Supplementary Figure 13: Alternative computational model trained with lower-variation training data.
From: Explicit information for category-orthogonal object properties increases along the ventral stream
(a) Training performance curve for an alternative computational model, trained using a dataset containing a large number of categories but less overall background and object pose variation that the original model shown in Supplementary Fig. 7 (see Online Methods for more information). Axes and labels are as in Supplementary Fig. 7d. (b) Performance of alternative model on the high-variation testing set on which neural data was collected, for categorization task (left panel) and horizontal position estimation task (right panel). Performance for layers 1, 3, 4, 6 are shown. Axes are as in Supplementary Fig. 8. (c) Correlation between performance on testing-set categorization and horizontal position estimation tasks, for each of four model layers shown in panel b. The y axis and error bars are as in Fig. 6e.
