Abstract
Selective visual attention is generally conceptualized to control the flow of information with respect to the task at hand. Various studies in the space-based and feature-based domain have demonstrated that the visual system achieves this via gain-control mechanisms. These mechanisms are supposed to result in an enhanced neural representation of relevant stimuli or features while irrelevant ones are suppressed. For example, attending to a specific direction of motion results in an enhanced response of those neurons whose tuning characteristics match with the attended direction, while neurons which prefer the opposite direction are inhibited. However, some models predict that attention can alter the perceived feature as well (e.g. Hamker, Adv. Cogn. Psychol., 2007).
Our study provides psychophysical evidence that global feature-based attention does indeed not simply result in a more salient representation of attended features, but that global feature-based attention is able to dynamically alter the entity of an encoded stimulus in feature space.
While subjects attended to the direction of a target random-dot kinematogram (RDK) presented in one hemisphere, another adaptor RDK was presented in the opposite hemisphere. After a certain time, subjects indicated the direction of the perceived motion aftereffect of the unattended adaptor RDK. We observed that directions close to the attended target are attracted (32 degrees on average) while directions farther away are repelled (29 degrees on average) resulting in an effective expansion of the feature space between these directions. We explain these effects by model simulations in which gain-modulations lead to distortions of the neural population responses if the adaptor direction differs from the one of the target. Furthermore, consistent with recent electrophysiological observations this model predicts changes of tuning curves for cells which are driven by these distorted responses with the consequence that more cells are recruited to process the attended feature.
F. H. was supported by the German Science Foundation (DFG HA 2630/4-1), the European Comission (FP7-ICT: Eyeshots), and the Federal Ministry of Education and Research (BMBF 01GW0653).