Visual perceptual learning: A new perspective
Cong Yu
Department of Psychology and Peking-Tsinghua Center for Life Sciences, Peking University, Beijing, China
Abstract: For many years perceptual learning (PL) researchers are excited by the observations that PL is specific to the trained retinal location and orientation, and regard these specificities as indications of neural plasticity in the early visual cortex. However, we have created a “double training” design to enable complete learning transfer to untrained conditions, suggesting that PL is mainly a cognitive learning process beyond the retinotopic visual areas.
More recently we studied why PL is specific in the first place, and what is really learned in PL. In double training, PL becomes transferable if the subjects are exposed to the untrained location or orientation through an irrelevant task, indicating that learning specificity may be to do with the untrained conditions. We used a continuous flash suppression technique to create “bottom-up only” or “top-down only” exposure stimuli. We found that bottom-up exposure of an untrained location or orientation, or pure top-down influences, can both enable significant and sometimes complete learning transfer. These results suggest that learning specificities may result from neural under-activations at the untrained conditions that receive insufficient bottom-up stimulation and/or top-down attention during training.
In addition, we demonstrated that PL is a form of concept learning. We studied orientation learning with orientation features defined by either luminance gratings or symmetric dot patterns, and motion direction learning with first-order luminance stimuli and second-order contrast stimuli. Each pair of stimuli had distinctive physical properties that are encoded by different neuronal mechanisms. Using a variation of the double training method, we observed complete mutual transfer of learning between grating and dot-pattern defined orientations, and between first- and second-order motion directions. These results indicate that what is learned in PL is the abstract concept of a trained visual feature, such as orientation or motion direction.