Predictive Hebbian Learning
Terry Sejwnoski   Peter Dayan   P Read
Montague
Invited talk to COLT 8, 15-18.
Abstract
A creature presented with an uncertain and variable environment needs
to anticipate important future events or risk diminished chances for
survival. These events can include the presence of food, destructive
stimuli, and potential mates. In short, a nervous system must have means
to generate guesses about its most likely next state and the most likely
next state of the world. Psychologists have studied conditions under which
animals can learn to predict future reward and punishment. We review the
computational theory that may be relevant for understanding this form of
learning. Some of the central mechanisms required for predictive learning
have been discovered in both vertebrate (Ljungberg et
al, 1992) and invertebrate brains (Hammer, 1994).