Abstract
Acquisition of perceptual expertise requires training of perceptual skills with related but different tasks and stimuli. Recent studies found that different stimuli and tasks interact in intermixed training protocols, in some cases eliminating learning (Adini et al. 2004; Yu et al. 2004). The Augmented Hebbian Reweighting Model (AHRM) makes detailed predictions about intermixed training protocols: Randomly mixing multiple stimuli/tasks may make competing and opposing demands on the reweighting structure, or cooperate to support a common weight structure. Optimizing for one task may help, conflict with, or be independent of optimizing for another task, depending on the exact task conditions. Here we test two AHRM predictions: Roving in a dimension separate from the judgment dimension or with tasks that rely on highly localized weight structures should not prevent learning. In experiment 1, two groups were trained in a 45 ± 5 deg Gabor orientation identification task in four spatial frequencies (SFs: 1, 2, 4 and 8 cpd). The SF was fixed within each 80-trial block in the blocked group but intermixed randomly in the roving group. After four sessions of training (2,560 trials), both groups showed substantial improvements in performance with no difference in the magnitude of threshold reduction between the groups. In experiment 2, two groups of participants were trained in a ± 2.5 degree Gabor orientation discrimination task with four base angles (22.5, 57.5, 112.5 and 157.5 deg) at a fixed SF (2 cpd). The base angle was fixed within each 80-trial block in the blocked group but randomized in the roving group. Participants in both groups improved their performance for all base angles. Although the thresholds of the roving group were higher early in learning, the two groups reached the same asymptotic thresholds following 2,560 trials of training. The results are consistent with the predictions of the AHRM.
Meeting abstract presented at VSS 2013