Extended Data Fig. 1: VEIs capture invariance observed in natural images with VEI-like activation.
From: Functional bipartite invariance in mouse primary visual cortex receptive fields
a, Examples of MEI, VEIs, and ‘natural VEIs’ for 4 example neurons. b, Neuronal responses to masked natural images are sparse, and their amplitudes are smaller than those evoked by MEIs and VEIs. The gray lines show the fraction out of 41 million masked images that elicit a given activation or higher for 100 model target neurons; black is the average. Responses from each cell are divided by the response to its MEI. On average, 1.2% of images produced activations above 50%, 0.02% above 75%, and 0.006% above 85% of the MEI activation. c, Natural VEIs maintained high specificity to their target neuron. Confusion matrices showed in silico representational similarity between original VEIs and highly activating natural crops. Representational similarity between original VEIs and highly activating natural crops for the same neurons (diagonal) was larger than cross-neuron similarity (off-diagonal) (two-sided permutation test, P < 10−4 for all conditions after BH corrections). d, Natural VEIs closely resembled the original VEIs. The original VEIs were more similar to natural VEIs of the same neuron than original VEIs of other random neurons (two-sided Wilcoxon signed-rank test, W = 0, P = 1.1 × 10−7). e, Original VEIs have higher mean Euclidean pairwise distances than those of natural VEIs (two-sided Wilcoxon signed-rank test, W = 179, P = 0.01). Data were pooled over 100 neurons randomly sampled from eight mice.
