Figure 2
From: Contextual associations represented both in neural networks and human behavior
(a) Examples of the contrasts of interest. Five different exemplars of each object were used. Category similarity was assessed by comparing representations in the CNN across exemplars of the same category (orange outlines). Context similarity was assessed by comparing the representations of exemplars across the object paired categories (e.g., one easel and one palette exemplar, blue outlines). (b) Example of an ideal similarity matrix for each comparison using simulated data. (c) Representational similarity of images in the same category (orange) and same context (blue) at each layer of VGG16 CNN. Similarity computed as ratio of average in-group versus out-group similarities for each group. Error bars represent standard error of the mean.
