Fig. 2: Accuracy dynamics in optimal and suboptimal learners of the Modeling dataset.

a Accuracy dynamics in the training dataset, illustrating the efficacy of the classifier, as optimal learners reach an accuracy optimum on the first training day, while suboptimal learners do so towards the end of training (red dashed line). b Accuracy dynamics in all groups of the Modeling dataset, contrasting the groups receiving verum stimulation to those receiving no stimulation (i.e., those shown in a). The results show a differential effect of stimulation, in which the benefit of stimulation is more pronounced in less efficient learners, while being detrimental to optimal learners. The accuracy dynamics are computed as the ratio of the accuracy of each training block to the accuracy of the first training block, reflecting the performance change relative to the first block. In all plots, the markers represent the average accuracy of each training block (i.e., 6 blocks per day), and the shaded regions reflect the 95% confidence interval of the fitted lines. The hollow markers correspond to groups receiving either no stimulation or placebo stimulation, while the full markers correspond to the groups receiving verum stimulation. The table depicts the number of individuals assigned to each learner tier, and highlights the groups used to train the classifier (in red). In the table, each column corresponds to young, middle-aged, and older adults receiving verum (V), placebo (P), or no stimulation (NS).