Drift-Diffusion Modeling Reveals Dissociable Components of Perceptual Learning

Visual perceptual learning leads to improved accuracy and faster reaction times. This effect is specific to the location and lower-level features of the training stimulus, as well as other task parameters. However, extensive practice using a specific set of stimulus conditions has been shown to partially generalize to new conditions in the form of faster learning rates and slightly improved performance at the outset. We hypothesized that this effect is caused by nuanced changes in the subject’s decision-making strategy that can be revealed through drift-diffusion modeling (DDM). To test this hypothesis, subjects were trained on a coarse direction discrimination task using random dot motion stimuli over several sessions. Stimulus location and direction of motion were fixed during training, and either of them changed on the testing session. A subset of participants were trained on separate days to test for the effect of sleep consolidation. We observed the typical training effects of improved accuracy and reaction times as a function of training. DDM modeling showed that learning induced increased drift rate, decreased boundary height, lowered integration leak, and decreased rate of the collapsing bound. Importantly, while changes to the drift rate and boundary conditions mostly reverted to their pre-training values on the testing session, integration leak and rate of the collapsing bound remained at their post-training values in the new stimulus conditions. Sleep consolidation exerted no meaningful difference on performance, reaction times, or parameter values. These results suggest that the observed changes to accuracy and reaction times are caused by at least two different components with different degrees of transferability, as well as the fruitfulness of DDM for understanding the nuances of perceptual learning.