Making Sense of a Variable but Structured World
Perception requires extracting information from variable and ambiguous sensory signals.
Speech is a prime example of this problem: many different kinds of variability affect the speech signal, from environmental noise, to intrinsic variability of the motor systems, to systematic differences between talkers, dialects, registers, etc. I'll talk about a theory grounded in the tradition of Bayesian ideal observer models for understanding how perceptual systems function in the face of such structured variability. In short, this analysis suggests that effective perception in a variable but structured world requires a tight integration between learning, memory, and perception, in order to benefit from the structure present in systematic variability across contexts.
Author: Dave Kleinschmidt