Approximate Inference through Active Sampling of Likelihoods Accounts for Human Categorization Behavior

Bayesian computations are intractable and expensive, but this is rarely accounted for in existing Bayesian observer models. In this work, we propose that a) the brain only computes imprecise (noisy) estimates of likelihoods and posteriors, and, b) since computations are expensive, the brain actively chooses which computations to perform to refine such estimates. We call our framework approximate inference through active sampling (AIAS) and study its implications in N-alternative categorization. AIAS accounts for several empirical findings. First, we account for a puzzling recent finding that decision confidence follows the difference between the two highest posteriors, rather than the highest posterior itself. AIAS not only provides better fits, but also yields an accurate prediction of response times based on the number of iterations. Second, we show that AIAS is able to explain how categorization behavior changes when the visual contrast varies. Third, we find that the mean response times predicted by AIAS grows approximately logarithmically with the number of categories $N$, as per Hick's law. Overall, AIAS provides a novel approach to explain human categorization by casting approximate inference as an active-sampling process with imprecise computations.