Abstract
Purpose: To constrain models, and localization of function, of perceptual learning and aftereffects by studying their interaction. Methods: We measured tilt aftereffects produced by adaptation to a heavily masked Gabor patch (low contrast and buried in white Gaussian noise), both before and after discrimination training with the same patch. Thresholds were determined by varying the contrast of the noise-masked patch, and asking observers to discriminate +20/−20 deg tilts (thresholds = ∼ 2–3% contrast). During the pre-adapt phase, observers adapted to this patch (+20 or −20 deg, varied across session), then made an orientation judgment of a subsequent supra-threshold test patch of identical spatial frequency (and ∼ 0 deg tilt). In the training phase, observers were briefly shown the patch, and were asked to discriminate +20/−20 deg tilts, with feedback; ten 300 trial sessions were run, across a week. In the post-adapt phase, observers repeated the adaptation procedure. Results: In spite of their near-invisibility, the pre-adapt phase yielded tilt aftereffects of about 1 deg (for comparison, the tilt aftereffect for a high-contrast adaptor is about 4–5 deg). During training, three out of our four observers showed significant learning, increasing on average from 74% correct to 81%. Importantly, for post-adapt, all four observers (interestingly, including the observer who did not show learning) showed a significant increase in their tilt aftereffect, doubling to 2 deg. Conclusions: The mechanisms that are ‘tuned’ during external noise-limited discrimination training may be the same as those underlying the tilt aftereffect. We are developing a model (taking into account current thinking on perceptual learning), where learning occurs because the more informative channels are given greater weight; if one of these more heavily weighted channels is subsequently adapted, it results in a greater-than-usual unbalance in pooling statistics, yielding larger aftereffects.