Performance-monitoring integrated reweighting model of perceptual learning.

Perceptual learning (PL) has been traditionally thought of as highly specific to stimulus properties, task and retinotopic position. This view is being progressively challenged, with accumulating evidence that learning can generalize (transfer) across various parameters under certain conditions. For example, retinotopic specificity can be diminished when the proportion of easy to hard trials is high, such as when multiple short staircases, instead of a single long one, are used during training. To date, there is a paucity of mechanistic explanations of what conditions affect transfer of learning. Here we present a model based on the popular Integrated Reweighting Theory model of PL but departing from its one-layer architecture by including a novel key feature: dynamic weighting of retinotopic-location-specific vs location-independent representations based on internal performance estimates of these representations. This dynamic weighting is closely related to gating in a mixture-of-experts architecture. Our dynamic performance-monitoring model (DPMM) unifies a variety of psychophysical data on transfer of PL, such as the short-vs-long staircase effect, as well as several findings from the double-training literature. Furthermore, the DPMM makes testable predictions and ultimately helps understand the mechanisms of generalization of PL, with potential applications to vision rehabilitation and enhancement.

Contrast dependency and prior expectations in human speed perception.

The perceived speed of moving objects has long been known to depend on image contrast. Lowering the contrast of first-order motion stimuli typically decreases perceived speed - the well-known "Thompson effect". It has been suggested that contrast-dependent biases are the result of optimal inference by the visual system, whereby unreliable sensory information is combined with prior beliefs. The Thompson effect is thought to result from the prior belief that objects move slowly (in Bayesian terminology, a "slow speed prior"). However, there is some evidence that the Thompson effect is attenuated or even reversed at higher speeds. Does the effect of contrast on perceived speed depend on absolute speed and what does this imply for Bayesian models with a slow speed prior? We asked subjects to compare the speeds of simultaneously presented drifting gratings of different contrasts. At low contrasts (3-15%), we found that the Thompson effect was attenuated at high speeds: at 8 and 12deg/s, perceived speed increased less with contrast than at 1 and 4deg/s; however, at higher contrasts (15-95%), the situation was reversed. A semi-parametric Bayesian model was used to extract the subjects' speed priors and was subsequently improved by combining it with a model of speed tuning. These novel findings regarding the dual, contrast-dependent effect of high speeds help reconcile existing conflicting literature and suggest that physiologically plausible mechanisms of representation of speed in the visual cortex may need to be incorporated into Bayesian models to account for certain subtleties of human speed perception.

Perceptual learning in visual hyperacuity: A reweighting model.

Improvements of visual hyperacuity are a key focus in research of perceptual learning. Of particular interest has been the specificity of visual hyperacuity learning to the particular features of the trained stimuli as well as disruption of learning that occurs in some cases when different stimulus features are trained together. The implications of these phenomena on the underlying learning mechanisms are still open to debate; however, there is a marked absence of computational models that explore these phenomena in a unified way. Here we implement a computational learning model based on reweighting and extend it to enable direct comparison, by means of simulations, with a variety of existing psychophysical data. We find that this very simple model can account for a diversity of findings, such as disruption of learning of one task by practice on a similar task, as well as transfer of learning across both tasks and stimulus configurations under certain conditions. These simulations help explain existing results in the literature as well as provide important insights and predictions regarding the reliability of different hyperacuity tasks and stimuli. Our simulations also shed light on the model's limitations, for example in accounting for temporal aspects of training procedures or dependency of learning with contextual stimuli, which will need to be addressed by future research.