Psychology 190(4): "Neural Networks"

Winter quarter 1997


Efforts to understand the brain draw upon multiple scientific methodologies and disciplines. Neurobiological studies focus on characterizing various aspects of neural function and their relation to behavior. Psychological studies attempt to characterize various aspects of behavior or performance with the hope of revealing underlying processes or mechanisms. More recently, there has been a growing effort to draw upon principles from mathematics, physics, engineering and computer science in order to formulate neural network models capable of elucidating the link between neurons and behavior. This course will focus on this latter approach, and will introduce basic neural network paradigms that have been used as models of sensory coding, pattern recognition, associative memory, motor control, learning, and self-organization. Specific topics to be covered include Adaline/perceptron learning, backpropagation, Hebbian learning, competitive learning, redundancy reduction, self-organizing maps, Hopfield networks, probabilistic modeling, Boltzmann machines, Helmholtz machines, and Bayes nets. The emphasis will be on understanding the key conceptual ideas behind these models and what insights they lend into brain and behavior.


Instructor:

Bruno A. Olshausen, 268J Young Hall, 757-8749

Units:

4

Time:

Monday, 6-9, in room 145 of Young Hall.

Prerequisites:

The course is aimed at advanced undergraduates or graduate students in psychology and neuroscience, although enrollment is open to students from all disciplines. Some familiarity with basic concepts of calculus, linear algebra, or probability theory is desirable.

Texts:

"Parallel Distributed Processing," by J.L. McClelland and D.A. Rumelhart. MIT Press.

"An Introduction to Neural Networks," by J.A. Anderson. MIT Press