Figure 4.

Classification of between-category trials. A, C, Mean classifier estimates of delay-period activity are shown for the two validation subjects who were instructed to solve the working memory task using a prospective strategy (A) and for the eight experimental subjects who received no instruction on which coding strategy to prioritize during the task (C). Note that the results in A validate the method, in that delay-period activity classification matches the cognitive strategy that these two subjects were instructed to adopt. The vertical axis shows the classifier estimates of the correspondence between the delay-period brain activity and the learned patterns of category-specific activity (for people, locations, and objects) associated with the stimuli from each trial. Estimates are collapsed across all 36 between-category trials for each subject into three categories: retrospective (the category of the target), prospective (the category of its associate), and other (task-irrelevant). The horizontal axis shows the timing of the working memory trial. The black bars along this axis indicate the presentations of target and probe stimuli (target, 0–1 s; probe, 12–13 s). C, Bottom, The bar graphs indicate the number of subjects that demonstrated each type of category selectivity at each time point (black, retrospective; purple, prospective; gray, other). These selectivity statistics were calculated by assigning each subject to a particular category based on the largest average estimate of the three categories for each point. B, Statistical comparisons for validation subjects focused on the within-subject difference between the prospective classifier estimate and the average of the nonprospective estimates (retrospective and other). D, For experimental subjects, statistical comparisons focused on the within-subject difference between the average of the task-relevant estimates (retrospective and prospective) and the task-irrelevant estimate (other). *p < 0.01 (B), and *p < 0.05 (D). Error bars represent a 95% confidence interval around the within-subject difference scores.