Differential Network Metrics as Predictors of Specificity and Transfer in Perceptual Learning

Visual perceptual learning is characterized by specificity to trained conditions. Understanding the mechanisms behind this specificity remains a fundamental topic in perceptual learning research. In a previous ERP study (Zhang-et-al., 2013), half the participants (n=14) in Vernier learning exhibited location specificity (TI<0.5), while the other half demonstrated learning transfer (TI>0.5) across the visual hemifield. Here, we examined small-world network metrics to compare the changes in the topological properties of the whole-brain network between the specificity and transfer groups. We established functional connectivity between each pair of EEG channels by calculating the phase-locking value using wavelet transform in the beta band. We then applied a threshold to obtain undirected binary small-world networks for each participant at trained and transfer locations before and after training. From these networks, we calculated the clustering coefficient, indicating local interconnectedness, and the shortest path length, reflecting global information integration. Before training, both groups exhibited similar small-world properties, characterized by high clustering coefficients and short path lengths. After training, while both groups displayed similar threshold reductions, the specificity group showed higher clustering coefficient (p=.003) and longer path length (p<.001), whereas the network metrics of the transfer group remained unchanged (ps>.111). Additionally, when the transfer effects were tested at the untrained location, the specificity group demonstrated trends (ps<.007) in network metrics similar to those at the trained location, while the transfer group displayed higher clustering coefficient (p=.035) with unchanged path length. We conclude that whether learning is location-specific is pre-determined by training-induced changes in network properties. While the transfer group maintained both local and global efficiencies, the specificity group exhibited over-optimized local efficiency at the expense of global efficiency, which likely hinders learning transfer. These results provide crucial insights into the mechanisms of specificity and transfer in perceptual learning from the perspective of brain network dynamics.