Abstract:

Perception is commonly understood as the process of inferring the state of the world from incomplete, noisy sensory information. Bayesian models of perceptual behavior have shown that human observers can exploit statistical regularities of their natural environment (i.e., priors) to reduce uncertainty and support rational perceptual decisions. In this talk, I will present experimental and theoretical work from my laboratory showing that incorporating resource constraints into these rational models substantially improves their quantitative predictions. I will argue that the resource limitations of biological perceptual systems do not merely constrain perceptual performance, but instead impose computational principles that fundamentally shape perceptual behavior.