Specificity and transfer in perceptual learning of motion
56.3010, Tuesday, 19-May, 2:45 pm - 6:45 pm, Banyan Breezeway
Ruyuan Zhang1, Oh-Sang Kwon1, Duje Tadin1,2; 1Center for Visual Science & Dept. of Brain and Cognitive Sciences, University of Rochester, 2Department of Ophthalmology, University of Rochester, Rochester, NY, USA 14627
Transfer of learning is a key characteristic of perceptual learning that is often used to reveal the underlying mechanisms. For instance, well-established feature and location specificity of perceptual learning highlight the importance of low-level feature and spatial tuning in learning. Here, we investigated the mechanisms governing the perceptual learning of motion by examining its transfer effects. Results reveal two distinct mechanisms that are selectively responsible for the learning process, which evidently depends on the contrast and size of learning stimuli. METHODS: Participants were instructed to identify motion direction of random texture patterns (6 °/s speed), which moved in one of four directions (up, down, left or right). In pre- and post-tests, participants’ duration thresholds (i.e., minimum stimulus duration for participants to detect motion direction) were assessed at seven contrast levels (2% – 100%) and six stimulus radius levels (1° – 8°). During the 8 day training phase, one group of participants (N=5) were trained with small, low-contrast stimuli (radius drawn from Gaussian distribution: u=1.33°, σ=0.11°; contrast: u=2.85%, σ=0.5%) while another group of participants (N=6) were trained on large, high-contrast stimuli (radius: u=6°, σ=0.5°; contrast: μ=70%, σ=10%). RESULTS and DISCUSSION: Surprisingly, training effects on small, low-contrast stimuli generalized to large, high-contrast stimuli but not vice versa. Such dissociable transfer patterns suggest two distinct learning mechanisms. Training on small, low-contrast stimuli generally improves motion signal processing, which may account for the observed transfer of learning. On the other hand, we speculate, that training on large, high-contrast stimuli alters center-surround antagonism in motion processing, which is mainly specific for large, high-contrast stimuli (Tadin et al., 2003). We are currently working on a unified computational model to accommodate the two learning mechanisms observed in the data.