
L. and Rumelhart, D. E. (1981) “An Interactive Activation Model of Context
Effects in Letter Perception: Part 1. An Account of Basic Findings.”, Psych.
Rev. v. 88, no. 5, 357.
These
sessions will be mainly devoted to the use of connectionist simulation
environments (Tlearn, PDP). Several simulations will be carried out that
illustrate models and principles presented during the lectures.
Readings:
Appendices 1 and 3 in McLeod, P., Plunkett, K., and
Rolls, E.T. (1998), “Introduction to Connectionist Modelling of Cognitive
Processes”, Oxford
Plunkett, K., and Elman, J.L. (1998), “Exercises in
Rethinking Innateness: A Handbook for Connectionist Simulations”, MIT
Students
who desire credit can create a simple neural network simulation, based on the
examples that have been considered and present it in a paper like form,
stressing the presentation of the network architecture, adequacy of the
representation, potential for scaling etc.
Maurice
Grinberg is Associate Professor in the Department of Cognitive Science and
Psychology in the New Bulgarian University. His research interests are in the
field of cognitive modeling.