Non-stationary Salience Processing During Perceptual Training in Humans

Abstract

Performance in sensory tasks improves with practice. Some theories suggest that the generalization of learning depends on task difficulty. In consequence, most studies have focused on measuring learning specificity, and perceptual impact after training completes. However, how exactly sustained changes in task difficulty influence the learning curves and how this affects the efficiency of perceptual discrimination is not well understood. Here, we adapted a visual task for humans by creating monocular training programs with increasing (SIMinc) and decreasing (SIMdec) stimulus similarities. We found a marked improvement in all participants after 10 days of training, with an almost complete transfer of learning to the untrained eyes. Interestingly, the training paradigms led to drastically different learning curves for the SIMinc and SIMdec groups. The learning curves were best predicted by an associative learning model that allowed stimuli to gain or lose salience depending on how the subject’s learned about them. On addition, a non-stationary sequential sampling model that jointly accounts for choice and RT distributions revealed that the SIMinc group led to faster evidence accumulation rate relative to the SIMdec group. Altogether, our results illustrate how different learning trajectories influenced attentional salience processing leading to distinctive stimulus processing efficiencies. This crucial interdependence determines how observers learn to guide their attention towards visual stimuli in search for a decision.