Abstract
Specificity or partial specificity to either a retinal location or to stimulus dimensions is one hallmarks of perceptual learning. Recent literature suggests that the degree of specificity (or transfer) may depend on task precision, extent of initial training, or pre-training of locations. Position and feature (i.e., orientation) specificity has been taken to imply a key locus of visual representation in early visual cortex. For learned changes rooted in early representations, specificity is the default and the challenge is to explain transfer. Previously, an augmented Hebbian reweighting model (AHRM) (Petrov, Dosher, & Lu, 2005, 2006) has modeled perceptual learning at a single location as reweighting of inputs from early representations to a decision unit. This model successfully predicts the effect of feedback and training accuracy on learning. Here, we report results of a new multi-location-extension to the AHRM (m-AHRM) (Dosher, Jeter, Liu, & Lu, ms) that includes higher-precision location-specific representations and a lower-precision representation shared over a broad set of locations. The m-AHRM then uses reweighting mechanisms to generate implications for transfer to tasks that differ in position or in feature from the original training task. The model predicts differences in level of specificity (or transfer) when a new task involves changes in orientation, changes in location, or both – correctly predicting our results of more transfer to same-orientation judgments in new locations and least transfer to new orientation judgments in the same location. The model also accounts broadly for limitations in transfer to high-precision judgments due to the lower precision of the multi-location representations. In a related abstract (Liu, Lu, & Dosher, VSS 2011), we consider how the m-AHRM accounts for position pre-training (‘double training’) tasks. The m-AHRM advances our understanding of the nature and mechanisms of transfer in perceptual learning.
National Eye Institute Grant # EY-17491.