(Approximately) Optimal Models of Animal Conditioning

Peter Dayan

Ideas from statistics and engineering, such as the Kalman filter and dynamic programming, offer appealing formal frameworks within which aspects of conditioning behavior look roughly optimal, and aspects of the neural substrate can be closely captured. I will discuss this program from the perpsective of two rather puzzling conditioning
paradigms: downwards unblocking and incentive learning.

Presentation Files (Zip Archive)

Back to Probabilistic Models of Cognition: The Mathematics of Mind