Dramatic Changes in Mechanisms of Task-Irrelevant Visual Perceptual Learning from Childhood to Adulthood

Unsupervised learning as a computational principle works in visual learning of natural scenes, but not of artificial stimuli

“…We found that exposure to natural scene images resulted in VPL of the dominant orientation in these images. Conversely, exposure to artificial images, despite matching local orientation and spatial frequency distribution with natural scene images, did not lead to VPL, as has been found in previous studies using artificial supra-threshold stimuli (Astorga et al, 2022;Chang et al, 2014;Frank et al, 2020;Tsushima et al, 2008). These findings suggest that the occurrence of exposure-based task-irrelevant learning depends on whether the exposed stimuli were natural scenes or artificial images.…”

“…As discussed above, consistent results show that exposure to supra-threshold task-irrelevant artificial stimuli, including Gabor patches, does not lead to VPL due to attentional suppression on the task-irrelevant stimuli (Ahissar & Hochstein, 1993;Astorga et al, 2022;Chang et al, 2014;Frank et al, 2020;Huang et al, 2007;Lewis et al, 2009;Li et al, 2004;Mukai et al, 2011;Paffen et al, 2008;Schoups et al, 2001;Shiu & Pashler, 1992;Tsushima et al, 2006Tsushima et al, , 2008. Meanwhile, the dominant orientation derived from the higher-order statistics from NS images may have spatially and/or temporarily differences in quality compared to that derived from the lower-order statistics in FS and KS images (Barbot et al, 2011;Dakin et al, 1999;Landy & Graham, 2004).…”

“…Whether unsupervised learning is used for human VPL has been empirically tested (Herzog & Fahle, 1998;Shibata et al, 2009Shibata et al, , 2014Tsodyks & Gilbert, 2004;Watanabe & Sasaki, 2015). As a result, these studies concluded that unsupervised learning model appears to be contradicted by a biological model; this conclusion is based on a significant body of findings showing that VPL does not occur as a result of exposure to supra-threshold and task-irrelevant features due to attentional suppression on those features (Astorga et al, 2022;Chang et al, 2014;Frank et al, 2020;Tsushima et al, 2008). In the current study, we demonstrate a resolution to this controversy with a new finding: Exposure to a dominant global orientation as task-irrelevant leads to VPL of the orientation, only when the orientation is derived from natural scene images.…”

“…In the exposure stage, participants were asked to perform a rapid serial visual presentation (RSVP) task, designed to make participants' attention focused on the central task (Chang et al, 2014;Frank et al, 2020;Nishina et al, 2007;Seitz & Watanabe, 2003;Tsushima et al, 2008;Watanabe et al, 2001). In each trial, a sequence of eight alphabetical letters and two digits were presented in a random order for each participant at the center of the display (Fig.…”

“…One possibility is that the dominant orientation generated from these higher-order statistics was sub-threshold and invisible. Task-irrelevant VPL of a feature in an artificial stimulus occurs only if the feature is sub-threshold (Chang et al, 2014;Frank et al, 2020;Tsushima et al, 2006Tsushima et al, , 2008. That might have been why task-irrelevant VPL of the dominant orientation in the images containing these higher-order statistics have occurred.…”