Abstract
Perceptual templates for identifying stimuli varying in one dimension may also be tuned along other stimulus dimensions. We examine perceptual learning of orientation judgments in the presence of low- and high-pass filtered external noise to estimate the spatial-frequency sensitivity of the perceptual template. External noise in spatial frequencies overlapping with the template causes elevated thresholds. In this study, we use the integrated reweighting theory (IRT, Dosher et al., 2013), elaborated for multi-alternative choice, to account for the spatial frequency sensitivity of perceptual judgements and learning in a 4-alternative identification (-67.5°, -22.5° +22.5°, or +67.5° from vertical) task for Gabors displayed with filtered external noise. Contrast thresholds (3/1 and 2/1 staircases, averaged for 75% correct) were measured in five low-pass and five high-pass filtered external noise conditions. Observers participated in 14 training sessions with Gabor stimuli (2.66 cpd, sigma = 0.42°) in one of two precued locations at opposite diagonals in periphery, then 2 sessions in locations on the other diagonal, and 2 sessions with Gabors in a new spatial frequency (0.67 cpd, sigma = 0.42°). Threshold vs cutoff spatial frequency functions in the low- and high-pass conditions crossed over near the center frequency of the Gabor, indicating the tuning of the decision template; perceptual learning showed reduced thresholds in both the low and high noise limbs of the functions; switched locations showed partial transfer, and the cross over point shifted towards the center frequency of the new Gabors after the switch. Here we show that the 4-AFC IRT model predicted the observed sensitivity to masking and the improvements in contrast thresholds during learning, and performance in the transfer tests (despite some quantitative departures for transfer to the new spatial frequency). The decision weights of activations in different spatial-frequency and orientation-tuned representation units revealed the “template” used in the task.