Abstract
Search for a conjunction of visual features is generally slow and inefficient. Here we show that with practice, feature conjunctions pop out: visual search for them becomes rapid and efficient. Participants completed multiple experimental sessions (approx. 5 minutes each, a day or more apart). They viewed a circular array of disks that were green on the left and red on the right. One stimulus, the target, was reversed. While fixating, subjects searched as quickly as they could for the target amongst distracter stimuli. Initially, subjects were slower to detect the target with many distracters (31) than with few distracters (1). However, after one week of training, subjects identified the target as quickly amongst many distracters as amongst few distracters, though the speed of identification for few distracters did not change. Search performance then became asymptotic, after which we rotated both target and distracter stimuli 90° for some subjects and 180° for others. Performance remained asymptotic upon 90° rotation, suggesting generalization of learning. However, performance rebounded to pre-learning levels upon 180° rotation (which swapped targets for distracters and vice versa), suggesting an interference effect where the distracters (heretofore targets) popped out, necessitating active inhibitory processing. In a second experiment, we tested new subjects in a protocol like the first condition of the main experiment, but with more trials in each session (approx. 1 hour). Those subjects did not learn efficient search more quickly than the subjects in the original experiment. In total, our results suggest that inefficient visual search for feature conjunctions can be made efficient via a week of training, and that this learning can generalize to novel stimuli. However, the failure to speed learning via additional exposures in each session suggests that the mechanism for this learning is speed-limited, which provides a clue to its possible neuronal substrate.
Alexander von Humboldt Foundation, NSF GRFP.