Adaptation and learning priors in visual inference

Users’ prior expectations are an important but understudied degree of freedom in visual inferences. We ask: To what extent are priors learned through visual experience? How do they impact behavior? Can we design visual analytics systems to manipulate users’ priors and calibrate their sensitivity to the signal in data? We connect theoretical accounts of priors in visual inference with the empirical results from psychophysical and physiological studies of visual adaptation: a ubiquitous process by which the neural code calibrates to statistics of the immediate environment. The way the visual system adapts its internal representations based on experience explains implicit learning of empirical priors for the purpose of visual inference. Drawing on the visual adaptation literature, we present a framework for researching and designing for priors in visual inference.