Fig. 8.

Faster learners show greater differentiation in large-scale brain network configuration between trained and novel problems. (a) Task-related brain activation was used to identify 23 network nodes. (b–c) Task-related effective connectivity was used to compute a network connectivity matrix for trained (b) and novel (c) problems. (d) 23 links showed significant differences in connectivity between trained and novel problems (p < .05, in yellow). These links were used as input features in a linear SVM classifier. Leave one out cross validation and permutation tests (1000 samples) revealed a classification accuracy of 70% (p =  0.01). (e) Learning rate predicted greater differentiation of brain networks between trained and novel problems. Connectivity pattern distance was computed using sum of absolute distance from the SVM hyperplane (separating connectivity between trained and novel problems) to each condition. Dashed lines represent 95% confidence interval. Abbreviations as in Fig. 5.