Abstract
Identification accuracy of band-limited noise patterns embedded in Gaussian noise increases substantially in a 10AFC task over two training sessions, when the same set of ten patterns is used throughout. Such perceptual learning is stimulus-specific, little improvement transfers to different stimuli with similar spatial attributes. Ostensibly, learning might be invariant for stimulus transformations that leave intact the exemplar identities by preserving the spatial distribution of information in the stimulus. Here, we test the effects of such transformations by measuring the amount of transfer to stimuli reversed in contrast and to stimuli rotated by 180 degrees. Four groups of observers performed the task on two consecutive days. On Day 1, all observers practiced the 10AFC task with a given set of textures. On Day 2 separate groups identified the same textures, different textures, rotated textures, or contrast-reversed textures. In all conditions, stimuli were displayed at one of seven contrasts, in one of three external noise levels, using method of constant stimuli. Results: The effects of contrast reversal and stimulus rotation were the same as replacing the trained stimuli with an unexposed set. The same texture group showed a continuous learning profile across both days, with performance on Day 2 exceeding that measured on Day 1. All other groups' performance dropped sharply on Day 2 relative to performance at the end of Day 1. Although there was some recovery for all three groups, they did not achieve the level of performance of the same texture group on Day 2. This result is particularly surprising for the contrast-reversed stimuli, in which the information distinguishing stimuli is spatially unchanged, unlike the rotated stimuli. Therefore, perceptual learning is sensitive to the presence of stimulus alterations, yet indiscriminate as to the type of stimulus transformation.