Purchase this article with an account.
Jiajuan Liu, Zhonglin Lu, Barbara Dosher; Augmented Hebbian Learning Hypothesis in Perceptual Learning: Interaction between feedback and training accuracy. Journal of Vision 2008;8(6):1124. doi: 10.1167/8.6.1124.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Previous analyses of the role of feedback have suggested that perceptual learning may be accomplished through augmented Hebbian learning. (Petrov, Dosher, & Lu, 2005; Vaina et al., 1995) When there is feedback, the product of the feedback and the input is used to update the weights in the neural network; in the absence of feedback, the product of the value generated by the output unit and the input is used to update the weights. One prediction of this learning rule is that the ability to exhibit perceptual learning without feedback may depend on the training accuracy level. In contrast, a pure supervised error correction model will not learn without feedback, and a pure Hebbian learning model will not depend on the presence of feedback. We tested the predictions of these learning rules. The accelerated stochastic approximation method was used to track threshold contrasts at particular performance accuracy levels in a Gabor orientation identification task over 6 training days. Subjects were divided into 4 groups: high training accuracy (85% correct) with and without feedback, and low training accuracy (65%) with and without feedback. Contrast thresholds improved in the high training accuracy groups, independent of the feedback condition. However, threshold improved in the low training accuracy condition only in the presence of feedback but not in the absence of feedback. Furthermore, the learning rate did not depend on training accuracy in the feedback condition, nor did it depend on the feedback condition in the high training accuracy condition. The results are both qualitatively and quantitatively consistent with the predictions of an augmented Hebbian learning model, and are not consistent with pure supervised error correction and pure Hebbian learning models. The results lend further support for the augmented Hebbian learning hypothesis in perceptual learning.
This PDF is available to Subscribers Only