Rapid long lasting learning in a collinear edge-detection task

We have developed a detection task in which subjects identify a pair of collinear edges in a field of polygons. Five of our six subjects showed significant, rapid learning at this task. Four showed evidence of retention a day and a week later. In several transfer tests, we found that disruption of the distractors produced a significant drop-off in performance. These results are consistent with a model in which collinear targets initially produce a salience signal too weak to be reliably detected over the noise of the distractors. As the experiment proceeds, the visual system learns to dampen the distractor signals, allowing for more reliable detection.