Abstract
Attention to a visual object sharpens the representation of that object's features, e.g., location, orientation, hue, velocity, etc. These advantages can be spread to multiple objects in the visual field, but there is a trade-off between the number of items attended and the degree of benefit. Here, we show evidence that visual attention can be allocated strategically and fluidly between objects of variable importance. Observers attended to cued arrows among similar distractor arrows. Following a brief blank period, a single target arrow reappeared with a changed direction, and observers indicated whether its displacement was clockwise or counterclockwise. Accuracy was used to assess the precision of representation of direction, which decreases as the number of attended objects increases. Then, the precision of direction representation for equal-valued targets was compared with the precision for targets of variable value. In the variable value condition, subjects were instructed to maximize their scores based on a point scheme where correctly assessing the displacement direction of High-Value targets was worth twice as many points as Low-Value targets. While subjects' performance showed a relatively precise representation of High-Value targets, the representation of Low-Value targets was impaired. The combined accuracy for judging both High-Value and Low-Value targets was comparable to the previously recorded accuracy for the same total number of equal-valued targets. Analysis via a mixture model that distinguishes between the precision of target representation and probability of failure to represent a target at all (Bays & Husain, 2008; Zhang & Luck, 2008), suggests that this differential benefit of attention is due to strategic allocation of attentional resources, rather than simply attending to High-Value targets and ignoring Low-Value targets. The results of these experiments demonstrate that while visual attention is a limited resource, it can be strategically and fluidly allocated to benefit attended items to varying degrees.
Research supported by CELEST, an NSF Science of Learning Center (NSF SBE-0354378). EM was supported in part by HP (DARPA prime HR001109-03-0001) and HRL Labs LLC ( DARPA prime HR001-09-C-0011).