Sergei Gepshtein, Luis Lesmes, Ivan Tyukin, Thomas Albright

Introduction. It has been proposed that sensory adaptation optimizes visual sensitivity to properties of the variable environment (Sakitt and Barlow, 1982; Wainwright, 1999). On this view, motion adaptation is expected to improve the ability to perceive motion at the adapting conditions. Yet experimental evidence of motion adaptation has been controversial. In speed adaptation, for example, sensitivity to adapting speeds can either increase or decrease; it can also change for speeds very different from the adapting speed (Krekelberg, van Wezel, and Albright, 2006). We presently test a normative theory of motion adaptation which implements the premise that motion adaptation improves the ability to perceive motion in a new environment but which also predicts that (a) sensitivity to motion must deteriorate at some adapting conditions, and (b) adaptation-induced changes are global so increments and decrements of sensitivity are expected also away from the adapting conditions.

Theory. According to a new normative-economic theory of motion perception (Gepshtein, Tyukin, and Kubovy, 2007), spatiotemporal sensitivity manifests an optimal allocation of limited resources (such as motion-sensitive cells) to conditions of visual stimulation. The allocation is optimal in two respects. First, it balances the errors of estimating stimulus location and stimulus content, satisfying the uncertainty principle of measurement (Gabor, 1946). Second, it places more resources at the conditions where the resources are more likely to be used, by taking into account the statistics of visual stimulation. The theory predicts that motion adaptation should induce a characteristic pattern of changes in the spatiotemporal contrast sensitivity function (Kelly, 1979), forming well-defined foci of increased and decreased sensitivity across a map of sensitivity to spatial and temporal frequencies of stimulation.

Experiments. We tested the predictions by measuring human contrast sensitivity over a large range of spatial and temporal frequencies (0.25-8 c/deg and 0.5-32 Hz). The observers viewed drifting luminance gratings of variable contrast and discriminated the direction of motion. We varied the statistics of motion speed: In some blocks of trials low speeds were more common than high speeds, and in other blocks high speeds were more common than low speeds. To rapidly measure the entire contrast sensitivity function in both statistical contexts, we used a novel adaptive procedure that combined Bayesian adaptive inference with a trial-to-trial information-gain strategy (Lesmes, Lu, Baek, and Albright, 2008). We compared the spatiotemporal sensitivity functions obtained in the different statistical contexts and found that sensitivity changed similar to our predictions. The changes were global and they formed foci of increased and decreased sensitivity, so the map of observed changes was similar to the map of predicted changes.

Conclusions. These findings support the normative-economic theory and the view that motion adaptation amounts to reallocation of neural computational resources in the visual system: The allocation of sensitivity takes into account both the uncertainty principle of measurement and the statistics of stimulation. Since computational resources of the visual system are limited, improvement of sensitivity to some stimuli is accompanied by deterioration of sensitivity to other stimuli.

COSYNE 2009