Extended Data Fig. 5: RSMs for untrained AlexNet.
From: Infants have rich visual categories in ventrotemporal cortex at 2 months of age
(a) AlexNet was initialised with random weights and had no exposure to visual stimuli. Layer wise activations were calculated in response to the 36 stimuli, while keeping weights frozen. (b) Untrained RSMs for each layer. X and Y axes are the 36 objects, nested by tripartite class (3), category (4) and exemplar (3). Pairwise correlations were calculated between activations in response to each stimulus. We plot the z-scored correlation to highlight representational content rather than strength, and to aid comparison to brain-derived RSMs. Raw correlation ranges: all diagonal values are 1, indicating within exemplar comparisons, off-diagonal ranges were conv1: (−0.1,0.5), conv2: (0.1,0.7), conv3: (0.3,0.9), conv4: (0.4,0.9), conv5: (0.4,0.9), fc6: (0.6,1) fc7: (0.7,1). conv, convolutional. fc, fully connected.
