Zorzi, M. 1, 2 & Stoianov, I. 1
1 Dipartimento di Psicologia Generale, University of Padova
2 Center for Cognitive Science, University of Padova
Many animal species have evolved a capacity to estimate the numerosity of visual objects. Though foundational to mathematical learning in humans, the nature of the computations underlying this “visual sense of number” remains controversial. Here we show that visual numerosity emerges as a statistical property of images in “deep networks” that learn a hierarchical generative model of the sensory input. Emergent numerosity detectors in the network’s deepest layer had response profiles resembling that of monkey parietal neurons and their activity supported a numerosity estimation task with the same behavioral signature and performance level shown by humans and animals. The unsuspected simplicity of the underlying neural mechanism fits well with the long phylogenetic history of numerosity perception.