Supplementary Figure 8: Performance time-courses for computational model on test set tasks.

y-axis represents performance of linear regressors and classifiers trained on the top hidden layer of the computational model, on each task defined on the testing set (see Supplementary Fig. 1a). x-axis represents timepoints taken during training for categorization on the ImageNet dataset (as described in Methods). Performance was estimated by building top-level regularized classifiers and regressors (as described in the methods text) separately at each time step. Note that the x-axis is the same for all panels, representing the same training trajectory; the various y-axis panels are all based on the single feature set produced by the categorization training. The first two panels, with gray backgrounds, indicate categorical tasks (8-way basic categorization and subordinate category identifications); the remaining white-background panels indicate non-categorical tasks.