Introduction to Neural Network Models in Cognitive Science (Psyc 4254)
TTh 1:00 - 2:50, Spring 1998
Frontier Hall 354 and RIAS Room, University of Denver
Yuko Munakata
munakata@kore.psy.du.edu
Frontier Hall 343, 871-4151
Goals: How does the brain think? This course will introduce you to the ideas and methods in computational cognitive neuroscience that have been applied to answering this question. Specifically, we will focus on simulating cognitive and perceptual processes using neural network models, which provide a bridge between behavioral and biological levels of analysis. A core set of computational principles based largely on well-established properties of neural processing in the cortex will be introduced gradually, and used throughout the course to account for a wide range of cognitive phenomena. This focused and unified approach makes potentially difficult material easier to learn, and allows us to explore more complex and interesting phenomena. We will start by understanding the basic computational and biological properties of individual neurons and networks of neurons, which give rise to basic processing mechanisms like spreading activation, inhibition, and multiple constraint satisfaction. We will then discuss processes of learning, which all networks of neurons require to perform any reasonably complex task. Two types of experience-based learning (Hebbian and error-driven) will be covered and related to phenomena in the early perceptual system (e.g., development of simple cell receptive fields in V1) and in higher cognitive tasks (e.g., learning to correctly pronounce and inflect words). We will then examine a range of other cognitive phenomena using this framework, including attention, memory, priming, language, and higher-level cognition (``executive'' control, planning, etc).
Requirements: Although the models are mathematically based, only algebra and some simple calculus-level concepts are required. The focus will be on intuitive and practical applications, not on theoretical derivations. Computer programming experience is not required, because the models are accessible via a graphical interface.
Text: O'Reilly, R. C., Munakata, Y., and McClelland, J.L. (in preparation). Computational Cognitive Neuroscience: Understanding the Mind by Simulating the Brain. MIT Press, Cambridge, MA.
Evaluation Your grade will consist of the following:
The simulation exercises are in the text. You are encouraged do these on your own, but you are also free to talk with one another about them. Aim for clarity and conciseness in writing up your answers; a sentence or two should suffice to answer most questions. Exercises are to be turned in at the beginning of class on the due date (shown in the schedule below). Exercises turned in late will be penalized 5% for each day after the due date, starting immediately after class has started (i.e., if you turn it in at the end of class on the day it's due, it is already 5% off).
Productive participation is encouraged to help you get the most out of this course, both in contributing to class discussion and in actively evaluating and providing feedback on the text (its content, clarity, etc.). Feel free to communicate about any of this as a group by emailing pdp@kore.psy.du.edu (includes me) or pdp-students@kore.psy.du.edu (excludes me).
Class Schedule
In general, we will meet for the first part of each class in Frontier Hall 354 for lecture/discussion, and for the second part in the RIAS room for hands-on simulation.
To Run Simulations on kore.psy.du.edu
from RIAS Room PC's
On PC:
MI/X
TWM
telnet kore.psy.du.edu
On kore:
disp<computer number>
cd sims/chapter_x
leabra++ <prj name>.proj.gz
from other DU PC's
On PC:
helvR10.pcf and
helvR14.pcf) from http://www.microimages.com/freestuf/mix/
MI/X
TWM
telnet kore.psy.du.edu
On kore:
winipcfg.exe to find IP address
setenv DISPLAY <IP address>:0
cd sims/chapter_x
leabra++ <prj name>.proj.gz
from XTerminals
On Xterm:
xhost + kore.psy.du.edu
telnet kore.psy.du.edu
On kore:
setenv DISPLAY <xterminal name>:0
cd sims/chapter_x
leabra++ <prj name>.proj.gz
Handy UNIX commands
Tab completion
^d (to logout or to show completions)
^p (previous command)
^n (next command)
Yuko Munakata