Abstract
In visual search tasks, attention is guided by a limited set of attributes (color, size, etc.). Shape is one of those attributes, but our understanding of shape guidance has been restricted to shape features (e.g. line termination and closure) that do not fully describe preattentive shape processing. In an effort explore shape space and to generate new hypotheses about shape guidance, we have used a novel genetic algorithm method. We start with a target and twelve distractors, generated randomly in a radial frequency space defined by 10 radial frequencies each with an amplitude and a relative phase. On each trial, observers search for the target in arrays composed of one of the distractors. Reaction time (RT) is the measure of 'fitness'. To evolve an easier search task, distractors that produce faster RTs survive into the next generation, mate, and have children. To evolve a harder search task, distractors yielding longer RTs survive. Items also 'mutate' at a modest rate. The 30 "genes" are the ten frequencies, amplitudes, and phases. The method works. Eight generations of evolution (~20 minute task) can produce search tasks either much easier or much harder than the starting task. For some targets, these results are easily interpretable in radial frequency space. Easy distractors evolve amplitude X frequency spectra that are dissimilar from the target. Hard distractors evolve spectra that are more similar to the target. However, other targets suggest different rules. For instance, when the target is a rabbit silhouette, the hard distractors do not resemble the target in radial frequency space. Moreover, distractors that make rabbit search inefficient do not look much like rabbits. Inefficient rabbit search may arise when distractors have rabbit 'parts' (oblong body, ear-like structures) even if the whole is quite unrecognizable. The same holds true for other target shapes.
Meeting abstract presented at VSS 2016