Dissertation/Thesis Abstract

Mode of visual perceptual learning: Augmented Hebbian learning explains the function of feedback and beyond
by Liu, Jiajuan, Ph.D., University of Southern California, 2011, 133; 3477949
Abstract (Summary)

This thesis concerns the augmented Hebbian reweighting model (AHRM) in perceptual learning. The development of AHRM was inspired by two sets of research endeavors: first, converging researches in perceptual learning suggest that subjects' performance may be improved through specific channel reweighting between representation areas and decision units in visual system (Dosher & Lu, 1998; 1999); second, the complex pattern of empirical results concerning the role of feedback in perceptual learning rules out both a purely supervised mode and a purely unsupervised mode of learning, and leads some researchers to suggest that feedback may change the learning rate through top-down control instead of acting as a teacher signal (Herzog & Fahle, 1998). Constructed by Petrov, Dosher and Lu (Petrov, Dosher & Lu, 2005), the AHRM utilizes feedback to influence the effective rate of learning by serving as an additional input instead of a direct teacher signal. In this thesis, we discuss two predictions of the AHRM: one regarding the interactions between feedback and levels of training accuracy, and the other concerning the facilitatory effect from trials of high training accuracy to low ones. Training in a Gabor orientation identification task over six days, subjects were divided into groups of different feedback conditions and levels of training accuracy (or mixtures of different levels of training accuracy). In the tasks with a single training accuracy level, contrast thresholds improved in the condition of high training accuracy independent of feedback conditions, but thresholds improved in the condition of low training accuracy only in the presence of feedback. In the tasks with levels of low and high training accuracy mixed, however, even without feedback, contrast thresholds in the low training accuracy level still improved roughly the same amount as those in high training accuracy levels and/or with feedback. The results are both qualitatively and quantitatively consistent with the predictions of the AHRM, but inconsistent with purely supervised error correction models or purely unsupervised Hebbian learning models.

The AHRM was also applied in modelling the perceptual learning in external noise and helped us to understand several interesting observations: (1) perceptual learning improved contrast thresholds at all levels of external noise in peripheral orientation identification (Dosher & Lu, 1998, 1999), (2) training with low external noise improved performance in high external noise, while training with high external noise did not affect performance in low external noise (Dosher & Lu, 2005), and (3) pre-training with high external noise only reduced subsequent learning in high external noise, whereas pre-training with zero external noise practically eliminated any or left very little additional learning in all the external noise conditions (Lu, Chu & Dosher, 2006).

The behavioral confirmation of the predictions from the AHRM, together with its ability to interpret a range of experimental results in perceptual learning, provides strong support for further developments and applications of AHRM in future research of visual perceptual learning and other related areas.

Indexing (document details)
Advisor: Lu, Zhong-Lin
Commitee: Baker, Robert, Grzywacz, Norberto, Hirsch, Judith, Tjan, Bosco
School: University of Southern California
Department: Neuroscience
School Location: United States -- California
Source: DAI-B 73/01, Dissertation Abstracts International
Subjects: Cognitive psychology
Keywords: Augmented Hebbian learning, Feedback, Visual perceptual learning
Publication Number: 3477949
ISBN: 9781124950815
Copyright © 2019 ProQuest LLC. All rights reserved. Terms and Conditions Privacy Policy Cookie Policy