Proceedings of the Annual Meeting of the Cognitive Science Society

There is large agreement among vision scientists that biolog-ical perception is capacity-limited and that attentional mecha-nisms control how that capacity is allocated. Despite the factthat Bayesian models generally do not include capacity limits,many researchers model perceptual attention as the result ofoptimal Bayesian inference. This inconsistency arises becausevision science currently lacks a feasible and principled com-putational framework for characterizing optimal attentional al-location in the presence of capacity constraints. Here, weintroduce such a framework based on rate-distortion theory(RDT), a theory of optimal lossy compression developed in theengineering literature. Our approach defines Bayes-optimalperformance when an upper limit on information processingrate is imposed. Here, we compare Bayesian and RDT ac-counts in a visual search task, and highlight a typical shortcom-ing of unlimited-capacity Bayesian models that is not sharedby RDT models, namely that they often over-estimate task-performance when information-processing demands are in-creased. In this study, we asked human subjects to find eitherone or two targets in a collection of distractors in a single-fixation search task. We predicted relative performance be-tween one- and two-target conditions based on both RDT andBayesian models. Performance differed between conditions ina way that was well accounted for by the capacity-limited RDTmodel but not by the capacity-unlimited Bayesian model.