Purchase this article with an account.
Jiajuan Liu, Zhong-Lin Lu, Barbara Dosher; Multi-location Augmented Hebbian Re-Weighting Accounts for Transfer of Perceptual Learning following Double Training. Journal of Vision 2011;11(11):992. https://doi.org/10.1167/11.11.992.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
The hallmark finding in perceptual learning has been the widespread observation of specificity of learned improvements to either a retinal location or to a basic stimulus dimension. Recent literature suggests that the degree of specificity may depend on task difficulty, task precision, number of trials, the extent of initial training and double training. This complex pattern of results cries out for a coherent theoretical account. Here, we further develop a multi-location Augmented Hebbian Reweighting Model (m-AHRM) (Dosher, Jeter, Liu, & Lu, ms) to account for the specificity and transfer of perceptual learning to different spatial regions and stimuli/tasks. In the m-AHRM, several location-specific and one location-invariant representations are connected through weight structures to the decision unit, along with inputs from a bias unit and a feedback unit. The location-invariant representation receives gated inputs from all the location-specific representations. Learning at one location changes the weights between the location-specific representation for that location and the decision unit, and weights between the location-invariant representation and the decision unit. Also, the gain of the gate between the location-specific and location-invariant representation increases. We applied the m-AHRM to account for the observed transfer of perceptual learning following double training (Xiao et al., 2008). Specifically, pre-training on task T2 in location L2 allows subsequent training on a different task T1 in another location L1 to generalize to location L2, while it would not have generalized without the pre-training. The m-AHRM can reproduce the results from all three double training experiments in Xiao et al. (2008). In another study (Dosher et al., VSS 2011), we show that the m-AHRM is also able to account for specificity results following location and/or feature changes, as well as task precision dependent specificity in perceptual learning. The m-AHRM provides a general framework to understand specificity and transfer in perceptual learning.
This PDF is available to Subscribers Only