Abstract
Adaptation is a widespread property of sensory systems, generally defined as any change in processing that follows a change in recent input statistics. In the visual system, the effects of adaptation have been described across a variety of different levels of investigation, ranging from the perceptual judgments made by human observers to the spiking activity of individual neurons. As a consequence, the study of adaptation phenomena provides an important testbed for theories that aim to bridge the gap between perception and its underlying neurophysiological mechanisms. The perceptual consequences of visual motion adaptation are often modeled within a simple population-coding framework in which adaptation selectively alters the responsivity of direction-selective cortical neurons (e.g. reducing the gain or tuning bandwidth of neurons with preferred direction that are similar to the adapting stimulus) and downstream mechanisms tasked with decoding population activity fail to compensate for these changes. Here we examine how successful this approach is in accounting for adaptation-induced biases in perceived motion direction (i.e. direction aftereffects) and changes in direction discriminability. We show that observed patterns of bias are consistent with a large set of neural adaptation parameter values, including (but not limited to) those previously reported in single-unit recording studies. The same parameter sets however, systematically underestimate changes in observers’ direction discrimination threshold. Conversely, parameter values that permit an approximation of the profile of changes in discriminability dramatically over-predict the magnitude of shifts in perceived direction. To address this discrepancy, we consider the potential impact of additional factors including the nature of the read-out mechanism and modifications of the pattern and/or correlation structure of neural response variability. Our results suggest that the quantitative relationship between adaptation-induced changes in bias and discriminability places tight constraints on the nature of the underlying neural mechanisms.
Meeting abstract presented at VSS 2013