Mental Rotation as Bayesian Quadrature

Given a computational resource–for example, the ability to visualize an object rotating–how do you best make use of it? We explored how mental simulation should be used in the classic psychological task of determining if two images depict the same object in different orientations. We compared two models on this mental rotation task, and found that a model based on an optimal experiment design for Bayesian quadrature is qualitatively more consistent with classic behavioral data than a simpler model. We suggest that rational models which adaptively exploit available resources are promising in their ability to characterize metacognitive processes like mental simulation.