Abstract
Performance improvement in perceptual learning may be measured as increases in accuracy for a constant stimulus, or as reductions in threshold contrast for a given threshold accuracy. Conditions that lead to better perceptual learning are of both theoretical and practical concern. Reviews of perceptual learning (Dosher & Lu, 2009) suggest that feedback is not necessary for learning, although it may be helpful under some circumstances. The rate of learning may also be influenced by the ease of the training regime, yielding higher or lower accuracy of response. The current experiment measured contrast thresholds in an orientation discrimination task as a function of feedback and of the presence of different accuracy trials during practice. Observers discriminated the orientation of a Gabor stimulus (+/− 12 deg from vertical at one of two precued locations 5 deg in the periphery). Performance was tested in zero external noise and in high external noise. Four groups of observers were tested. Feedback (on correct trials) was either provided or not, and training with a 75% accuracy adaptive staircase of contrast was mixed either with a 65% adaptive staircase or an 85% adaptive staircase. Learning in the 75% accuracy staircase condition thresholds was estimated through a power function model of perceptual learning. The 75% threshold showed larger and faster improvements when it was measured in the context of the higher accuracy 85% trials than when mixed with the lower accuracy 65% trials, and benefited from feedback in some cases. The results are consistent with the predictions of an augmented Hebbian reweighting model of perceptual learning (Petrov, Dosher, & Lu, 2005, 2006). The model predicts that perceptual learning can occur without feedback, but may be improved by it, and that mixing in practice trials at a higher accuracy level can also lead to better learning. Supported by NEI.
Supported by National Eye Institute.