Supplementary Figure 6: Comparison of task overlap to random and minimal control models.
From: Explicit information for category-orthogonal object properties increases along the ventral stream
(a) Comparison of weight overlap for categorical vs non-categorical tasks, relative to random overlap (top) and minimal overlap (bottom) models. Color represents t-statistic of separation of each pairwise overlap from the indicated control model (random in the top panel, minimal overlap in lower panel), with red color representing more overlap than random and blue representing less. (b) Average overlap for (i) (non-face) categorical tasks, (ii) faces-vs-non-face categorical tasks, (iii) non-face categorical tasks vs non-categorical tasks and (iv) faces vs non-categorical tasks. Shown are actual neural overlap (blue bars) in comparison to random overlap (gray) and minimal overlap (orange) models. Error bars for neural data overlap are due to variation in unit sampling and classifier training split. Error bars in model overlaps are due to variation of model input data (per-task and per-unit weight constrains) due to unit sampling and classifier training split, as well as random initial conditions of model weights.
